Skip to content

armgit / Orkas

Public Imported from GitHub
0

Orkas is an open-source (MIT), local-first AI team for your desktop: a Commander turns your goal into to-dos and coordinates 16 built-in specialist agents plus Claude Code, Codex, OpenCode, OpenClaw and Hermes, with projects, scheduled automations and connectors. Free app for macOS & Windows; built-in models or your own keys.

feat(reflection orchestrator + window state + signals patch suggester + plan polish)

wyta0911ac905b9
diff --git a/CLAUDE.md b/CLAUDE.md
index 5d809d76..0a932371 100644
--- a/CLAUDE.md
+++ b/CLAUDE.md
@@ -85,22 +85,10 @@ util/ Pure-function utilities (stdlib only or single third-party d
Exception: env-var names with the project prefix (`ORKAS_NODE` / `ORKAS_PC_DIR`) are allowed when actually referenced in bash. Before adding/changing a prompt, `grep "Orkas\|/Users/\|/home/\|PC/data\|PC/src" src/main/prompts/*.md` should come back clean.
-**Platform-editing language constraint**: when the inline edit chat targets a platform-installed skill / agent (dev mode only — see §6), the system-prompt tail uses `i18n.ts::buildBuiltinEditingConstraint()` instead of `buildLanguageDirective()`. The directive forces English output for every field except `description_zh`, regardless of the user's UI language. **Why:** platform content ships to a multilingual user base; source files must stay English-only (see §6) and per-user UI language must not influence what gets published.
-
**Authoring rules live in platform skills, not in `chat_*.md`**: the canonical authoring rules for agents (`<agent>` container) and skills (`<skill>` container + `<<<skill-file>>>` blocks) live in two platform skills installed from the marketplace under ids `agent-creator` and `skill-creator`. The four prompts that previously copy-pasted these rules (`chat_commander.md`, `chat_agent_setup.md`, `chat_agent_setup_cli.md`, `chat_skill_setup.md`) now carry only a session-shell + `read_file <ROOT>/agent-creator/SKILL.md` (or skill-creator) pointer. **Why:** the long authoring rules are ~200 lines each and were duplicated across four prompts (~600 lines drift surface), kept in sync only by reviewer attention; the duplication caused the commander prompt to balloon past 460 lines, putting the rules in the lost-in-the-middle band. Moving them to platform skills cuts the always-loaded system prompt by ~590 lines and gives every authoring surface a single source of truth — the LLM `read_file`s the skill on demand. Field-level authoring details, including `category` candidate codes, belong in the matching creator skill; if the category set changes, update `agent-creator/SKILL.md` and `skill-creator/SKILL.md` together. The writer backfills a default category on missing/unsafe values; missing category is an advisory, not a write-blocking schema failure.
**Cross-prompt sync constraint for residual shared rules**: a few smaller cross-prompt concepts still live inline (`<agent-input-form>` / `<agent-input-submission>` extract-first protocol in `chat_agent_in_group.md`, the `chat_shared_rules.md` web/PDF/file-output block injected by `bus.ts::concatSharedRules`). For those, the historical principle stands: pick one authoritative prompt as truth, downstream sites carry only the principle (no field detail / schema tables / full translation tables). Before touching the truth source, run `grep -E "<agent-input-form>|<agent-input-submission>" src/main/prompts/*.md` to reconcile downstream.
-**Adding prompt text**: lean-prompt discipline applies to **every LLM-facing prompt** — `prompts/*.md`, SKILL.md bodies, agent / skill specs, and CLAUDE.md itself. Before adding any new clause, run two checks:
-
-1. **Is each clause necessary?** Cut narrative filler ("…and is what users complain about most"), redundant cross-references to globally-declared rules ("(in the user's UI language — the trailing language directive sets that)"), and double-emphasis ("even one token, even one field"). Keep the rule, the *Why* for counter-intuitive choices, and a concrete example when the rule is structural. **The test:** delete the clause; if the model behaves the same, drop it.
-
-2. **Does it conflict with or duplicate existing content?** Grep the same prompt md and its peers (`chat_*.md` share a domain; `chat_shared_rules.md` is injected into multiple roles — see the cross-prompt sync constraint above). Partial overlap → **merge** into one tighter rule, don't ship a parallel paragraph (parallel rules dilute each other; the model honours neither cleanly). Contradiction → **fix one to match the other**, don't ship both.
-
-3. **Show the audit (process enforcement)**: any response that submits a prompt-md edit MUST contain a visible per-clause audit — one row per newly added clause marked ✅ (necessary; which Keep category) or ❌ (cut; which redundancy reason from check 1). Without the audit in the response, treat the edit as incomplete. This binds checks 1 and 2 to a concrete moment so they aren't silently skipped.
-
-**Why:** prompt text sits in the cached prefix and shapes every turn. We shipped "don't paste absolute paths" + "don't enumerate bash scratch" as two parallel paragraphs in `chat_shared_rules.md` and watched the model honour neither until the merge. We also shipped this rule itself as advisory at first and watched the audit get skipped on every subsequent prompt-md edit until the audit-as-artifact step made skipping visible. Merge first, write second; show the audit.
-
**Prompt-md cache layout** (KV-cache prefix discipline): every `src/main/prompts/<name>.md` is split into two halves — **static rules first, then a single trailing `## Runtime injection` section that collects every `$variable` substitution** (with the most volatile field at the bottom of that section). Final assembled order = `<role>.md` static body → `chat_shared_rules.md` (when applicable) → `<role>.md` `## Runtime injection` → `buildLanguageDirective()`. The shared block is wedged in front of the runtime block by `bus.ts::concatSharedRules`; the language directive is appended last by `appendLanguageDirective`. **No `## Current xxx` / "current state" block at the top of the file** — anything the LLM itself rewrites mid-session (`<agent>` writeback / `<<<skill-file>>>` / `plan_set` / form submissions: `$name` / `$description*` / `$workflow` / `$skill_dir` / `$skill_files` / `$plan_state` / `$agents_index` / `$inputs_schema` / `$interactive`) goes only into the trailing runtime section. **Why:** putting volatile fields up front means the very next turn rewrites the cache prefix and we eat a full re-prefill; `chat_agent_setup.md` / `chat_agent_setup_cli.md` / `chat_skill_setup.md` shipped with a top-of-file `## Current agent` / `## Current skill` block, every edit turn invalidated the cache, and the fix was to consolidate those into the trailing `## Runtime injection` section. Same applies to any new prompt md added under `src/main/prompts/`. Path-constants that are fixed for the lifetime of a session (`$builtin_*_dir` / `$custom_*_dir` / `$working_dir`) are session-stable and may stay inline in the body; only fields that the LLM or bus mutate within a session are forbidden in the prefix.
---
@@ -142,7 +130,7 @@ Top-level three-way split: **cloud** (user-private, synced across devices) / **l
**Hard constraints**:
1. **At top level**, only `users.json` / `logs/` / `<uid>/` are allowed; per-user data must land under `<uid>/{cloud,local}/`. There is no shipped global `builtin/` tree (retired with the marketplace cutover; one-shot migration in `util/migrate-marketplace.ts` moves any legacy `data/builtin/` into the active uid's cloud tree on first launch, then deletes it).
2. **Platform agents / skills are marketplace-only**: published to the server via `marketplace_dev.upload*` (dev only), installed via `marketplace.installMarketplace*` to `<uid>/local/marketplace/{agents,skills}/<id>/`. The on-disk dir contains the same shape as a custom one (`agent.json` for agents, `SKILL.md` + script files for skills) plus a `_install.json` with `(version, published_at)` used by reconcile to detect drift. The cloud-synced manifest `<uid>/cloud/marketplace/installs.json` records `{id, version, published_at, url}` per entry; `marketplace_reconcile.reconcileInstalls(uid)` runs on startup, parallel-fetches anything missing from the local tree, and re-pulls when the manifest version disagrees with `_install.json`. Immediately before reconcile, `checkServerUpdatesForInstalls(uid)` sweeps every installed id against the server `/marketplace/{agents,skills}/list` endpoint and bumps the manifest's `(version, published_at)` when the server is newer — admin republishes propagate to every client on next startup without user action; offline boots swallow the network error and fall through to a normal reconcile against the stale manifest. **Why this split**: content is binary-blob / catalog-versioned and would tear cloud sync if every machine wrote the same bytes; per-machine local copy + tiny cloud-synced manifest = single source of truth at the manifest level, machine-private at the content level. The loaders scan `[<uid>/cloud/{agents,skills}/, <uid>/local/marketplace/{agents,skills}/]`; custom takes precedence over a platform install with the same id.
-3. **Agent directory shape (per-agent asset aggregation)**: `<uid>/cloud/agents/<aid>/` contains `agent.json` (the sole UI display source — pure spec) + `meta/` (metacognition: COMPETENCE.md + LEARNING_STRATEGIES.md) + `skills/` (SkillStore self-evolved skills, visible only to this agent, not injected into the SkillLoader system-prompt block). Deleting an agent = `rm -rf <aid>/`, no cascade. See `docs/plans/agent-as-directory.md`. **Do not** revert to the old top-level `meta/` or `PC/skills/` (the SkillStore default cwd).
+3. **Agent directory shape (per-agent asset aggregation)**: `<uid>/cloud/agents/<aid>/` contains `agent.json` (the sole UI display source — pure spec) + `meta/` (metacognition: COMPETENCE.md + LEARNING_STRATEGIES.md) + `skills/` (SkillStore self-evolved skills, visible only to this agent, not injected into the SkillLoader system-prompt block). Deleting an agent = `rm -rf <aid>/`, no cascade. **Do not** revert to the old top-level `meta/` or `PC/skills/` (the SkillStore default cwd).
4. **`search/` indexes are purely derived**: never synced; mtime+size reconcile self-heals before each query; 1-second debounced flush + force flush on `before-quit`; rebuild automatically on schema change or corruption.
5. **The KB vector store is NOT cloud-synced (batch 2 decision)**: only `cloud/contexts/` source files cross devices; the `vector.db` + its `config.json` live at `<uid>/local/contexts/.kb/` (machine-private, NOT under cloud/). Each device's startup `kb_indexer.reconcile(uid)` rebuilds the index from the synced sources idempotently by sha1. One-shot migration `util/migrate-kb-to-local.ts` moves a pre-existing `<uid>/cloud/contexts/.kb/` directory into local/ on first launch (stamped under `.migrations`). **Why path mirrors cloud/contexts/ under local/contexts/**: keeps the mental model "KB lives under .../contexts/.kb/" intact across the cloud/local divide, so the cloud→local split is reflected in the path without renaming the `.kb` segment. **Why this changed from the earlier "vector.db IS synced" rule**: a single-doc edit kicks a hundreds-of-MB whole-DB re-upload — economics broke down once we wired the actual sync engine. Index rebuild is per-spec idempotent, so reuse it. **Journal mode is still DELETE, not WAL** (no benefit from switching; `kb_files` rows already use it).
6. **Multi-device sync (plan: `PC/docs/plans/multi-device-sync.md`)**: COS bucket prefix = **`private/<oauth_user_id>/`** (NOT the local 8-digit `uid`); per-machine path stays `<container>/data/<local_uid>/cloud/`, and `private/<oauth_user_id>/` is the cloud-side mirror. **Why the `private/` namespace**: leaves the rest of the bucket (e.g. `public/marketplace/skills/`) freely STS-scoped for non-user-private content; without it, widening any future public scope risks colliding with user data prefixes. **Why oauth_user_id (not local uid)**: same OAuth user has different local `uid` on each device — using local uid would scatter cloud data. The `features/sync/` engine is the **sole** writer of cloud→COS traffic and `private/<oauth_user_id>/cloud/*` keys; `features/sync/manifest.ts::commitManifest` (POST `/sync/manifest/commit`) is the **sole** writer of `private/<oauth_user_id>/manifest.json` (Server enforces atomicity via per-user lock + generation CAS + quota delta; see `Server/CLAUDE.md` §4). PC's STS scope is `private/<oauth_user_id>/*` for read, `private/<oauth_user_id>/cloud/*` for write — `manifest.json` is read-only for PC by design. **Five canonical sync triggers** funnel through `DirtyBus` / `engine.syncNow`: startup (post-marketplace reconcile), debounce-fired dirty event, manual ("立即同步" button), 5-minute periodic safety-net ticker (`engine.ts::PERIODIC_INTERVAL_MS`, bound to engine lifetime so the master switch gates it for free), before-quit. The 30s global floor between auto runs (`engine.ts::GLOBAL_FLOOR_MS`) applies to all except manual + before-quit (which bypass it intentionally; periodic respects it — a recent pass for any reason defers the next tick). **3-way pull classification** (`manifest.ts::classifyPullCandidates`, called from `_passBody` step 5): every path `diff()` puts into `need_pull` is routed by the (base, disk, remote) sha tuple — case A `base==disk` (local untouched, remote moved) → straight overwrite, **never push**; case B `base==remote` (local moved, remote untouched) → reclassify to push (legacy 3-way diff); case C three-way divergence → content-aware merge, then push the merge result back. **Why case A must not push**: a path whose on-disk bytes differ from base usually came from a prior merge's residue, NOT a fresh user edit — pushing it would publish the engine's own historical output as if it were a user change. We shipped this bug for a while and saw `pulled: 4 / pushed: 0` every pass for the same set of jsonl + mutable-json paths; the case A bucket is the load-bearing fix.
@@ -150,7 +138,7 @@ Top-level three-way split: **cloud** (user-private, synced across devices) / **l
- **`LOCAL_DATA_VERSION` data-version gate** (`features/sync/data_version.ts`): every PC build carries an integer schema-version constant; on each pass the engine compares it to `manifest.data_version` post-fetch and aborts with `pause_reason = 'version_outdated'` when `remote > LOCAL_DATA_VERSION` (Server mirrors the same gate on commit → `client_outdated` / 50404, which the engine handles as a sticky pause rather than a retry-able error). **When to bump + when NOT to + bumping procedure** are the MUST / MUST-NOT checklists in `data_version.ts`'s header comment — that file is the single source of truth for the call; don't restate the criteria elsewhere. **Why a third version field** when `manifest_v` and per-entry `schema_v` exist: they answer different questions — `manifest_v` is the manifest JSON's own structure, `schema_v` is per-file content format (with consumption logic still TODO in `merge/`), and `data_version` is the *global* application-data contract gate. Don't conflate them.
8. **`<uid>/local/cache/<bucket>/` is the user-clearable cache umbrella**. Anything dropped under `cache/` is fair game for the "clear cache" UI button (`features/cache_clearable.ts::clearAllClearable`). Distinct from `config/` (user prefs — never auto-cleared) and `biz/` (server-sourced reference data — refresh-on-need but not user-clearable). The marketplace content cache lives at `cache/marketplace/{agents,skills}/<id>/` with a per-entry `_cache.json` carrying `(server_id, version, published_at, fetched_at, last_used_at)`; entry-time sweep evicts entries idle > 7d AND LRU-trims when total > 100 MB (see `features/marketplace_cache.ts::sweepIfNeeded`, called once per `openMarketplace`). **Note**: `local/marketplace/{agents,skills}/<id>/` (the install target — constraint 2) is NOT under `cache/` and is NOT user-clearable; only the staging cache for browsing detail is. Future migration: `file_cache/` and `tool-results/` will move under `cache/` so a single "clear cache" hits all of them — DON'T yet rely on this; each owner module updates its path helper before the move.
-9. **`<uid>/local/signals/<yyyy-mm-dd>.jsonl` is the expert-signal stream** (T0/T1 user-behavior records — retry / skip / form_left_blank / silence / tool_failure / accept / correction / reject / edit / skill_advertised / skill_invoked / skill_ineffective / agent_dispatched). Local-only, append-only, daily-rotated. **Sole writer for the features layer = `features/expert_signals/emitSignal`**; chokepoint sites are `features/group_chat/bus.ts::runTurn` end + `_enqueueBody` USER_ID branch + `features/group_chat/index.ts::{retryStep, skipStep, markFormSubmittedAndDispatch}` + `features/group_chat/plan_executor.ts::dispatchReady`. **Model-layer signals route through `ChatOptions.{onSkillAdvertised, onSkillInvoked}` callbacks** drained at bus turn-end (`expert_signals/turn_hooks.ts::createSkillTurnBuffer`); **the model layer must never `import { emitSignal }` directly** (would reverse-import features/ and violate §3 layering). **Why centralized**: scattered `emitSignal` across `features/*` (or, post-callback, scattered model-layer imports) violates the single-dispatch-primitive rule (§5) and produces signals whose order disagrees with turn order — downstream reflection / patch suggester / critic (phase 1+) all assume turn-aligned causality. **`turn_id` convention**: every signal's `turn_id` is the agent msg id the signal reacts to or is produced by (agent-side = self; user-reaction = previous agent msg; retry/skip = `step.output_msg_id`; agent_dispatched = synthetic `<cid>:plan:dispatch:<ts>:<i>` since no commander msg id is plumbed through plan_executor yet). Direct JOIN on `turn_id` recovers cross-signal causality (skill_invoked × correction, retry × tool_failure). Phase 0 has NO redaction layer (see §10). Plans: `docs/plans/expert-signals-phase-0.md`, `docs/plans/expert-signals-skill-attribution.md` (phase 0 follow-up for skill / agent attribution).
+9. **`<uid>/local/signals/<yyyy-mm-dd>.jsonl` is the expert-signal stream** (T0/T1 user-behavior records — retry / skip / form_left_blank / silence / tool_failure / accept / correction / reject / edit / skill_advertised / skill_invoked / skill_ineffective / agent_dispatched). Local-only, append-only, daily-rotated. **Sole writer for the features layer = `features/expert_signals/emitSignal`**; chokepoint sites are `features/group_chat/bus.ts::runTurn` end + `_enqueueBody` USER_ID branch + `features/group_chat/index.ts::{retryStep, skipStep, markFormSubmittedAndDispatch}` + `features/group_chat/plan_executor.ts::dispatchReady`. **Model-layer signals route through `ChatOptions.{onSkillAdvertised, onSkillInvoked}` callbacks** drained at bus turn-end (`expert_signals/turn_hooks.ts::createSkillTurnBuffer`); **the model layer must never `import { emitSignal }` directly** (would reverse-import features/ and violate §3 layering). **Why centralized**: scattered `emitSignal` across `features/*` (or, post-callback, scattered model-layer imports) violates the single-dispatch-primitive rule (§5) and produces signals whose order disagrees with turn order — downstream reflection / patch suggester / critic (phase 1+) all assume turn-aligned causality. **`turn_id` convention**: every signal's `turn_id` is the agent msg id the signal reacts to or is produced by (agent-side = self; user-reaction = previous agent msg; retry/skip = `step.output_msg_id`; agent_dispatched = synthetic `<cid>:plan:dispatch:<ts>:<i>` since no commander msg id is plumbed through plan_executor yet). Direct JOIN on `turn_id` recovers cross-signal causality (skill_invoked × correction, retry × tool_failure). Phase 0 has NO redaction layer (see §10).
**uid lifecycle** (`features/users.ts`): on startup, `initActiveUser()` reads or creates `users.json` and calls `activateUser(uid)` (skeleton mkdir + injecting `process.env.CORE_AGENT_AUTH_DIR` + clearing caches + running migrations including `migrateLegacySessionIds`). All user-scoped features get the uid via `getActiveUserId()` (throws if not activated). One active uid at a time for now. **uid shape is unconstrained** — OrkasOpen first-boot generates 8-digit numeric via `genUserId()`; Hosted Orkas reconciles the local uid with the OAuth-issued user_id via `reconcileDevUid`, which today carries through whatever the server emits (UUID with dashes is real). Code that uses `<uid>/...` as a path segment is fine with any shape (single segment, no parsing). Code that USED to embed `<uid>-` into session_id has been removed (§5 — session_id is now `<kind>-<tail>`); never reintroduce that pattern.
@@ -320,7 +308,7 @@ Third-party integrations attach via MCP servers (`@modelcontextprotocol/sdk`); O
**Catalog bundles** (`CatalogEntry.bundle_member_ids`): a catalog entry can declare itself a **UI-only grouping** that fans out on connect. The bundle entry has `transport_template: null` and never becomes a ConnectorInstance — `manager.connectViaOAuth` runs OAuth once for the bundle (Server returns a grant with the union of all members' scopes, e.g. `_SCOPES_BY_CATALOG_ID['google-workspace']`) and provisions one instance per member with a deep-cloned grant. **Model view is unchanged**: each member stays a distinct connector with its own tools. Renderer hides member entries, shows the bundle card; disconnect / enable-toggle fan out across all members. Used today for `google-workspace`.
-**OrkasOpen**: nothing in `features/connectors/` is stripped — connectors are 100% client-side. (A future marketplace-style recommended-connectors UI would gate the same way `marketplace_dev` does — see §9 OrkasOpen contract.)
+**OrkasOpen**: nothing in `features/connectors/` is stripped — connectors, including Google Workspace catalog entries and stdio adapters, sync to the open-source build. (A future marketplace-style recommended-connectors UI would gate the same way `marketplace_dev` does — see §9 OrkasOpen contract.)
---
@@ -332,7 +320,7 @@ Third-party integrations attach via MCP servers (`@modelcontextprotocol/sdk`); O
- **Embedder is fixed at `bge-small-zh-v1.5`, 512 dims**; switching models requires a full rebuild (`config.json` lock prevents accidental swaps). The model is ~95 MB and ships via the installer's `extraResources`, so zero download / zero network at runtime.
- **Journal mode is DELETE, not WAL** — kept for continuity with existing `kb_files` rows (the original "WAL would tear cloud sync" rationale is moot now that vector.db is machine-private, but switching brings no benefit).
- **No `worker_threads` for multiple ONNX sessions**: empirically the native layer SIGSEGVs (OpenMP threadpool + concurrent allocator init is a known dangerous combination); for true parallelism use `child_process`.
-- The model side may use only `kb_search` / `kb_read` tools; **`cat` / `rg` access to `$contexts_dir/` is forbidden** (stated in `chat_core.md`).
+- The model side may use only `kb_search` / `kb_read` tools; **`cat` / `rg` access to `$contexts_dir/` is forbidden**.
- Chunk cap is `EMBED_MAX_CHARS=400` chars (matching the 512-token window); no overlap across segments to avoid topic pollution.
- `_INDEX.md` is generated only at the root of contexts for users browsing in Finder; **the model does not read it**.
- Cloud-sync conflicts take the newer mtime; the loser runs `reconcile` at startup to backfill via sha1; the `kb_files` table is the manifest, no separate manifest file is needed.
diff --git a/bin/run-skill.cjs b/bin/run-skill.cjs
index 4047dba1..38c11d15 100755
--- a/bin/run-skill.cjs
+++ b/bin/run-skill.cjs
@@ -61,13 +61,6 @@ function parseArgs(argv) {
return { skillId, scriptBase, scriptArgs };
}
-// Sibling directories of WS_ROOT that are NOT user-id dirs. See paths.ts —
-// users.json / window-state.json are files (filtered by isDirectory above);
-// these are the directories we must not treat as uids.
-const RESERVED_WS_DIRS = new Set([
- 'logs', 'config', 'user_workspaces', 'builtin',
-]);
-
function locateSkillScript(skillId, scriptBase) {
const wsRoot = process.env.ORKAS_WORKSPACE_ROOT
|| process.env.ORKAS_WS_ROOT
@@ -75,17 +68,26 @@ function locateSkillScript(skillId, scriptBase) {
// Candidate skill dirs — mirror SkillRegistry's
// [<uid>/cloud/skills, <uid>/local/marketplace/skills] resolution, with a
// dev/source-tree fallback for unpacked builtins.
- const skillDirs = [];
+ //
// Per-user skills live under <ws>/<uid>/cloud/skills/<id> and
// <ws>/<uid>/local/marketplace/skills/<id>. uid can be numeric or a UUID,
- // so we scan every non-reserved subdirectory rather than regex-matching.
+ // so we scan every subdirectory rather than regex-matching. Skip the deny-
+ // list approach (had to hand-maintain a list of `logs / config / ...`
+ // sibling names that would drift the moment a new top-level data dir gets
+ // added under PC/CLAUDE.md §4); instead, only KEEP an entry if it actually
+ // contains the skill under one of the two SkillRegistry-aware shapes. This
+ // is the strict invariant — a real uid dir always has one of those shapes
+ // when the skill is installed for that user; any sibling dir that doesn't
+ // is irrelevant by construction.
+ const skillDirs = [];
try {
for (const entry of fs.readdirSync(wsRoot, { withFileTypes: true })) {
if (!entry.isDirectory()) continue;
if (entry.name.startsWith('.')) continue;
- if (RESERVED_WS_DIRS.has(entry.name)) continue;
- skillDirs.push(path.join(wsRoot, entry.name, 'cloud', 'skills', skillId));
- skillDirs.push(path.join(wsRoot, entry.name, 'local', 'marketplace', 'skills', skillId));
+ const cloud = path.join(wsRoot, entry.name, 'cloud', 'skills', skillId);
+ const marketplace = path.join(wsRoot, entry.name, 'local', 'marketplace', 'skills', skillId);
+ if (fs.existsSync(cloud)) skillDirs.push(cloud);
+ if (fs.existsSync(marketplace)) skillDirs.push(marketplace);
}
} catch { /* no data dir yet */ }
if (process.env.ORKAS_PC_DIR) {
diff --git a/bootstrap.cjs b/bootstrap.cjs
index 0fed8e91..bd55dcd6 100644
--- a/bootstrap.cjs
+++ b/bootstrap.cjs
@@ -18,6 +18,38 @@
'use strict';
require('./src/main/install-data-root.cjs');
+const fs = require('node:fs');
+const path = require('node:path');
+
+function configurePackagedEsbuildBinary() {
+ if (!process.versions.electron || !process.resourcesPath || process.env.ESBUILD_BINARY_PATH) {
+ return;
+ }
+
+ const platformPackages = {
+ 'darwin:arm64': ['@esbuild', 'darwin-arm64', 'bin', 'esbuild'],
+ 'darwin:x64': ['@esbuild', 'darwin-x64', 'bin', 'esbuild'],
+ 'linux:arm64': ['@esbuild', 'linux-arm64', 'bin', 'esbuild'],
+ 'linux:x64': ['@esbuild', 'linux-x64', 'bin', 'esbuild'],
+ 'win32:arm64': ['@esbuild', 'win32-arm64', 'esbuild.exe'],
+ 'win32:ia32': ['@esbuild', 'win32-ia32', 'esbuild.exe'],
+ 'win32:x64': ['@esbuild', 'win32-x64', 'esbuild.exe'],
+ };
+ const parts = platformPackages[`${process.platform}:${process.arch}`];
+ if (!parts) return;
+
+ const bin = path.join(
+ process.resourcesPath,
+ 'app.asar.unpacked',
+ 'node_modules',
+ ...parts,
+ );
+ if (fs.existsSync(bin)) {
+ process.env.ESBUILD_BINARY_PATH = bin;
+ }
+}
+
+configurePackagedEsbuildBinary();
require('tsx/cjs');
require('tsx/esm/api').register();
diff --git a/package-lock.json b/package-lock.json
index 6c6e181a..b33337d6 100644
--- a/package-lock.json
+++ b/package-lock.json
@@ -1,12 +1,14 @@
{
"name": "orkas",
"version": "2026.4.30",
+ "version": "2026.5.21",
"lockfileVersion": 3,
"requires": true,
"packages": {
"": {
"name": "orkas",
"version": "2026.4.30",
+ "version": "2026.5.21",
"hasInstallScript": true,
"license": "MIT",
"dependencies": {
@@ -19,7 +21,6 @@
"fastembed": "^2.1.0",
"jimp": "^1.6.1",
"mammoth": "^1.12.0",
- "onnxruntime-node": "^1.24.3",
"pdfjs-dist": "^5.6.205",
"sqlite-vec": "^0.1.9",
"tsx": "^4.21.0",
@@ -6444,29 +6445,6 @@
"node": ">=6"
}
},
- "node_modules/onnxruntime-common": {
- "version": "1.24.3",
- "resolved": "https://registry.npmjs.org/onnxruntime-common/-/onnxruntime-common-1.24.3.tgz",
- "integrity": "sha512-GeuPZO6U/LBJXvwdaqHbuUmoXiEdeCjWi/EG7Y1HNnDwJYuk6WUbNXpF6luSUY8yASul3cmUlLGrCCL1ZgVXqA==",
- "license": "MIT"
- },
- "node_modules/onnxruntime-node": {
- "version": "1.24.3",
- "resolved": "https://registry.npmjs.org/onnxruntime-node/-/onnxruntime-node-1.24.3.tgz",
- "integrity": "sha512-JH7+czbc8ALA819vlTgcV+Q214/+VjGeBHDjX81+ZCD0PCVCIFGFNtT0V4sXG/1JXypKPgScQcB3ij/hk3YnTg==",
- "hasInstallScript": true,
- "license": "MIT",
- "os": [
- "win32",
- "darwin",
- "linux"
- ],
- "dependencies": {
- "adm-zip": "^0.5.16",
- "global-agent": "^3.0.0",
- "onnxruntime-common": "1.24.3"
- }
- },
"node_modules/openai": {
"version": "6.26.0",
"license": "Apache-2.0",
diff --git a/package.json b/package.json
index d1cb43ec..942e4424 100644
--- a/package.json
+++ b/package.json
@@ -1,6 +1,7 @@
{
"name": "orkas",
"version": "2026.4.30",
+ "version": "2026.5.21",
"private": true,
"license": "MIT",
"description": "Orkas desktop — Electron + Node one-piece app",
@@ -19,6 +20,7 @@
"rebuild:sqlite:electron": "node scripts/swap-sqlite-abi.mjs electron",
"rebuild:sqlite:node": "node scripts/swap-sqlite-abi.mjs node",
"postinstall": "node scripts/swap-sqlite-abi.mjs electron && node scripts/fetch-embedding-model.mjs",
+ "prepare": "node scripts/install-git-hooks.mjs",
"prepack": "node scripts/swap-sqlite-abi.mjs electron",
"predist": "node scripts/swap-sqlite-abi.mjs electron",
"pack": "electron-builder --dir",
@@ -48,7 +50,6 @@
"fastembed": "^2.1.0",
"jimp": "^1.6.1",
"mammoth": "^1.12.0",
- "onnxruntime-node": "^1.24.3",
"pdfjs-dist": "^5.6.205",
"sqlite-vec": "^0.1.9",
"tsx": "^4.21.0",
@@ -102,7 +103,6 @@
"node_modules/bindings/**/*",
"node_modules/file-uri-to-path/**/*",
"node_modules/sqlite-vec*/**/*",
- "node_modules/onnxruntime-node/**/*",
"node_modules/fastembed/**/*"
],
"extraResources": [
@@ -136,6 +136,9 @@
],
"icon": "src/resources/icons/icon.icns"
},
+ "dmg": {
+ "writeUpdateInfo": false
+ },
"win": {
"target": "nsis",
"icon": "src/resources/icons/icon.ico"
diff --git a/scripts/codesign-adhoc.cjs b/scripts/codesign-adhoc.cjs
index fcc4657b..170993ec 100644
--- a/scripts/codesign-adhoc.cjs
+++ b/scripts/codesign-adhoc.cjs
@@ -11,16 +11,76 @@
'use strict';
const { execSync } = require('node:child_process');
+const fs = require('node:fs');
const path = require('node:path');
+const ARCH_NAMES = {
+ 0: 'ia32',
+ 1: 'x64',
+ 2: 'armv7l',
+ 3: 'arm64',
+ 4: 'universal',
+};
+
+function removeIfExists(target) {
+ if (fs.existsSync(target)) fs.rmSync(target, { recursive: true, force: true });
+}
+
+function listDirs(dir) {
+ if (!fs.existsSync(dir)) return [];
+ return fs.readdirSync(dir, { withFileTypes: true })
+ .filter((entry) => entry.isDirectory())
+ .map((entry) => entry.name);
+}
+
+function pruneOnnxRuntimePackage(pkgDir, targetArch) {
+ const binDir = path.join(pkgDir, 'bin');
+ if (!fs.existsSync(binDir)) return;
+
+ const keepArch = targetArch === 'arm64' || targetArch === 'x64' ? targetArch : null;
+ for (const napiDirName of listDirs(binDir)) {
+ const napiDir = path.join(binDir, napiDirName);
+ for (const platformDirName of listDirs(napiDir)) {
+ if (platformDirName !== 'darwin') {
+ removeIfExists(path.join(napiDir, platformDirName));
+ }
+ }
+
+ if (!keepArch) continue;
+ const darwinDir = path.join(napiDir, 'darwin');
+ for (const archDirName of listDirs(darwinDir)) {
+ if (archDirName !== keepArch) {
+ removeIfExists(path.join(darwinDir, archDirName));
+ }
+ }
+ }
+}
+
+function prunePackedNativePayload(appPath, targetArch) {
+ const nodeModules = path.join(appPath, 'Contents', 'Resources', 'app.asar.unpacked', 'node_modules');
+ const packages = [
+ path.join(nodeModules, 'onnxruntime-node'),
+ path.join(nodeModules, 'fastembed', 'node_modules', 'onnxruntime-node'),
+ ];
+
+ for (const pkgDir of packages) {
+ pruneOnnxRuntimePackage(pkgDir, targetArch);
+ }
+}
+
module.exports = async function afterPack(context) {
if (context.electronPlatformName !== 'darwin') return;
+ const appName = context.packager.appInfo.productFilename;
+ const appPath = path.join(context.appOutDir, `${appName}.app`);
+ const targetArch = ARCH_NAMES[context.arch] || process.arch;
+
+ prunePackedNativePayload(appPath, targetArch);
+
if (process.env.CSC_LINK || process.env.CSC_NAME) {
console.log('[codesign-adhoc] formal signing env detected; skipping ad-hoc signing');
return;
}
- const appName = context.packager.appInfo.productFilename;
- const appPath = path.join(context.appOutDir, `${appName}.app`);
+
console.log(`[codesign-adhoc] signing ${appPath}`);
execSync(`codesign --force --deep --sign - "${appPath}"`, { stdio: 'inherit' });
// 校验签名(失败时直接 throw 中断 build,避免发出未签的包)
diff --git a/scripts/install-git-hooks.mjs b/scripts/install-git-hooks.mjs
new file mode 100644
index 00000000..89d72035
--- /dev/null
+++ b/scripts/install-git-hooks.mjs
@@ -0,0 +1,52 @@
+#!/usr/bin/env node
+/**
+ * Auto-install the repo-local git hooks directory.
+ *
+ * Wired into PC/package.json `prepare`, which `npm install` runs whenever
+ * a contributor sets up the repo. Effect:
+ *
+ * git config core.hooksPath <repo-root>/.githooks
+ *
+ * so the commit-msg hook (the one that enforces `Prompt audit:` on
+ * prompt-facing commits) is on by default — no per-machine ceremony.
+ *
+ * Best-effort: silently no-ops when
+ * - cwd isn't inside a git checkout (e.g. tarball install)
+ * - the .githooks/ directory isn't present
+ * - `git` isn't on PATH
+ * so a botched local setup never blocks `npm install`.
+ */
+
+import { execSync } from 'node:child_process';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+
+function tryGitTopLevel() {
+ try {
+ return execSync('git rev-parse --show-toplevel', { stdio: ['ignore', 'pipe', 'ignore'] })
+ .toString()
+ .trim();
+ } catch {
+ return null;
+ }
+}
+
+const repoRoot = tryGitTopLevel();
+if (!repoRoot) {
+ // Not a git checkout — npm install from a published tarball, CI install
+ // with .git stripped, etc. Silently no-op.
+ process.exit(0);
+}
+
+const hooksDir = path.join(repoRoot, '.githooks');
+if (!fs.existsSync(hooksDir)) {
+ // No hooks vendored in this checkout — nothing to wire up.
+ process.exit(0);
+}
+
+try {
+ execSync(`git config core.hooksPath "${hooksDir}"`, { stdio: 'ignore' });
+ console.log(`[install-git-hooks] core.hooksPath -> ${hooksDir}`);
+} catch (err) {
+ console.warn(`[install-git-hooks] skipped: ${(err && err.message) || err}`);
+}
diff --git a/src/core-agent/src/agent/runner.ts b/src/core-agent/src/agent/runner.ts
index 60b1eba7..7a728c47 100644
--- a/src/core-agent/src/agent/runner.ts
+++ b/src/core-agent/src/agent/runner.ts
@@ -10,7 +10,6 @@ import {
import { createLogger } from "../shared/logger.js";
import type { CoreAgentConfig } from "../config/schema.js";
import type { EvolutionConfig } from "../evolution/types.js";
-import { detectUserCorrection } from "../evolution/metacognition.js";
import type { LLMProvider, CompletionResult } from "../providers/base.js";
import { ProviderRegistry } from "../providers/registry.js";
import type { AgentTool, ToolContext } from "../tools/base.js";
@@ -18,8 +17,6 @@ import { toToolDefinition } from "../tools/base.js";
import { getBuiltinTools } from "../tools/builtin.js";
import { SkillStore } from "../evolution/skill-store.js";
import { createSkillManageTool } from "../evolution/skill-tools.js";
-import { shouldReflect, buildAdaptiveReviewPrompt } from "../evolution/metacognition.js";
-import type { MetacognitionConfig, RunMetrics } from "../evolution/types.js";
import { Session } from "./session.js";
import type { AgentRunParams, AgentRunResult, AgentRunMeta, AgentRunEvent } from "./types.js";
@@ -511,163 +508,6 @@ export class AgentRunner {
return this.skillStore;
}
- /**
- * Evaluate whether a completed run warrants metacognitive reflection.
- * Converts AgentRunResult.meta into RunMetrics, calls shouldReflect(),
- * and if triggered, returns the adaptive review prompt. Returns null
- * if no reflection is needed.
- *
- * @param result The completed run result
- * @param competence Current COMPETENCE.md content (optional)
- * @param strategies Current LEARNING_STRATEGIES.md content (optional)
- */
- evaluateReflection(
- result: AgentRunResult,
- competence?: string,
- strategies?: string,
- /** The user message that triggered this run. When provided, `buildRunMetrics`
- * runs `detectUserCorrection` on it to populate `userCorrections`. Wired
- * through from the caller per plan §6.1 so the same heuristic feeds both
- * this metric and the bus-side `correction` signal (no double-judgment). */
- inboundMessage?: string,
- ): { prompt: string; primaryFocus: string; score: number } | null {
- const metaConfig = this.config.evolution.metacognition as MetacognitionConfig;
- if (!metaConfig.enabled) {
- log.debug('evaluateReflection: metacognition disabled in config');
- return null;
- }
-
- const metrics = this.buildRunMetrics(result, inboundMessage);
- log.debug(`evaluateReflection: metrics toolCalls=${metrics.toolCalls} errors=${metrics.errorCount} errorKind=${metrics.errorKind} transient=${metrics.transientErrorCount} corrections=${metrics.userCorrections} skills=[${metrics.skillsLoaded.join(',')}]`);
-
- const reflection = shouldReflect(metrics, metaConfig, competence);
- if (!reflection.shouldReflect) {
- log.info(`evaluateReflection: no reflection needed (score=${reflection.score.toFixed(2)} threshold=${metaConfig.reflectThreshold} signals=${reflection.signals.length})`);
- return null;
- }
-
- // Build a conversation digest so the reflection LLM has context
- // about what actually happened (it runs in an ephemeral session).
- const digest = this.buildConversationDigest(result, metrics, reflection.signals);
-
- const prompt = buildAdaptiveReviewPrompt(
- reflection.primaryFocus,
- competence || '',
- strategies || '',
- undefined, // skillHealthReport — future
- digest,
- );
- return {
- prompt,
- primaryFocus: reflection.primaryFocus,
- score: reflection.score,
- };
- }
-
- /**
- * Build a concise digest of the conversation for the reflection LLM.
- * Since the reflection runs in an ephemeral session, it needs this
- * context to understand what happened and make meaningful updates.
- */
- private buildConversationDigest(
- result: AgentRunResult,
- metrics: RunMetrics,
- signals: Array<{ name: string; reason: string }>,
- ): string {
- const lines: string[] = [];
-
- // What tools were used
- if (metrics.toolNames.length > 0) {
- lines.push(`Tools used: ${metrics.toolNames.join(', ')}`);
- }
-
- // Which skills were loaded
- if (metrics.skillsLoaded.length > 0) {
- lines.push(`Skills loaded: ${metrics.skillsLoaded.join(', ')}`);
- }
-
- // Error info with classification
- if (result.meta.error) {
- lines.push(`Error: [${result.meta.error.kind}] ${result.meta.error.message}`);
- if (metrics.recovered) lines.push('Outcome: recovered from the error and completed the task');
- }
- if (metrics.errorKind !== 'none') {
- const label = metrics.errorKind === 'transient' ? 'transient (network / timeout / rate limit)'
- : metrics.errorKind === 'mixed' ? 'mixed (transient + permanent)'
- : 'permanent (non-transient)';
- lines.push(`Error kind: ${label} (${metrics.transientErrorCount} transient occurrence(s))`);
- }
-
- // Trigger signals
- lines.push(`Trigger signals: ${signals.map(s => `${s.name}(${s.reason})`).join('; ')}`);
-
- // The agent's final response (truncated) — this is the most important
- // context for understanding what actually happened
- if (result.text) {
- const maxLen = 1500;
- const text = result.text.length > maxLen
- ? result.text.slice(0, maxLen) + '\n…(truncated)'
- : result.text;
- lines.push('', '--- Final assistant reply ---', text);
- }
-
- // Session message history (user messages only, for additional context)
- const messages = this.session.getMessages();
- const userMessages = messages
- .filter(m => m.role === 'user')
- .map(m => {
- const textParts = m.content
- .filter((c: any) => c.type === 'text')
- .map((c: any) => c.text)
- .join('');
- return textParts;
- })
- .filter(t => t.length > 0);
-
- if (userMessages.length > 0) {
- lines.push('', '--- User messages ---');
- // Include last few user messages (most relevant)
- const recent = userMessages.slice(-3);
- for (const msg of recent) {
- const truncated = msg.length > 300 ? msg.slice(0, 300) + '…' : msg;
- lines.push(`- ${truncated}`);
- }
- }
-
- return lines.join('\n');
- }
-
- /** Convert AgentRunResult.meta into RunMetrics for the trigger evaluator.
- * When `inboundMessage` is provided, runs `detectUserCorrection` to set
- * `userCorrections` (otherwise 0). Reuses the same heuristic the bus-side
- * expert_signals extractor uses — same input, same boolean. */
- private buildRunMetrics(result: AgentRunResult, inboundMessage?: string): RunMetrics {
- const meta = result.meta;
- const transient = meta.transientToolErrors ?? 0;
- const permanent = meta.permanentToolErrors ?? 0;
- const hadErrors = meta.error !== undefined || transient + permanent > 0;
- const totalToolErrors = transient + permanent;
- let errorKind: import('../evolution/types.js').ErrorKind = 'none';
- if (totalToolErrors > 0) {
- if (transient > 0 && permanent === 0) errorKind = 'transient';
- else if (permanent > 0 && transient === 0) errorKind = 'permanent';
- else errorKind = 'mixed';
- } else if (meta.error) {
- errorKind = 'permanent';
- }
- return {
- toolCalls: meta.toolLoops,
- toolNames: meta.toolNames ?? [],
- skillsLoaded: meta.skillsLoaded ?? [],
- hadErrors,
- recovered: hadErrors && result.text.length > 0,
- errorCount: totalToolErrors + (meta.error ? 1 : 0),
- userCorrections: inboundMessage && detectUserCorrection(inboundMessage) ? 1 : 0,
- errorKind,
- transientErrorCount: transient,
- };
- }
-
private async buildSystemPromptWithEvolution(basePrompt: string): Promise<string> {
if (!this.skillStore || !this.config.evolution.enabled) {
return basePrompt;
diff --git a/src/core-agent/src/evolution/index.ts b/src/core-agent/src/evolution/index.ts
index 113a4f14..c96e655c 100644
--- a/src/core-agent/src/evolution/index.ts
+++ b/src/core-agent/src/evolution/index.ts
@@ -15,5 +15,5 @@ export {
detectUserCorrection,
emptyRunMetrics,
shouldReflect,
- buildAdaptiveReviewPrompt,
+ buildReviewPrompt,
} from "./metacognition.js";
diff --git a/src/core-agent/src/evolution/metacognition.ts b/src/core-agent/src/evolution/metacognition.ts
index 78c702cd..34709b0e 100644
--- a/src/core-agent/src/evolution/metacognition.ts
+++ b/src/core-agent/src/evolution/metacognition.ts
@@ -201,103 +201,81 @@ function extractSection(text: string, headingKeywords: string[]): string | null
return captured.length > 0 ? captured.join('\n').trim() : null;
}
-// ── Adaptive review prompt generation ───────────────────────────────────
+// ── Review prompt generation ────────────────────────────────────────────
/**
- * Build a review prompt tailored to the trigger reason, the agent's
- * self-assessment, and available learning strategies.
+ * Build the reflection LLM prompt for one agent over a recent activity
+ * transcript.
*
- * @param primaryFocus Trigger reason (error_recovery, user_correction, etc.)
- * @param competence Current COMPETENCE.md content
- * @param strategies Current LEARNING_STRATEGIES.md content
- * @param skillHealthReport Optional skill health report
- * @param conversationDigest Summary of what happened in the conversation
+ * Single open-ended prompt (no `primaryFocus` branching) per
+ * `docs/plans/reflection-redesign.md` §2.3 — the LLM reads the transcript
+ * and decides what's worth saving. Eliminating the 7-case switch removes
+ * a window-aggregation ambiguity (multi-focus runs would need contradictory
+ * directives) and keeps the system prompt cache-stable across calls.
+ *
+ * @param competence Current COMPETENCE.md content (capped 3000 chars by the meta tool)
+ * @param strategies Current LEARNING_STRATEGIES.md content (capped 4000 chars)
+ * @param transcript Activity transcript from `reflection-transcript.buildTranscript`
*/
-export function buildAdaptiveReviewPrompt(
- primaryFocus: string,
+export function buildReviewPrompt(
competence: string,
strategies: string,
- skillHealthReport?: string,
- conversationDigest?: string,
+ transcript: string,
): string {
- const base = 'Based on the conversation digest below, consider improving your skills and self-awareness.\n\n'
+ const base = 'Based on the activity transcript below, review and update your self-understanding.\n\n'
+ '**Important: transient errors (network timeouts, dropped connections, rate limits) are environmental issues, not skill or capability deficits. '
+ 'Do not mark them as weaknesses in COMPETENCE.md, do not modify or delete the related skills, '
+ 'and do not advise avoiding the tool in LEARNING_STRATEGIES.md.**\n\n';
- let focus: string;
- switch (primaryFocus) {
- case 'error_recovery':
- focus = 'You recovered from an error during this task. '
- + 'Extract the resolution into a reusable skill, focusing on the non-obvious steps that made recovery succeed.';
- break;
- case 'user_correction':
- focus = 'The user corrected your approach. '
- + 'Identify what you got wrong, then create or update a skill that captures the correct way. '
- + 'Also update COMPETENCE.md to record this weakness.';
- break;
- case 'skill_ineffective':
- focus = 'A skill you loaded did not help on this task. '
- + 'Diagnose why: outdated? too generic? mismatched? '
- + 'Patch or delete the ineffective skill.';
- break;
- case 'known_weakness':
- focus = 'This task hit one of your known weaknesses. '
- + 'Strengthen this area. If you succeeded, update COMPETENCE.md to reflect the progress.';
- break;
- case 'weakness_succeeded':
- focus = 'A capability you previously flagged as a weakness performed normally on this task. '
- + 'Update COMPETENCE.md: remove or downgrade that weakness entry. '
- + 'Do not modify or delete the related skills — they are working as expected.';
- break;
- case 'periodic_review':
- focus = 'Some time has passed since the last reflection — do a periodic review. '
- + 'Look at COMPETENCE.md and LEARNING_STRATEGIES.md: '
- + 'merge similar entries, drop content that no longer applies or is duplicated, '
- + 'and distill recently-recurring stable patterns into new skills. '
- + 'If there is nothing new worth saving, just say "nothing to save" — do not reflect for the sake of reflecting.';
- break;
- case 'complexity':
- default:
- focus = 'This is a complex task. '
- + 'Consider whether any approach or pattern is worth saving as a skill.';
- break;
- }
-
- const parts = [
+ const focus = 'Look at COMPETENCE.md and LEARNING_STRATEGIES.md: '
+ + 'merge similar entries, drop content that no longer applies or is duplicated, '
+ + 'and distill recently-recurring stable patterns into new skills. '
+ + 'Identify user preferences and domain constraints surfaced in this window — '
+ + 'red lines the user revealed, edits made to your draft, repeated corrections, '
+ + 'workflow preferences. '
+ + 'If there is nothing new worth saving, just say "nothing to save" — '
+ + 'do not reflect for the sake of reflecting.';
+
+ // Imperative writing guidance — prose injection of COMPETENCE/STRATEGIES
+ // into the runtime system prompt is a weak nudge unless entries are written
+ // as actionable rules anchored to trigger conditions. Without this section
+ // reflection tends to produce vague descriptions ("agent should be careful
+ // about X") that the agent reliably ignores. See docs/plans/reflection-redesign.md §9.
+ const writingStyle = 'When you update COMPETENCE.md / LEARNING_STRATEGIES.md or '
+ + 'author a skill, write actionable imperatives — not descriptions.\n'
+ + ' ✗ "Agent should be careful about overly verbose output."\n'
+ + ' ✓ "NEVER exceed 5 bullet points in family-office context replies."\n'
+ + ' ✗ "User seems to prefer concise output."\n'
+ + ' ✓ "WHEN replying to family-office context, ALWAYS lead with the '
+ + 'bottom-line answer before rationale."\n'
+ + 'Use NEVER / ALWAYS / WHEN-THEN structures with concrete trigger conditions.\n'
+ + 'For a new skill\'s `description` field specifically: one imperative line '
+ + 'stating WHEN to use the skill, not what it is about (the description is '
+ + 'the only thing the next turn\'s agent sees before deciding to load the body).\n'
+ + ' ✗ "Q4 earnings analysis skill."\n'
+ + ' ✓ "WHEN handling Q4 earnings for utilities, use trailing-4-quarters not annualized."';
+
+ return [
base,
`**Focus**: ${focus}`,
- ];
-
- if (conversationDigest) {
- parts.push(
- '',
- '**Conversation digest**:',
- conversationDigest,
- );
- }
-
- parts.push(
+ '',
+ `**Writing style**: ${writingStyle}`,
+ '',
+ '**Activity transcript**:',
+ transcript || '(No new activity in the window.)',
'',
'**Current self-assessment (COMPETENCE.md)**:',
competence || '(No self-assessment yet — consider creating one.)',
'',
'**Available learning strategies (LEARNING_STRATEGIES.md)**:',
strategies || '(No strategy log yet — use your own judgment.)',
- );
-
- if (skillHealthReport) {
- parts.push('', '**Skill health report**:', skillHealthReport);
- }
-
- parts.push(
'',
'After reflecting, you can:',
- '1. Create / patch / delete skills via the skill_manage tool',
+ '1. Create / patch / delete skills via the skill_manage tool '
+ + '(no user confirmation needed during reflection — the tool\'s default '
+ + '"confirm with user" guidance is for live turns, not for this call).',
'2. Update COMPETENCE.md via the metacognition tool',
'3. Update LEARNING_STRATEGIES.md via the metacognition tool',
'4. If nothing is worth saving, just say "nothing to save"',
- );
-
- return parts.join('\n');
+ ].join('\n');
}
diff --git a/src/core-agent/src/index.ts b/src/core-agent/src/index.ts
index bc7b8d74..8f97f46c 100644
--- a/src/core-agent/src/index.ts
+++ b/src/core-agent/src/index.ts
@@ -66,7 +66,7 @@ export {
// Evolution (self-improvement)
export { SkillStore, createSkillManageTool } from "./evolution/index.js";
export type { Skill, SkillSummary, SkillFrontmatter } from "./evolution/index.js";
-export { detectUserCorrection, emptyRunMetrics, shouldReflect, buildAdaptiveReviewPrompt } from "./evolution/index.js";
+export { detectUserCorrection, emptyRunMetrics, shouldReflect, buildReviewPrompt } from "./evolution/index.js";
export type { MetacognitionConfig, RunMetrics, TriggerSignal, MetacognitiveReflection } from "./evolution/index.js";
// CLI
diff --git a/src/core-agent/test/evolution.test.ts b/src/core-agent/test/evolution.test.ts
index 1fadafd4..bcacd3b6 100644
--- a/src/core-agent/test/evolution.test.ts
+++ b/src/core-agent/test/evolution.test.ts
@@ -543,74 +543,10 @@ describe("Evolution: AgentRunner integration", () => {
expect(runner.getSkillStore()).toBeNull();
});
- it("evaluateReflection returns null when metacognition is disabled", async () => {
- const store = new SkillStore(skillsDir, makeConfig(skillsDir));
- await store.init();
-
- const registry = new ProviderRegistry();
- const config = createConfig({
- agent: { defaultProvider: "mock", defaultModel: "mock-model" },
- evolution: { enabled: true, skillsDir, metacognition: { enabled: false } },
- });
-
- const runner = new AgentRunner({ config, providers: registry, tools: [], skillStore: store });
- const result = makeRunResult({ toolLoops: 10 });
- expect(runner.evaluateReflection(result)).toBeNull();
- });
-
- it("evaluateReflection returns prompt when metacognition triggers", async () => {
- const store = new SkillStore(skillsDir, makeConfig(skillsDir));
- await store.init();
-
- const registry = new ProviderRegistry();
- const config = createConfig({
- agent: { defaultProvider: "mock", defaultModel: "mock-model" },
- evolution: {
- enabled: true,
- skillsDir,
- metacognition: { enabled: true, reflectThreshold: 0.7 },
- },
- });
-
- const runner = new AgentRunner({ config, providers: registry, tools: [], skillStore: store });
- // Error recovery (weight 0.8) should trigger
- const result = makeRunResult({
- toolLoops: 3,
- error: { kind: "provider_error", message: "temporary" },
- toolNames: ["bash"],
- skillsLoaded: [],
- });
- // result.text is non-empty → recovered = true
- const evaluation = runner.evaluateReflection(result);
- expect(evaluation).not.toBeNull();
- expect(evaluation!.primaryFocus).toBe("error_recovery");
- // Adaptive review prompt = English head (buildAdaptiveReviewPrompt) +
- // English digest body (buildConversationDigest's "Outcome: recovered
- // from the error and completed the task") + the user-supplied
- // result.text appended verbatim.
- expect(evaluation!.prompt).toContain("conversation digest");
- expect(evaluation!.prompt).toContain("recovered from the error");
- expect(evaluation!.prompt).toContain("Task completed");
- });
-
- it("evaluateReflection returns null for simple runs", async () => {
- const store = new SkillStore(skillsDir, makeConfig(skillsDir));
- await store.init();
-
- const registry = new ProviderRegistry();
- const config = createConfig({
- agent: { defaultProvider: "mock", defaultModel: "mock-model" },
- evolution: {
- enabled: true,
- skillsDir,
- metacognition: { enabled: true, reflectThreshold: 0.7 },
- },
- });
-
- const runner = new AgentRunner({ config, providers: registry, tools: [], skillStore: store });
- const result = makeRunResult({ toolLoops: 2 });
- expect(runner.evaluateReflection(result)).toBeNull();
- });
+ // Per-turn `evaluateReflection` path removed in the reflection redesign
+ // (`docs/plans/reflection-redesign.md` §2.4). Reflection now runs only
+ // from the orchestrator (`features/reflection-orchestrator.ts`); the
+ // runner no longer owns reflection triggering.
it("agent can create a skill via tool call", async () => {
const mockProvider = createMockProvider([
diff --git a/src/core-agent/test/metacognition.test.ts b/src/core-agent/test/metacognition.test.ts
index 97d524c3..44f2485a 100644
--- a/src/core-agent/test/metacognition.test.ts
+++ b/src/core-agent/test/metacognition.test.ts
@@ -3,7 +3,7 @@ import {
detectUserCorrection,
emptyRunMetrics,
shouldReflect,
- buildAdaptiveReviewPrompt,
+ buildReviewPrompt,
} from '../src/evolution/metacognition.js';
import type { RunMetrics, MetacognitionConfig } from '../src/evolution/types.js';
@@ -287,91 +287,83 @@ describe('shouldReflect', () => {
});
});
-// ── buildAdaptiveReviewPrompt ───────────────────────────────────────────
+// ── buildReviewPrompt ───────────────────────────────────────────────────
-describe('buildAdaptiveReviewPrompt', () => {
- it('builds error_recovery focused prompt', () => {
- const prompt = buildAdaptiveReviewPrompt('error_recovery', '', '');
- expect(prompt).toContain('recovered from an error');
- expect(prompt).toContain('skill_manage');
- expect(prompt).toContain('metacognition');
+describe('buildReviewPrompt', () => {
+ it('renders the transcript section with content when provided', () => {
+ const transcript = '## Activity since 2026-05-19 14:00\n\n### c001\n[14:23 user]\nq';
+ const prompt = buildReviewPrompt('', '', transcript);
+ expect(prompt).toContain('Activity transcript');
+ expect(prompt).toContain('14:23 user');
});
- it('builds user_correction focused prompt', () => {
- const prompt = buildAdaptiveReviewPrompt('user_correction', '', '');
- expect(prompt).toContain('corrected your approach');
- expect(prompt).toContain('COMPETENCE');
- });
-
- it('builds skill_ineffective focused prompt', () => {
- const prompt = buildAdaptiveReviewPrompt('skill_ineffective', '', '');
- expect(prompt).toContain('did not help');
- expect(prompt).toContain('Patch or delete');
- });
-
- it('builds known_weakness focused prompt', () => {
- const prompt = buildAdaptiveReviewPrompt('known_weakness', '', '');
- expect(prompt).toContain('known weaknesses');
- expect(prompt).toContain('Strengthen');
- });
-
- it('builds complexity default prompt', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).toContain('complex task');
+ it('shows placeholder when transcript is empty', () => {
+ const prompt = buildReviewPrompt('', '', '');
+ expect(prompt).toContain('No new activity in the window');
});
it('includes competence content', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', 'I am strong at Python', '');
+ const prompt = buildReviewPrompt('I am strong at Python', '', '');
expect(prompt).toContain('I am strong at Python');
expect(prompt).not.toContain('No self-assessment yet');
});
it('shows placeholder when no competence', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
+ const prompt = buildReviewPrompt('', '', '');
expect(prompt).toContain('No self-assessment yet');
});
it('includes strategies content', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', 'Error extraction pattern');
+ const prompt = buildReviewPrompt('', 'Error extraction pattern', '');
expect(prompt).toContain('Error extraction pattern');
expect(prompt).not.toContain('No strategy log yet');
});
- it('includes skill health report when provided', () => {
- const report = '- docker-debug: effectiveness 0.83, healthy';
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '', report);
- expect(prompt).toContain('Skill health report');
- expect(prompt).toContain('docker-debug');
+ it('always includes transient-error guidance (preserved invariant)', () => {
+ const prompt = buildReviewPrompt('', '', '');
+ expect(prompt).toContain('transient errors');
+ expect(prompt).toContain('Do not mark them as weaknesses in COMPETENCE.md');
});
- it('omits skill health section when not provided', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).not.toContain('Skill health report');
+ it('always offers the four post-reflection actions', () => {
+ const prompt = buildReviewPrompt('', '', '');
+ expect(prompt).toContain('skill_manage tool');
+ expect(prompt).toContain('metacognition tool');
+ expect(prompt).toMatch(/nothing to save/i);
});
- it('includes conversation digest when provided', () => {
- const digest = 'Tools used: bash, read_file\nSkills loaded: docker-debug\n--- final reply ---\nFixed Docker container that failed to start';
- const prompt = buildAdaptiveReviewPrompt('error_recovery', '', '', undefined, digest);
- expect(prompt).toContain('docker-debug');
- expect(prompt).toContain('Docker container');
- expect(prompt).toContain('final reply');
+ it('directs LLM to look for user preferences + domain constraints', () => {
+ const prompt = buildReviewPrompt('', '', '');
+ // Plan §2.3: prompt should nudge LLM to extract red lines / edits / etc.
+ expect(prompt).toMatch(/user preferences|domain constraints|red lines/i);
});
- it('omits conversation digest content when not provided', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).not.toContain('final reply');
+ it('directs LLM to write imperatives, not descriptions', () => {
+ // Plan §9.x: prose injection of COMPETENCE/STRATEGIES is too soft unless
+ // entries are written as actionable rules with trigger conditions.
+ const prompt = buildReviewPrompt('', '', '');
+ expect(prompt).toContain('Writing style');
+ // The NEVER / ALWAYS / WHEN-THEN formula is the load-bearing instruction.
+ expect(prompt).toMatch(/NEVER\s*\/\s*ALWAYS\s*\/\s*WHEN-THEN/);
+ // Concrete bad/good pair is the second load-bearing piece.
+ expect(prompt).toMatch(/✗.*✓/s);
});
- it('builds weakness_succeeded focused prompt', () => {
- const prompt = buildAdaptiveReviewPrompt('weakness_succeeded', '', '');
- expect(prompt).toContain('performed normally');
- expect(prompt).toContain('remove or downgrade');
- expect(prompt).toContain('Do not modify or delete the related skills');
+ it('overrides skill_manage tool\'s "confirm with user" default for reflection', () => {
+ // skill_manage tool description says "Confirm with user before creating
+ // or deleting" — written for live turns. Without an explicit override
+ // here, reflection LLM gets inhibited and never creates skills. See
+ // reflection-redesign plan: skill-creation soft pitfall 1.
+ const prompt = buildReviewPrompt('', '', '');
+ expect(prompt).toMatch(/no user confirmation needed|does NOT need user confirmation/i);
});
- it('includes transient error guidance in all prompts', () => {
- const prompt = buildAdaptiveReviewPrompt('complexity', '', '');
- expect(prompt).toContain('transient errors');
- expect(prompt).toContain('Do not mark them as weaknesses in COMPETENCE.md');
+ it('extends Writing style guidance to the skill description field', () => {
+ // The description field is the ONLY thing the next-turn agent sees
+ // before deciding to load a skill — vague descriptions make the skill
+ // effectively dead. Soft pitfall 2 in reflection-redesign plan.
+ const prompt = buildReviewPrompt('', '', '');
+ expect(prompt).toMatch(/description.*field|`description`/i);
+ expect(prompt).toMatch(/WHEN to use/i);
});
});
diff --git a/src/main/features/agents.ts b/src/main/features/agents.ts
index a60ee08a..094fe524 100644
--- a/src/main/features/agents.ts
+++ b/src/main/features/agents.ts
@@ -1,9 +1,9 @@
/**
- * Agents — listing, custom CRUD, builtin sync, inline edit chat.
+ * Agents — listing, custom CRUD, marketplace installs, inline edit chat.
*
* Mirrors the skills module shape. Two sources:
- * builtin — data/shared/agents/builtin/<id>.json (synced from PC/builtin/agents/)
- * custom — data/shared/agents/custom/<id>.json (user-created, editable)
+ * marketplace — <uid>/local/marketplace/agents/<id>/agent.json
+ * custom — <uid>/cloud/agents/<id>/agent.json
*
* Schema (one JSON per agent):
* { agent_id, name, description, workflow, created_at, updated_at }
@@ -60,7 +60,8 @@ import {
normalizeMarketplaceCategoryCode,
} from './marketplace_biz';
-export type AgentSource = 'builtin' | 'custom';
+export type AgentSource = 'marketplace' | 'custom';
+type AgentSourceInput = AgentSource | 'builtin';
export type AgentInputType = 'text' | 'textarea' | 'select' | 'multiselect' | 'number' | 'boolean' | 'file' | 'directory';
@@ -179,7 +180,7 @@ export interface Agent {
/** Marketplace author uid. Kept optional for install/reconcile compatibility; global UI
* surfaces must not render it. */
create_uid?: string;
- /** Marketplace install version for `source==='builtin'`. Read from `_install.json` so the
+ /** Marketplace install version for `source==='marketplace'`. Read from `_install.json` so the
* agents-tab card can render a `v1.0.0` chip alongside the category. Custom agents leave
* this undefined (version is a publish concept, not authored locally). */
version?: string;
@@ -244,21 +245,25 @@ export interface AgentChatMeta { session_id?: string; [k: string]: unknown }
// 1. Builtin agent sync (startup)
// ─────────────────────────────────────────────────────────────────────────
-// `syncBuiltinAgents` / `hashTree` / `_isMarketplaceInstalled` removed — there is no longer a
-// shipped-builtin tree to sync from (`PC/src/builtin/` is gone) or a sentinel to protect (the
-// install flow now writes to `<uid>/local/marketplace/agents/` directly, distinct from the
-// scrubbed builtin path that no longer exists). Source='builtin' still names the UI group
-// ("Platform"); on disk it resolves to `userMarketplaceAgentsDir(uid)`.
+// `syncBuiltinAgents` / `hashTree` / `_isMarketplaceInstalled` removed — marketplace installs
+// now write to `<uid>/local/marketplace/agents/` directly.
// ─────────────────────────────────────────────────────────────────────────
// 2. Read / list
// ─────────────────────────────────────────────────────────────────────────
-/** Resolve where a given source's agents live on disk. Source='builtin' is a legacy name
- * preserved across the codebase; it now points at the per-user marketplace install dir. */
-function agentBaseDir(source: AgentSource): string {
+function normalizeAgentSource(source: AgentSourceInput): AgentSource {
+ return source === 'builtin' ? 'marketplace' : source;
+}
+
+function isMarketplaceSource(source: AgentSourceInput): boolean {
+ return normalizeAgentSource(source) === 'marketplace';
+}
+
+/** Resolve where a given source's agents live on disk. */
+function agentBaseDir(source: AgentSourceInput): string {
const uid = getActiveUserId();
- return source === 'builtin' ? userMarketplaceAgentsDir(uid) : userAgentsDir(uid);
+ return isMarketplaceSource(source) ? userMarketplaceAgentsDir(uid) : userAgentsDir(uid);
}
/** Helpers replacing the removed `builtinAgentDir` / `builtinAgentDefinitionFile`.
@@ -340,7 +345,7 @@ function assertAgentNameCharsetValid(name: string): void {
}
}
-/** Reject names already in use by another agent (custom OR builtin).
+/** Reject names already in use by another agent (custom OR marketplace).
* Matches `_agentNameKey` (case-insensitive, all whitespace stripped) so
* "Code Helper" and "codehelper" collide. `excludeAgentId` lets the
* update path keep its own name. Caller owns the check at every write
@@ -479,7 +484,7 @@ export function validateAgentInputs(raw: unknown): AgentInput[] {
return out;
}
-export function normalizeAgent(raw: AgentRaw | null | undefined, source: AgentSource): Agent | null {
+export function normalizeAgent(raw: AgentRaw | null | undefined, source: AgentSourceInput): Agent | null {
if (!raw || typeof raw !== 'object' || !raw.agent_id) return null;
// Migrate legacy single-`description` into the matching language slot.
// Same Chinese-character heuristic as the skill loader; explicit > legacy.
@@ -494,7 +499,7 @@ export function normalizeAgent(raw: AgentRaw | null | undefined, source: AgentSo
description_en: explicitEn || (legacyDesc && !legacyHasChinese ? legacyDesc : ''),
workflow: typeof raw.workflow === 'string' ? raw.workflow : '',
category: typeof raw.category === 'string' ? raw.category : '',
- source,
+ source: normalizeAgentSource(source),
created_at: raw.created_at || '',
updated_at: raw.updated_at || '',
// Defaults to enabled; overlaid by listAgents / getAgent from the
@@ -784,7 +789,7 @@ export async function listAgents(): Promise<Agent[]> {
} else {
specs = [];
const seen = new Set<string>();
- const sources: Array<[AgentSource, string]> = [['custom', CUSTOM_AGENTS_DIR()], ['builtin', userMarketplaceAgentsDir(getActiveUserId())]];
+ const sources: Array<[AgentSource, string]> = [['custom', CUSTOM_AGENTS_DIR()], ['marketplace', userMarketplaceAgentsDir(getActiveUserId())]];
for (const [source, dir] of sources) {
if (!fs.existsSync(dir)) continue;
const entries = (await fsp.readdir(dir, { withFileTypes: true }))
@@ -798,8 +803,8 @@ export async function listAgents(): Promise<Agent[]> {
const norm = normalizeAgent(data, source);
if (!norm) continue;
if (seen.has(norm.agent_id)) {
- if (source === 'builtin') {
- log.warn(`id conflict: custom and builtin both define "${norm.agent_id}" — custom wins, rename one`);
+ if (isMarketplaceSource(source)) {
+ log.warn(`id conflict: custom and marketplace both define "${norm.agent_id}" — custom wins, rename one`);
}
continue;
}
@@ -807,7 +812,7 @@ export async function listAgents(): Promise<Agent[]> {
// Marketplace-installed agents carry an `_install.json` sidecar with `version`.
// Author uid may also be present there for install/reconcile compatibility, but the
// global UI intentionally does not surface it.
- if (source === 'builtin') {
+ if (isMarketplaceSource(source)) {
try {
const metaFile = path.join(dir, e.name, '_install.json');
if (fs.existsSync(metaFile)) {
@@ -837,8 +842,8 @@ export async function listAgents(): Promise<Agent[]> {
*/
export async function getAgent(agentId: string | null | undefined): Promise<Agent | null> {
if (!agentId) return null;
- for (const source of ['custom', 'builtin'] as AgentSource[]) {
- const f = source === 'builtin'
+ for (const source of ['custom', 'marketplace'] as AgentSource[]) {
+ const f = isMarketplaceSource(source)
? _platformAgentSpecFile(agentId)
: agentDefinitionFile(getActiveUserId(), agentId);
if (!fs.existsSync(f)) continue;
@@ -1357,6 +1362,7 @@ export async function _applyAgentUpdatesAndInvalidate(
catch (err) { log.warn(`rename roster sweep failed id=${agentId}: ${(err as Error).message}`); }
}
return normalizeAgent(data, 'custom');
+ return normalizeAgent(data, 'marketplace');
}
/**
@@ -1704,6 +1710,8 @@ export function buildAgentEditSystemPrompt(agent: {
interactive: agent.interactive === true ? 'true' : 'false',
});
return `${body}\n\n---\n\n${buildLanguageDirective()}`;
+ const tail = buildLanguageDirective();
+ return `${body}\n\n---\n\n${tail}`;
}
export interface AgentEditResult {
diff --git a/src/main/features/chat_attachments.ts b/src/main/features/chat_attachments.ts
index 0038ad49..8eeb767c 100644
--- a/src/main/features/chat_attachments.ts
+++ b/src/main/features/chat_attachments.ts
@@ -35,6 +35,7 @@
import * as fs from 'node:fs';
import * as path from 'node:path';
+import * as crypto from 'node:crypto';
import { chatAttachmentDir, userChatAttachmentsDir, userChatsDir } from '../paths';
import { createLogger } from '../logger';
@@ -146,6 +147,85 @@ function uniqueTarget(dir: string, name: string): string {
return path.join(dir, `${stem}-${stamp}${ext}`);
}
+const attachmentWriteLocks = new Map<string, Promise<void>>();
+
+async function withAttachmentWriteLock<T>(
+ userId: string,
+ cid: string,
+ fn: () => Promise<T> | T,
+): Promise<T> {
+ const key = JSON.stringify([userId, cid]);
+ const prev = attachmentWriteLocks.get(key) || Promise.resolve();
+ let release!: () => void;
+ const current = new Promise<void>((resolve) => { release = resolve; });
+ attachmentWriteLocks.set(key, current);
+ try {
+ await prev;
+ return await fn();
+ } finally {
+ release();
+ if (attachmentWriteLocks.get(key) === current) {
+ attachmentWriteLocks.delete(key);
+ }
+ }
+}
+
+function attachmentInfoForPath(absPath: string, name = path.basename(absPath)): AttachmentInfo | null {
+ const ext = path.extname(name).toLowerCase();
+ if (!ALLOWED_EXTENSIONS.has(ext)) return null;
+ let st: fs.Stats;
+ try { st = fs.statSync(absPath); }
+ catch { return null; }
+ if (!st.isFile()) return null;
+ return {
+ name,
+ bytes: st.size,
+ kind: kindOf(ext),
+ mtime: Math.floor(st.mtimeMs / 1000),
+ };
+}
+
+function hashBuffer(buf: Buffer): string {
+ return crypto.createHash('sha256').update(buf).digest('hex');
+}
+
+function hashFile(absPath: string): Promise<string> {
+ return new Promise((resolve, reject) => {
+ const h = crypto.createHash('sha256');
+ const s = fs.createReadStream(absPath);
+ s.on('data', (chunk) => h.update(chunk));
+ s.on('error', reject);
+ s.on('end', () => resolve(h.digest('hex')));
+ });
+}
+
+async function findDuplicateByHash(
+ dir: string,
+ bytes: number,
+ sha256: string,
+): Promise<AttachmentInfo | null> {
+ let items: fs.Dirent[];
+ try { items = fs.readdirSync(dir, { withFileTypes: true }); }
+ catch { return null; }
+ items.sort((a, b) => a.name.toLowerCase().localeCompare(b.name.toLowerCase()));
+ for (const e of items) {
+ if (!e.isFile() || e.name.startsWith('.')) continue;
+ const ext = path.extname(e.name).toLowerCase();
+ if (!ALLOWED_EXTENSIONS.has(ext)) continue;
+ const abs = path.join(dir, e.name);
+ let st: fs.Stats;
+ try { st = fs.statSync(abs); }
+ catch { continue; }
+ if (st.size !== bytes) continue;
+ try {
+ if (await hashFile(abs) === sha256) {
+ return attachmentInfoForPath(abs, e.name);
+ }
+ } catch { /* skip unreadable */ }
+ }
+ return null;
+}
+
// ── Public API ───────────────────────────────────────────────────────────
export async function uploadAttachment(
@@ -153,7 +233,7 @@ export async function uploadAttachment(
cid: string,
name: string,
raw: Buffer | Uint8Array | null | undefined,
-): Promise<Result<{ info: AttachmentInfo }>> {
+): Promise<Result<{ info: AttachmentInfo; reused?: boolean }>> {
let safeName: string;
let safeConvId: string;
try {
@@ -175,36 +255,115 @@ export async function uploadAttachment(
}
const dir = ensureDir(userId, safeConvId);
- const target = uniqueTarget(dir, safeName);
- const finalName = path.basename(target);
- try { fs.writeFileSync(target, buf); }
- catch (err) {
- // Roll back the mkdir from `ensureDir` if the directory ended up empty
- // (i.e., this was the first attachment for the cid and the write failed
- // before producing anything). Best-effort: a non-empty dir means a
- // previous successful upload owns it and must not be touched.
- try {
- if (fs.readdirSync(dir).length === 0) fs.rmdirSync(dir);
- } catch { /* best-effort */ }
- return { ok: false, error: (err as Error).message };
+ const incomingHash = hashBuffer(buf);
+ return withAttachmentWriteLock(userId, safeConvId, async () => {
+ const duplicate = await findDuplicateByHash(dir, buf.length, incomingHash);
+ if (duplicate) {
+ log.info(`upload dedupe user=${userId} cid=${safeConvId} reuse=${duplicate.name} bytes=${duplicate.bytes}`);
+ return { ok: true, info: duplicate, reused: true };
+ }
+
+ const target = uniqueTarget(dir, safeName);
+ const finalName = path.basename(target);
+ try { fs.writeFileSync(target, buf); }
+ catch (err) {
+ // Roll back the mkdir from `ensureDir` if the directory ended up empty
+ // (i.e., this was the first attachment for the cid and the write failed
+ // before producing anything). Best-effort: a non-empty dir means a
+ // previous successful upload owns it and must not be touched.
+ try {
+ if (fs.readdirSync(dir).length === 0) fs.rmdirSync(dir);
+ } catch { /* best-effort */ }
+ return { ok: false, error: (err as Error).message };
+ }
+
+ // No upload-time preprocessing: extract / grayscale happen lazily in
+ // features/file_indexer when the model's tools first touch this file.
+ // Video is display-only and never preprocessed.
+
+ const st = fs.statSync(target);
+ const kind = kindOf(ext);
+ log.info(`upload user=${userId} cid=${safeConvId} name=${finalName} kind=${kind} bytes=${st.size}`);
+ return {
+ ok: true,
+ info: {
+ name: finalName,
+ bytes: st.size,
+ kind,
+ mtime: Math.floor(st.mtimeMs / 1000),
+ },
+ };
+ });
+}
+
+export async function importAttachmentFromPath(
+ userId: string,
+ cid: string,
+ sourcePath: string,
+ name?: string,
+): Promise<Result<{ info: AttachmentInfo; reused?: boolean }>> {
+ let safeName: string;
+ let safeConvId: string;
+ try {
+ safeName = safeAttachmentName(name || path.basename(sourcePath || ''));
+ safeConvId = safeCid(cid);
+ } catch (err) { return { ok: false, error: (err as Error).message }; }
+
+ const absSource = path.resolve(String(sourcePath || ''));
+ let sourceStat: fs.Stats;
+ try { sourceStat = fs.statSync(absSource); }
+ catch { return { ok: false, error: 'file not found' }; }
+ if (!sourceStat.isFile()) return { ok: false, error: 'file not found' };
+
+ const ext = path.extname(safeName).toLowerCase();
+ const cap = maxBytesFor(ext);
+ if (sourceStat.size > cap) {
+ return { ok: false, error: t('errors.file_too_large_mb', { mb: Math.round(cap / 1024 / 1024) }) };
+ }
+ if (TEXT_EXTS.has(ext)) {
+ let buf: Buffer;
+ try { buf = fs.readFileSync(absSource); }
+ catch (err) { return { ok: false, error: (err as Error).message }; }
+ const s = buf.toString('utf8');
+ if (Buffer.from(s, 'utf8').length !== buf.length) {
+ return { ok: false, error: t('errors.not_utf8') };
+ }
}
- // No upload-time preprocessing: extract / grayscale happen lazily in
- // features/file_indexer when the model's tools first touch this file.
- // Video is display-only and never preprocessed.
+ const dir = ensureDir(userId, safeConvId);
+ let incomingHash: string;
+ try { incomingHash = await hashFile(absSource); }
+ catch (err) { return { ok: false, error: (err as Error).message }; }
+ return withAttachmentWriteLock(userId, safeConvId, async () => {
+ const duplicate = await findDuplicateByHash(dir, sourceStat.size, incomingHash);
+ if (duplicate) {
+ log.info(`import dedupe user=${userId} cid=${safeConvId} reuse=${duplicate.name} bytes=${duplicate.bytes}`);
+ return { ok: true, info: duplicate, reused: true };
+ }
- const st = fs.statSync(target);
- const kind = kindOf(ext);
- log.info(`upload user=${userId} cid=${safeConvId} name=${finalName} kind=${kind} bytes=${st.size}`);
- return {
- ok: true,
- info: {
- name: finalName,
- bytes: st.size,
- kind,
- mtime: Math.floor(st.mtimeMs / 1000),
- },
- };
+ const target = uniqueTarget(dir, safeName);
+ const finalName = path.basename(target);
+ try { fs.copyFileSync(absSource, target); }
+ catch (err) {
+ try {
+ if (fs.readdirSync(dir).length === 0) fs.rmdirSync(dir);
+ } catch { /* best-effort */ }
+ return { ok: false, error: (err as Error).message };
+ }
+
+ const st = fs.statSync(target);
+ const kind = kindOf(ext);
+ log.info(`import user=${userId} cid=${safeConvId} name=${finalName} kind=${kind} bytes=${st.size}`);
+ return {
+ ok: true,
+ info: {
+ name: finalName,
+ bytes: st.size,
+ kind,
+ mtime: Math.floor(st.mtimeMs / 1000),
+ },
+ };
+ });
}
export function listAttachments(userId: string, cid: string): AttachmentInfo[] {
diff --git a/src/main/features/chats.ts b/src/main/features/chats.ts
index 0ce6a522..26e0f1ea 100644
--- a/src/main/features/chats.ts
+++ b/src/main/features/chats.ts
@@ -28,6 +28,9 @@ import {
} from '../storage';
import { createLogger } from '../logger';
import { t } from '../i18n';
+import {
+ ZH_FILLER_RE, EN_FILLER_RE, CLAUSE_RE, TITLE_MAX,
+} from '../util/auto-title';
const log = createLogger('chats');
import * as search from './search';
@@ -115,7 +118,14 @@ export async function listConversations(userId: string): Promise<Conversation[]>
let since: string | null = null;
try {
const s = await readState(userId, c.conversation_id);
- const bus = await import('./group_chat/bus');
+ // CJS require (not dynamic import) so bus.ts resolves through the same
+ // module cache as the static-import chain. Node's dynamic `import()` is
+ // always ESM, which would load bus.ts as a SECOND module instance with
+ // its own _cids Map — splitting the bus state into two and silently
+ // losing every event that's emitted on the wrong half (see bus.ts comment
+ // at planExecutor.bindBusHooks for why ESM-vs-CJS duplication corrupts
+ // plan_executor's hooks).
+ const bus = require('./group_chat/bus') as typeof import('./group_chat/bus');
const busBusy = !bus.isQuiescent(userId, c.conversation_id);
processing = s.status === 'running' || busBusy;
since = processing ? s.last_active_at : null;
@@ -277,8 +287,11 @@ export async function deleteConversation(userId: string, cid: string): Promise<b
await saveConversations(userId, kept);
// Purge group dir (members.json / state.json / plan.md / visibility/) + bus state.
+ // CJS require (same reason as listConversations above) — dynamic `import()`
+ // would load bus.ts as a second ESM module and dropConv would clear the
+ // wrong _cids.
try {
- const bus = await import('./group_chat/bus');
+ const bus = require('./group_chat/bus') as typeof import('./group_chat/bus');
bus.dropConv(userId, cid);
await purgeGroupDir(userId, cid);
}
@@ -394,10 +407,7 @@ export async function deleteConversationsByAgent(agentId: string): Promise<numbe
// doesn't match a shorter prefix when a longer one is also valid. Single-
// character fillers (e.g. bare `请`) are intentionally NOT listed — too
// likely to clip real content like `请教...`.
-const ZH_FILLER_RE = /^(帮我看一下|可不可以|帮我看下|帮我看看|麻烦你|帮我看|我想要|想问问|能不能|可以不|可以吗|请帮我|看一下|麻烦|帮我|我想|想问|请问|看下|看看)\s*/;
-const EN_FILLER_RE = /^(could you|would you|can you|help me|i'?d like to|i want to|please)\s+/i;
-const CLAUSE_RE = /[,。?!;;,.?!]/;
-const TITLE_MAX = 30;
+// Heuristic auto-title regex/constants — see top-of-file import.
export function autoTitle(content: string): string {
const raw = (content || '').trim().replace(/\s+/g, ' ');
diff --git a/src/main/features/connectors/catalog-google.ts b/src/main/features/connectors/catalog-google.ts
index e919cfce..cc964ad7 100644
--- a/src/main/features/connectors/catalog-google.ts
+++ b/src/main/features/connectors/catalog-google.ts
@@ -1,9 +1,8 @@
/**
- * Google Workspace catalog entries — extracted from `catalog.ts` so they can be stripped
- * from the OrkasOpen build (Google requires Workspace Developer Preview Program enrollment
- * + production CASA verification, which the OSS distribution can't satisfy). PC builds
- * include this file via a try/require in `catalog.ts`; OrkasOpen ships without it (see
- * `OpenSource/SyncCode/strip-rules.json` `forbidden_files`).
+ * Google Workspace catalog entries — extracted from `catalog.ts` so the Google bundle and
+ * per-service connectors stay isolated while still syncing to OrkasOpen. Both PC and OrkasOpen
+ * include this file via a try/require in `catalog.ts`; the catch path only keeps older open-source
+ * checkouts without this file from crashing.
*
* Includes:
* - `google-workspace` (one independent all-in-one connector backed by
diff --git a/src/main/features/connectors/catalog.ts b/src/main/features/connectors/catalog.ts
index d59c335a..eddc0735 100644
--- a/src/main/features/connectors/catalog.ts
+++ b/src/main/features/connectors/catalog.ts
@@ -74,18 +74,15 @@ export const CONNECTOR_CATALOG: CatalogEntry[] = [
},
},
// Google Workspace entries (`google-workspace` bundle + 5 service members) live in
- // `catalog-google.ts` and are loaded via `_loadGoogleEntries` below — that file is
- // **stripped from the OrkasOpen build** because Google's production OAuth verification +
- // CASA assessment can't be satisfied for an OSS distribution (see
- // `OpenSource/SyncCode/strip-rules.json` `forbidden_files`). PC ships them; OrkasOpen
- // silently omits them (require throws → empty array).
+ // `catalog-google.ts` and are loaded via `_loadGoogleEntries` below. PC and OrkasOpen both
+ // ship them; the try/require keeps older OrkasOpen checkouts that lack the file from crashing.
..._loadGoogleEntries(),
];
function _loadGoogleEntries(): CatalogEntry[] {
try {
- // `require` (not `import`) so the path isn't statically resolved at bundle time. In the
- // OrkasOpen build the file is absent and the require throws — we swallow + return [].
+ // `require` (not `import`) so the path isn't statically resolved at bundle time. Older
+ // OrkasOpen builds may lack this file; swallow and return [] for that compatibility case.
// eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
const mod = require('./catalog-google') as { GOOGLE_ENTRIES?: CatalogEntry[] };
return Array.isArray(mod.GOOGLE_ENTRIES) ? mod.GOOGLE_ENTRIES : [];
diff --git a/src/main/features/connectors/tools-adapter.ts b/src/main/features/connectors/tools-adapter.ts
index d5738f97..b3e1bce8 100644
--- a/src/main/features/connectors/tools-adapter.ts
+++ b/src/main/features/connectors/tools-adapter.ts
@@ -112,34 +112,38 @@ const GOOGLE_SERVICE_BY_CONNECTOR_ID: Record<string, GoogleWorkspaceService> = {
gtasks: 'gtasks',
};
-const GOOGLE_WORKSPACE_TOOL_PREFIX_BY_SERVICE: Record<GoogleWorkspaceService, string> = {
- gmail: '[Gmail]',
- gcal: '[Calendar]',
- gdocs: '[Docs]',
- gsheets: '[Sheets]',
- gtasks: '[Tasks]',
-};
-
function _dedupeGoogleWorkspaceTools(
visible: Array<{ instance: ConnectorInstance; tools: ToolSchema[] }>,
): Array<{ instance: ConnectorInstance; tools: ToolSchema[] }> {
const workspace = visible.find((v) => v.instance.id === GOOGLE_WORKSPACE_ID);
if (!workspace) return visible;
- const shadowedPrefixes = new Set<string>();
- for (const { instance } of visible) {
- const service = GOOGLE_SERVICE_BY_CONNECTOR_ID[instance.id];
- if (!service) continue;
- shadowedPrefixes.add(GOOGLE_WORKSPACE_TOOL_PREFIX_BY_SERVICE[service]);
+ // Collect tool names from every visible single-Google-service connector.
+ // `bin/google-workspace-mcp-server.cjs` wraps the same five service
+ // adapters the standalone connectors use, preserving each adapter's
+ // original tool `name` and only adding a `[Gmail]` / `[Calendar]` / …
+ // prefix to the `description`. So `name` is the strict equality
+ // invariant — a service tool advertised under standalone `gmail`
+ // shares its name byte-for-byte with the workspace's wrapped copy.
+ // Matching by name is robust to a future UI polish that drops the
+ // description prefix (which would silently break the prior
+ // description-prefix dedup and re-expose both Gmail routes to the
+ // model — the 20–50 tool-selection-accuracy cliff anti-pattern that
+ // PC/CLAUDE.md §6.5 warns against).
+ const shadowedNames = new Set<string>();
+ for (const { instance, tools } of visible) {
+ if (instance.id === GOOGLE_WORKSPACE_ID) continue;
+ if (!GOOGLE_SERVICE_BY_CONNECTOR_ID[instance.id]) continue;
+ for (const t of tools) shadowedNames.add(t.name);
}
- if (!shadowedPrefixes.size) return visible;
+ if (!shadowedNames.size) return visible;
return visible
.map((v) => {
if (v.instance.id !== GOOGLE_WORKSPACE_ID) return v;
return {
instance: v.instance,
- tools: v.tools.filter((t) => !_hasShadowedWorkspacePrefix(t, shadowedPrefixes)),
+ tools: v.tools.filter((t) => !shadowedNames.has(t.name)),
};
})
// When all five single-service connectors are visible, the Workspace connector has no unique
@@ -147,10 +151,5 @@ function _dedupeGoogleWorkspaceTools(
.filter((v) => v.instance.id !== GOOGLE_WORKSPACE_ID || v.tools.length > 0);
}
-function _hasShadowedWorkspacePrefix(tool: ToolSchema, shadowedPrefixes: Set<string>): boolean {
- const desc = typeof tool.description === 'string' ? tool.description.trimStart() : '';
- for (const prefix of shadowedPrefixes) {
- if (desc.startsWith(prefix)) return true;
- }
- return false;
-}
+/** Test-only export — see `test/main/features/connectors/dedupe-google-workspace.test.ts`. */
+export const _dedupeGoogleWorkspaceToolsForTest = _dedupeGoogleWorkspaceTools;
diff --git a/src/main/features/conversation_files.ts b/src/main/features/conversation_files.ts
new file mode 100644
index 00000000..a8dd8cd2
--- /dev/null
+++ b/src/main/features/conversation_files.ts
@@ -0,0 +1,101 @@
+/**
+ * Conversation workspace file listing for the info side panel.
+ *
+ * The chat history only records files that passed through chip-producing
+ * tools (`write_file`, `edit_file`, PDF/image generators). Real runs often
+ * create batches through `bash` or CLI agents, so the panel needs a cheap
+ * disk snapshot of the conversation workspace as the source of truth.
+ */
+
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+
+export interface ConversationWorkspaceFile {
+ path: string;
+ relPath: string;
+ name: string;
+ bytes: number;
+ mtime: number;
+}
+
+export interface ConversationWorkspaceFileList {
+ root: string;
+ items: ConversationWorkspaceFile[];
+ count: number;
+ truncated: boolean;
+ rootExists: boolean;
+}
+
+const DEFAULT_MAX_FILES = 5000;
+const DEFAULT_MAX_DEPTH = 12;
+
+function toPosixRel(rel: string): string {
+ return rel.split(path.sep).filter(Boolean).join('/');
+}
+
+export function listWorkspaceFiles(
+ root: string,
+ opts: { maxFiles?: number; maxDepth?: number } = {},
+): ConversationWorkspaceFileList {
+ const rootAbs = path.resolve(root || '');
+ const maxFiles = Math.max(1, Math.floor(opts.maxFiles ?? DEFAULT_MAX_FILES));
+ const maxDepth = Math.max(0, Math.floor(opts.maxDepth ?? DEFAULT_MAX_DEPTH));
+ const items: ConversationWorkspaceFile[] = [];
+ let truncated = false;
+
+ let rootStat: fs.Stats;
+ try { rootStat = fs.statSync(rootAbs); }
+ catch {
+ return { root: rootAbs, items, count: 0, truncated: false, rootExists: false };
+ }
+ if (!rootStat.isDirectory()) {
+ return { root: rootAbs, items, count: 0, truncated: false, rootExists: false };
+ }
+
+ const walk = (dir: string, depth: number): void => {
+ if (items.length >= maxFiles) {
+ truncated = true;
+ return;
+ }
+ if (depth > maxDepth) {
+ truncated = true;
+ return;
+ }
+
+ let entries: fs.Dirent[];
+ try { entries = fs.readdirSync(dir, { withFileTypes: true }); }
+ catch { return; }
+ entries.sort((a, b) => a.name.localeCompare(b.name));
+
+ for (const e of entries) {
+ if (e.name.startsWith('.')) continue;
+ const abs = path.join(dir, e.name);
+ if (e.isSymbolicLink()) continue;
+ if (e.isDirectory()) {
+ walk(abs, depth + 1);
+ if (items.length >= maxFiles) {
+ truncated = true;
+ return;
+ }
+ } else if (e.isFile()) {
+ let st: fs.Stats;
+ try { st = fs.statSync(abs); }
+ catch { continue; }
+ items.push({
+ path: abs,
+ relPath: toPosixRel(path.relative(rootAbs, abs)),
+ name: e.name,
+ bytes: st.size,
+ mtime: Math.floor(st.mtimeMs),
+ });
+ if (items.length >= maxFiles) {
+ truncated = true;
+ return;
+ }
+ }
+ }
+ };
+
+ walk(rootAbs, 0);
+ return { root: rootAbs, items, count: items.length, truncated, rootExists: true };
+}
diff --git a/src/main/features/group_chat/bus.ts b/src/main/features/group_chat/bus.ts
index 79bcb962..bf19a164 100644
--- a/src/main/features/group_chat/bus.ts
+++ b/src/main/features/group_chat/bus.ts
@@ -47,9 +47,10 @@ import {
import * as planExecutor from './plan_executor';
import {
userChatsDir, userSkillsDir, userAgentsDir,
- userMarketplaceSkillsDir, userMarketplaceAgentsDir,
+ userMarketplaceSkillsDir, userMarketplaceAgentsDir, projectFilesDir,
} from '../../paths';
import * as agentsFeat from '../agents';
+import { isDevEnv } from '../devtools';
import { isAgentEnabled } from '../component_enabled';
import { buildLanguageDirective, descriptionLang, t } from '../../i18n';
import * as marketplaceFeat from '../marketplace';
@@ -923,6 +924,10 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
log.warn(`resolve project scope cid=${cid} pid=${turnProjectId}: ${(err as Error).message}`);
}
}
+ const turnProjectFilesRoot = turnProjectId ? projectFilesDir(uid, turnProjectId) : '';
+ const turnProjectFilesSystemBlock = turnProjectId
+ ? await buildProjectFilesSystemBlock(uid, cid, turnProjectId)
+ : '';
// First-turn replay: if the persistent session jsonl doesn't exist yet,
// prepend a `<group-chat-history>` block built from the visibility slice
@@ -941,13 +946,14 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
log.warn(`replay-prefix build failed cid=${cid} actor=${actor.id}: ${(err as Error).message}`);
}
- // Attach a `<attachments>` manifest block listing user-uploaded files
- // (text / pdf / docx / image with absolute paths + kinds). Commander uses
- // it to route work; agent uses it to extract values for `inputs_schema`
- // (especially `type=file` fields). Image bytes ride alongside via
- // ChatOptions.images so the vision model sees them on the same user turn
- // — the manifest entry carries `attached="inline"` so the LLM doesn't
- // waste a read_file round-trip re-fetching what it already has.
+ // Attach a `<attachments>` manifest block listing files uploaded on this
+ // user turn (text / pdf / docx / image with absolute paths + kinds).
+ // Project files are shared conversation context, so they live in the
+ // system prompt's runtime injection block instead of being prepended to
+ // the user message. Image bytes ride alongside via ChatOptions.images so
+ // the vision model sees them on the same user turn — the manifest entry
+ // carries `attached="inline"` so the LLM doesn't waste a read_file
+ // round-trip re-fetching what it already has.
let turnImages: Array<{ data: string; mediaType: 'image/png' | 'image/jpeg' | 'image/gif' | 'image/webp' }> = [];
if (item.attachments && item.attachments.length) {
try {
@@ -969,7 +975,7 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
// Hoisted here so the branch below can read it without re-fetching.
let cliAgent: import('../agents').Agent | null = null;
if (isCommander) {
- systemPrompt = await buildCommanderSystemPrompt(uid, cid, turnProjectScope?.agents ?? null);
+ systemPrompt = await buildCommanderSystemPrompt(uid, cid, turnProjectScope?.agents ?? null, turnProjectFilesSystemBlock);
extraTools = await buildCommanderExtraTools(state, w, item.attachments);
// skillList stays undefined for commander — every skill is globally
// visible (skills are NOT project-scoped this round; see CLAUDE.md §6).
@@ -1004,7 +1010,7 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
cliAgent = agent;
systemPrompt = ''; // unused on CLI path
} else {
- systemPrompt = await buildAgentInGroupSystemPrompt(uid, agent, workingDir);
+ systemPrompt = await buildAgentInGroupSystemPrompt(uid, agent, workingDir, turnProjectFilesSystemBlock);
// Pass agent.skill_list verbatim — it carries System A + System B
// ids the agent owns. Skills are NOT project-scoped this round
// (see CLAUDE.md §6); the runner's `projectAllowedSkillIds` hook is
@@ -1194,7 +1200,11 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
onSkillInvoked: (id, sys, trig) => skillBuffer.recordInvoked(id, sys, trig),
cacheRetention: 'short',
abortSignal: w.abortController.signal,
- readOnlyExtraRoots: [...skillRoots, ...agentRoots],
+ readOnlyExtraRoots: [
+ ...(turnProjectFilesRoot ? [turnProjectFilesRoot] : []),
+ ...skillRoots,
+ ...agentRoots,
+ ],
...(turnImages.length ? { images: turnImages } : {}),
...(extraTools.length ? { extraTools } : {}),
...(skillList !== undefined ? { skillList } : {}),
@@ -1306,12 +1316,19 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
const target = await agentsFeat.getAgent(editId);
if (!target) {
workingText = `${workingText}\n\n<span style="color:var(--danger)">⚠️ Agent edit failed: agent not found (id=${editId}).</span>`;
- } else if (target.source !== 'custom') {
- workingText = `${workingText}\n\n<span style="color:var(--danger)">⚠️ Built-in agents can't be edited from the main chat; fork one in the right-hand detail panel and edit there.</span>`;
+ } else if (target.source !== 'custom' && !false) {
+ workingText = `${workingText}\n\n<span style="color:var(--danger)">⚠️ Marketplace agents can't be edited from the main chat; fork one in the right-hand detail panel and edit there.</span>`;
} else if (agentsFeat.isCliAgent(target)) {
workingText = `${workingText}\n\n<span style="color:var(--danger)">⚠️ External agents can only be edited from the right-hand detail panel.</span>`;
} else {
- const updated = await agentsFeat.updateCustomAgent(editId, fields);
+ // `updateAgentSpec` dispatches: custom → updateCustomAgent,
+ // marketplace + dev → agents_dev.updateBuiltinAgentSpec
+ // (which writes the local marketplace install dir). The
+ // dev-env source guard above is what makes the marketplace
+ // branch reachable; the direct `updateCustomAgent` call this
+ // replaces would short-circuit to null because the custom
+ // agent.json doesn't exist for marketplace ids.
+ const updated = await agentsFeat.updateAgentSpec(editId, fields);
if (updated) {
createdAgents.push({ agent_id: updated.agent_id, name: updated.name, kind: 'updated' });
} else {
@@ -1694,7 +1711,10 @@ async function runTurn(state: CidState, w: WorkerState, item: QueueItem): Promis
// ── System prompts ───────────────────────────────────────────────────────
async function buildCommanderSystemPrompt(
- uid: string, cid: string, allowedAgentIds?: readonly string[] | null,
+ uid: string,
+ cid: string,
+ allowedAgentIds?: readonly string[] | null,
+ projectFilesBlock = '',
): Promise<string> {
const { prompts } = await import('../../prompts/loader');
const path_ = await import('node:path');
@@ -1726,11 +1746,24 @@ async function buildCommanderSystemPrompt(
os: process.platform === 'darwin' ? 'macOS' : process.platform === 'win32' ? 'Windows' : process.platform,
working_dir: workingDir,
local_exec_state: permState,
+ project_files_block: projectFilesBlock,
});
const shared = prompts.load('chat_shared_rules', {});
return appendLanguageDirective(concatSharedRules(main, shared));
}
+async function buildProjectFilesSystemBlock(uid: string, cid: string, projectId: string): Promise<string> {
+ try {
+ const projectFileFeature = await import('../project_files');
+ const manifest = await projectFileFeature.buildProjectFilesManifest(uid, projectId);
+ if (!manifest) return '';
+ return `### Project files\n\n${manifest}`;
+ } catch (err) {
+ log.warn(`project files system block failed cid=${cid} pid=${projectId}: ${(err as Error).message}`);
+ return '';
+ }
+}
+
/** Merge shared_rules into the per-role prompt. The shared block carries
* PDF / search / file-output rules that BOTH commander and agent need;
* duplicating them in two .md files would drift. We append them right
@@ -1756,9 +1789,9 @@ function appendLanguageDirective(prompt: string): string {
// Format:
// `\`read_file(<ROOT>/<id>/agent.json)\` — ROOT by Source:\n` +
// `- custom: <abs path>\n` +
-// `- builtin: <abs path>\n` +
+// `- marketplace: <abs path>\n` +
// `Use these ROOT values verbatim. \`id:\` is tool-call input only — prose mentions agents as @<name>.\n\n` +
-// per-entry lines `- @<name> (Source: custom|builtin, id: <agent_id>) — desc` + optional `\n inputs_schema: <slim json>`
+// per-entry lines `- @<name> (Source: custom|marketplace, id: <agent_id>) — desc` + optional `\n inputs_schema: <slim json>`
//
// Why expose id and ROOT inline (changed 2026-05): the prior layout hid
// agent_id (to discourage hex-id leak in user prose) and put paths in a
@@ -1789,11 +1822,11 @@ async function buildAgentsIndexBlock(uid: string, allowedIds?: readonly string[]
const { pickDescription } = await import('#core-agent');
const lang = descriptionLang(getCurrentLang());
const customRoot = path.resolve(userAgentsDir(uid));
- const builtinRoot = path.resolve(userMarketplaceAgentsDir(uid));
+ const marketplaceRoot = path.resolve(userMarketplaceAgentsDir(uid));
const header = [
'`read_file(<ROOT>/<id>/agent.json)` — ROOT by Source:',
`- custom: ${customRoot}`,
- `- builtin: ${builtinRoot}`,
+ `- marketplace: ${marketplaceRoot}`,
'Use these ROOT values verbatim. `id:` is tool-call input only — prose mentions agents as @<name>.',
'',
].join('\n');
@@ -1831,6 +1864,7 @@ async function buildAgentInGroupSystemPrompt(
_uid: string,
agent: { name?: string; description?: string; workflow?: string; agent_id: string; inputs?: unknown; output_format?: string },
workingDir: string,
+ projectFilesBlock = '',
): Promise<string> {
const { prompts } = await import('../../prompts/loader');
// Render the agent's declared inputs schema so the LLM knows when to
@@ -1856,6 +1890,7 @@ async function buildAgentInGroupSystemPrompt(
inputs_schema: inputsSchemaJson || '(none)',
working_dir: workingDir,
output_format_hint: buildOutputFormatHint(agent.output_format),
+ project_files_block: projectFilesBlock,
});
const shared = prompts.load('chat_shared_rules', {});
return appendLanguageDirective(concatSharedRules(main, shared));
@@ -2467,6 +2502,7 @@ export async function abort(uid: string, cid: string): Promise<void> {
const state = _cids.get(cidKey(uid, cid));
let cleared = 0;
let aborted = 0;
+ let abortedModelSessions = 0;
if (state) {
for (const [, w] of state.workers) {
cleared += w.queue.length;
@@ -2476,6 +2512,21 @@ export async function abort(uid: string, cid: string): Promise<void> {
try { w.abortController?.abort(); } catch { /* ignore */ }
}
}
+ // Belt-and-suspenders abort for model turns. In production traces we saw
+ // user stop requests reach this function while the bus worker map no longer
+ // exposed the live AbortController (`abortedWorkers=0`), even though the
+ // core-agent session kept running. The model client owns a per-session
+ // abort registry, so abort all active sessions for this conversation too:
+ // `gconv-<cid>` and every `gmember-<cid>-<agent>`.
+ try {
+ const model = await import('../../model/client');
+ const abortByCid = (model as {
+ abortActiveSessionsForConversation?: (cid: string) => number;
+ }).abortActiveSessionsForConversation;
+ if (typeof abortByCid === 'function') abortedModelSessions = abortByCid(cid);
+ } catch (err) {
+ log.warn(`abort model-session fallback failed cid=${cid}: ${(err as Error).message}`);
+ }
await setStatus(uid, cid, 'aborted');
if (state) {
emit(state, { type: 'aborted', cid });
@@ -2500,7 +2551,7 @@ export async function abort(uid: string, cid: string): Promise<void> {
await new Promise<void>((resolve) => setTimeout(resolve, 25));
}
}
- log.info(`abort user=${uid} cid=${cid} clearedQueue=${cleared} abortedWorkers=${aborted}`);
+ log.info(`abort user=${uid} cid=${cid} clearedQueue=${cleared} abortedWorkers=${aborted} abortedModelSessions=${abortedModelSessions}`);
}
// ── Cleanup ──────────────────────────────────────────────────────────────
@@ -2536,6 +2587,23 @@ export function _cidStateForTest(uid: string, cid: string): CidState | null {
// having to import bus internals (which would be a circular import). Bus is
// the only authority that mutates worker queues; executor speaks through
// these hooks.
+//
+// Catastrophic-if-broken invariant: bus.ts MUST be loaded exactly once per
+// process. If a second copy gets loaded (e.g. CJS + ESM dual load when
+// `bootstrap.cjs` requires src/main as CJS but somewhere does `await
+// import('./group_chat/bus')` — Node always resolves dynamic `import()` to
+// ESM regardless of caller context, producing a SEPARATE module instance
+// with its own `_cids`), the second load's bindBusHooks call wins because
+// plan_executor's `_hooks` is a module-level singleton. Then `groupChat.send`
+// (in module A) routes through `planExecutor.reconcile → _hooks.enqueue`
+// which now points at module B's enqueue. Module B has empty `_cids`, so
+// it creates a fresh state for the cid — agent workers run in B's state,
+// events emit to B's listeners (empty), and the IPC subscriber (registered
+// on module A) sees zero. Symptom: "loading 一直显示,中断后才看到 process
+// info"; chats sidebar sometimes-shows phantom `processing=true`. Bug fixed
+// at the load-path side: see chats.ts `require('./group_chat/bus')` (CJS
+// require, hits the same module cache as our static `from './bus'` import
+// chain) — never `await import` this file.
planExecutor.bindBusHooks({
async enqueue(params) {
@@ -2916,6 +2984,19 @@ async function _buildCliPrompt(
const attachmentsBlock = allAtts.length
? `## Attachments\n${allAtts.map(a => `- ${a}`).join('\n')}`
: '';
+ let projectFilesBlock = '';
+ try {
+ const { getConversation } = await import('../chats');
+ const conv = await getConversation(uid, cid);
+ const pid = (conv as any)?.project_id;
+ if (typeof pid === 'string' && pid) {
+ const projectFileFeature = await import('../project_files');
+ projectFilesBlock = await projectFileFeature.buildProjectFilesCliBlock(uid, pid);
+ }
+ } catch (err) {
+ log.warn(`cli project files block failed cid=${cid}: ${(err as Error).message}`);
+ }
+ const filesBlock = [attachmentsBlock, projectFilesBlock].filter(Boolean).join('\n\n');
// ── Task body — submission unwrap if the dispatch was a form-submit.
// When the user confirmed the input form, the dispatched payload is
@@ -2954,7 +3035,7 @@ async function _buildCliPrompt(
agent_name: agent.name || agent.agent_id,
agent_description: (agent.description_en || agent.description_zh || '').trim(),
output_protocol_block: outputProtocolBlock,
- attachments_block: attachmentsBlock,
+ attachments_block: filesBlock,
conversation_block: conversationBlock,
task_body: taskBody,
});
diff --git a/src/main/features/group_chat/conv_title.ts b/src/main/features/group_chat/conv_title.ts
index 08fedcfa..a0eaa6f4 100644
--- a/src/main/features/group_chat/conv_title.ts
+++ b/src/main/features/group_chat/conv_title.ts
@@ -1,20 +1,40 @@
// Placeholder conversation-title helpers shared by group-chat modules.
-// Keep this file dependency-free so `chats.ts` can import group_chat facade
-// without entering the conv_workspace -> chats cycle.
+// Keep this file's import surface narrow (only `i18n` — itself a leaf that
+// reads locale json) so `chats.ts` can import group_chat facade without
+// entering the conv_workspace -> chats cycle.
-// Match the literal default titles `chats.createConversation` writes when
-// no title is supplied (currently `t('chat.default_title')` resolved at
-// creation time per UI language) AND any historical / capitalization
-// variant. Match by string equality, not
-// lang lookup, because state can carry whatever the conv was named at
-// creation regardless of the current UI language.
-export const PLACEHOLDER_TITLES: ReadonlySet<string> = new Set([
- '新对话',
- 'New conversation',
+import { SUPPORTED_LANGS, t } from '../../i18n';
+
+// Capitalization variants from an older creation path. Live conversations
+// may still carry these literal titles from their creation time, so they
+// stay in the set as historical aliases.
+const HISTORICAL_PLACEHOLDERS: readonly string[] = [
'New Conversation',
- 'New chat',
'New Chat',
-]);
+];
+
+function buildPlaceholderTitles(): Set<string> {
+ const out = new Set<string>(HISTORICAL_PLACEHOLDERS);
+ for (const lang of SUPPORTED_LANGS) {
+ try {
+ const value = t('chat.default_title', undefined, lang);
+ // `t()` returns the raw key on miss — skip that to avoid polluting
+ // the placeholder set with the literal "chat.default_title" string.
+ if (value && value !== 'chat.default_title') out.add(value);
+ } catch { /* locale load failure — keep set partial rather than crash */ }
+ }
+ return out;
+}
+
+// Match the literal default title `chats.createConversation` writes when
+// no title is supplied (`t('chat.default_title')` resolved at creation
+// time per UI language) AND any historical / capitalization variant.
+// Built once at module load by enumerating every supported UI language —
+// adding a new locale auto-populates the set; nothing to remember to sync.
+// Match by string equality, not lang lookup, because state can carry
+// whatever the conv was named at creation regardless of the current UI
+// language.
+export const PLACEHOLDER_TITLES: ReadonlySet<string> = buildPlaceholderTitles();
/** True when `title` is a placeholder default written at conversation
* creation, i.e. the user hasn't named the chat yet. Locale-agnostic: it
diff --git a/src/main/features/group_chat/plan.ts b/src/main/features/group_chat/plan.ts
index 68608ff7..44be5c0d 100644
--- a/src/main/features/group_chat/plan.ts
+++ b/src/main/features/group_chat/plan.ts
@@ -76,6 +76,17 @@ export interface PlanStep {
* Reset to 0 on user-initiated retry / step success. */
transient_attempts?: number;
+ /** For `assignee === 'user'` steps only: the random `form_id` the
+ * executor stamped at dispatch time. Used by
+ * `plan_executor.ts::acceptsUserStepCompletion` to verify that the
+ * user's `<agent-input-submission>` text references the form we
+ * actually dispatched — O(1) lookup instead of reconstructing the
+ * mapping by scanning the conversation jsonl. Cleared on retryStep
+ * so a re-armed step gets a fresh form_id on the next dispatch.
+ * Absent on legacy plans authored before this field existed; in that
+ * case the gate falls back to "accept any user reply". */
+ pending_form_id?: string;
+
/** Legacy free-form notes (kept for tools that still use it). */
notes?: string;
}
@@ -150,6 +161,8 @@ function normalizeStep(raw: any, i: number): PlanStep {
...(typeof raw?.failure_reason === 'string' ? { failure_reason: raw.failure_reason } : {}),
...(Number.isFinite(Number(raw?.transient_attempts)) && Number(raw.transient_attempts) > 0
? { transient_attempts: Number(raw.transient_attempts) } : {}),
+ ...(typeof raw?.pending_form_id === 'string' && raw.pending_form_id
+ ? { pending_form_id: String(raw.pending_form_id) } : {}),
...(typeof raw?.notes === 'string' && raw.notes.trim() ? { notes: String(raw.notes) } : {}),
};
}
@@ -234,7 +247,7 @@ export async function setPlan(
export async function updateStep(
uid: string, cid: string, stepIndex: number, status: StepStatus,
- patch?: { notes?: string; output_summary?: string; output_files?: string[]; output_msg_id?: string; failure_reason?: string; transient_attempts?: number },
+ patch?: { notes?: string; output_summary?: string; output_files?: string[]; output_msg_id?: string; failure_reason?: string; transient_attempts?: number; pending_form_id?: string },
): Promise<PlanFile | null> {
const cur = await readPlan(uid, cid);
if (!cur) return null;
@@ -256,6 +269,10 @@ export async function updateStep(
if (patch.transient_attempts > 0) cur.steps[idx].transient_attempts = patch.transient_attempts;
else delete cur.steps[idx].transient_attempts;
}
+ if (patch?.pending_form_id !== undefined) {
+ if (patch.pending_form_id) cur.steps[idx].pending_form_id = patch.pending_form_id;
+ else delete cur.steps[idx].pending_form_id;
+ }
await writePlanRaw(uid, cid, cur);
log.info(`plan-update user=${uid} cid=${cid} step=${stepIndex} status=${status}`);
return cur;
diff --git a/src/main/features/group_chat/plan_executor.ts b/src/main/features/group_chat/plan_executor.ts
index be7c65f1..e4fe8443 100644
--- a/src/main/features/group_chat/plan_executor.ts
+++ b/src/main/features/group_chat/plan_executor.ts
@@ -23,12 +23,9 @@
import { Mutex } from 'async-mutex';
import * as crypto from 'node:crypto';
-import * as path from 'node:path';
import { createLogger } from '../../logger';
import { t } from '../../i18n';
-import { userChatsDir } from '../../paths';
-import { readJsonl } from '../../storage';
import { isTransientError } from '../../util/transient-errors';
import { COMMANDER_ID, USER_ID, readMembers } from './state';
import {
@@ -590,13 +587,23 @@ export async function retryStep(
if (!step) return { ok: false, error: 'step not found' };
if (step.status !== 'failed') {
// The UI can issue a duplicate retry from stale DOM while the first
- // retry has already re-armed the step. Treat those post-retry states as
- // success so users do not see a false failure after recovery started.
+ // retry has already re-armed the step (failed -> pending -> dispatch
+ // -> in_progress / blocked-on-deps). Treat those mid-recovery states
+ // as success so users do not see a false failure after recovery
+ // started.
+ //
+ // `done` is intentionally NOT in this set: by the time a step
+ // reaches `done`, the rail has already re-rendered with a hidden
+ // retry button (`_isStepActionable` in plan-rail.js returns
+ // `failed`-only), so there is no DOM path that produces a retry
+ // click on a done step. Accepting it here would also fire a retry
+ // expert-signal with the completed turn's output_msg_id, attributing
+ // user dissatisfaction to a successful execution — see PC/CLAUDE.md
+ // §10 expert-signals turn_id convention.
if (
step.status === 'pending'
|| step.status === 'in_progress'
|| step.status === 'blocked'
- || step.status === 'done'
) {
log.info(`plan-step retry-noop cid=${cid} step=${stepIndex} current=${step.status}`);
return { ok: true };
@@ -607,6 +614,7 @@ export async function retryStep(
failure_reason: '',
output_msg_id: '',
transient_attempts: 0,
+ pending_form_id: '',
});
// Unblock cascade-skipped downstream: when step N failed under
// `abort_plan` policy, applyTermination marked all not-yet-terminal
@@ -722,32 +730,23 @@ async function maybeMarkFinished(uid: string, cid: string, plan: PlanFile, ctx:
}
async function acceptsUserStepCompletion(
- uid: string,
+ _uid: string,
cid: string,
step: PlanStep,
ctx: ReconcileCtx,
): Promise<boolean> {
- const file = path.join(userChatsDir(uid), `${cid}.jsonl`);
- const messages = await readJsonl<GroupMessage>(file, 100_000);
- const userForms = messages.filter((m) => m.form?.agent_id === USER_ID);
- const stepForms = userForms.filter((m) => m.form?.plan_step_index === step.index);
- const candidates = stepForms.length ? stepForms : userForms;
- const relevant = candidates.length ? candidates[candidates.length - 1] : undefined;
- // Legacy conversations created before user-owned forms existed still need
- // plain-text replies to complete user steps.
- if (!relevant?.form) return true;
+ // Legacy conversations + plans authored before the stamp existed have no
+ // `pending_form_id`. Accept any user reply for those (the original
+ // behaviour before user-owned forms were introduced).
+ const expectedFormId = step.pending_form_id;
+ if (!expectedFormId) return true;
const text = ctx.finishedMessage?.text || '';
const submission = decodeSubmission(text);
- const expectedFormId = relevant.form.form_id;
if (!submission || submission.agent_id !== USER_ID || submission.form_id !== expectedFormId) {
log.info(`plan-user-step ignored non-matching user message cid=${cid} step=${step.index} expected_form=${expectedFormId}`);
return false;
}
- if (relevant.form.submitted !== true) {
- log.info(`plan-user-step ignored unmarked submission cid=${cid} step=${step.index} form=${expectedFormId}`);
- return false;
- }
return true;
}
@@ -901,6 +900,14 @@ async function dispatchStep(
// Step asks user for input. Commander voice; render goes to user with
// a user-owned form so the reply can route back to the plan machinery
// without waking commander as an extra side conversation.
+ //
+ // Stamp `pending_form_id` on the step BEFORE enqueueing so the user's
+ // reply can be matched in O(1) inside the reconcile mutex (see
+ // `acceptsUserStepCompletion`) — without this, that gate would have to
+ // scan the whole <cid>.jsonl to recover the mapping, with the linear
+ // scan happening inside `_planLock`.
+ const form = buildUserStepForm(step);
+ await updateStep(uid, cid, step.index, 'in_progress', { pending_form_id: form.form_id });
await _hooks.enqueue({
uid, cid,
fromActorId: COMMANDER_ID,
@@ -908,7 +915,7 @@ async function dispatchStep(
forceTo: [USER_ID],
triggered_step: step.index,
...(inheritedAttachments && inheritedAttachments.length ? { attachments: inheritedAttachments } : {}),
- form: buildUserStepForm(step),
+ form,
});
return;
}
diff --git a/src/main/features/metacognition.ts b/src/main/features/metacognition.ts
index 579f9d82..aab24c0c 100644
--- a/src/main/features/metacognition.ts
+++ b/src/main/features/metacognition.ts
@@ -38,7 +38,7 @@ export const STRATEGIES_CHAR_LIMIT = 4000; // ~1300 tokens — strategies is a
// 1. env `ORKAS_METACOGNITION='0'` — dev/CI kill switch, hard off.
// 2. user preference `preferences.json::metacognition_enabled` — UI setting;
// undefined (never written) → treated as on, preserving historical default.
-// runner.ts and reflection-trigger.ts both read from here; don't sprinkle the
+// runner.ts and reflection-orchestrator.ts both read from here; don't sprinkle the
// env check around again.
export function isFeatureEnabled(): boolean {
if (process.env.ORKAS_METACOGNITION === '0') return false;
diff --git a/src/main/features/project_files.ts b/src/main/features/project_files.ts
new file mode 100644
index 00000000..8c5b2002
--- /dev/null
+++ b/src/main/features/project_files.ts
@@ -0,0 +1,305 @@
+/**
+ * Project-scoped files.
+ *
+ * Storage: `<uid>/cloud/projects/<pid>/files/<name>`.
+ * These files belong to the project, not a single conversation, so every
+ * conversation inside the project receives a lightweight file-list prompt and
+ * file tools get read-only access to this directory.
+ */
+
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+
+import { projectFilesDir } from '../paths';
+import { createLogger } from '../logger';
+import { t } from '../i18n';
+import { invalidateFileCache } from './file_indexer';
+import { projectExists } from './projects';
+
+const log = createLogger('project_files');
+
+const TEXT_EXTS: ReadonlySet<string> = new Set([
+ '.md', '.markdown', '.txt', '.csv', '.tsv',
+ '.json', '.yaml', '.yml', '.log',
+]);
+const IMAGE_EXTS: ReadonlySet<string> = new Set(['.png', '.jpg', '.jpeg', '.webp', '.gif']);
+const VIDEO_EXTS: ReadonlySet<string> = new Set(['.mp4', '.webm', '.mov', '.m4v', '.ogv']);
+const PDF_EXT = '.pdf';
+const DOCX_EXT = '.docx';
+const ALLOWED_EXTENSIONS: ReadonlySet<string> = new Set([
+ ...TEXT_EXTS, PDF_EXT, DOCX_EXT, ...IMAGE_EXTS, ...VIDEO_EXTS,
+]);
+
+const MAX_BYTES_TEXT = 5 * 1024 * 1024;
+const MAX_BYTES_DOCX = 20 * 1024 * 1024;
+const MAX_BYTES_IMAGE = 20 * 1024 * 1024;
+const MAX_BYTES_PDF = 100 * 1024 * 1024;
+const MAX_BYTES_VIDEO = 200 * 1024 * 1024;
+const MAX_FILENAME_LEN = 200;
+
+export type ProjectFileKind = 'text' | 'pdf' | 'docx' | 'image' | 'video';
+
+export interface ProjectFileInfo {
+ name: string;
+ path: string;
+ bytes: number;
+ kind: ProjectFileKind;
+ mtime: number;
+}
+
+export type Result<T = {}> = ({ ok: true } & T) | { ok: false; error: string };
+
+function safeProjectId(projectId: unknown): string {
+ if (typeof projectId !== 'string' || !projectId) throw new Error('projectId required');
+ if (projectId.includes('/') || projectId.includes('\\') || projectId.includes('\x00') || projectId === '.' || projectId === '..') {
+ throw new Error('invalid projectId');
+ }
+ return projectId;
+}
+
+function safeFileName(name: unknown): string {
+ if (typeof name !== 'string') throw new Error('filename required');
+ const s = name.trim();
+ if (!s || s === '.' || s === '..') throw new Error('invalid filename');
+ if (s.includes('/') || s.includes('\\') || s.includes('\x00')) {
+ throw new Error('filename must not contain path separators');
+ }
+ if (s.startsWith('.')) throw new Error("filename must not start with '.'");
+ if (s.length > MAX_FILENAME_LEN) throw new Error('filename too long');
+ const ext = path.extname(s).toLowerCase();
+ if (!ALLOWED_EXTENSIONS.has(ext)) {
+ throw new Error(t('errors.unsupported_file_ext', { ext: ext || t('errors.unsupported_file_no_ext') }));
+ }
+ return s;
+}
+
+function kindOfName(name: string): ProjectFileKind {
+ const ext = path.extname(name).toLowerCase();
+ if (ext === PDF_EXT) return 'pdf';
+ if (ext === DOCX_EXT) return 'docx';
+ if (IMAGE_EXTS.has(ext)) return 'image';
+ if (VIDEO_EXTS.has(ext)) return 'video';
+ return 'text';
+}
+
+function maxBytesFor(name: string): number {
+ const ext = path.extname(name).toLowerCase();
+ if (ext === PDF_EXT) return MAX_BYTES_PDF;
+ if (ext === DOCX_EXT) return MAX_BYTES_DOCX;
+ if (IMAGE_EXTS.has(ext)) return MAX_BYTES_IMAGE;
+ if (VIDEO_EXTS.has(ext)) return MAX_BYTES_VIDEO;
+ return MAX_BYTES_TEXT;
+}
+
+function uniqueTarget(dir: string, name: string): string {
+ const original = path.join(dir, name);
+ if (!fs.existsSync(original)) return original;
+ const ext = path.extname(name);
+ const stem = name.slice(0, name.length - ext.length);
+ const now = new Date();
+ const pad = (v: number) => String(v).padStart(2, '0');
+ const stamp = `${now.getFullYear()}${pad(now.getMonth() + 1)}${pad(now.getDate())}-${pad(now.getHours())}${pad(now.getMinutes())}${pad(now.getSeconds())}`;
+ for (let i = 0; i < 1000; i += 1) {
+ const suffix = i === 0 ? stamp : `${stamp}-${i}`;
+ const target = path.join(dir, `${stem}-${suffix}${ext}`);
+ if (!fs.existsSync(target)) return target;
+ }
+ return path.join(dir, `${stem}-${stamp}-${Date.now()}${ext}`);
+}
+
+async function ensureProjectFilesDir(userId: string, projectId: string): Promise<string> {
+ const pid = safeProjectId(projectId);
+ if (!await projectExists(userId, pid)) throw new Error('not_found');
+ const dir = projectFilesDir(userId, pid);
+ fs.mkdirSync(dir, { recursive: true });
+ return dir;
+}
+
+function _notifyDirty(projectId: string): void {
+ try {
+ // eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
+ const sync = require('./sync') as { markDirty?: (domain: string, relPath: string) => void };
+ sync.markDirty?.('projects', `cloud/projects/${projectId}/files`);
+ } catch { /* features/sync stripped */ }
+}
+
+function infoFor(absPath: string): ProjectFileInfo | null {
+ let st: fs.Stats;
+ try { st = fs.statSync(absPath); }
+ catch { return null; }
+ if (!st.isFile()) return null;
+ const name = path.basename(absPath);
+ return {
+ name,
+ path: absPath,
+ bytes: st.size,
+ kind: kindOfName(name),
+ mtime: Math.floor(st.mtimeMs / 1000),
+ };
+}
+
+export async function listProjectFiles(userId: string, projectId: string): Promise<ProjectFileInfo[]> {
+ let dir: string;
+ try {
+ const pid = safeProjectId(projectId);
+ if (!await projectExists(userId, pid)) return [];
+ dir = projectFilesDir(userId, pid);
+ } catch { return []; }
+
+ let items: fs.Dirent[];
+ try { items = fs.readdirSync(dir, { withFileTypes: true }); }
+ catch { return []; }
+
+ const out: ProjectFileInfo[] = [];
+ items.sort((a, b) => a.name.toLowerCase().localeCompare(b.name.toLowerCase()));
+ for (const e of items) {
+ if (!e.isFile() || e.name.startsWith('.')) continue;
+ const ext = path.extname(e.name).toLowerCase();
+ if (!ALLOWED_EXTENSIONS.has(ext)) continue;
+ const info = infoFor(path.join(dir, e.name));
+ if (info) out.push(info);
+ }
+ return out;
+}
+
+export async function uploadProjectFile(
+ userId: string,
+ projectId: string,
+ name: string,
+ raw: Buffer | Uint8Array | null | undefined,
+): Promise<Result<{ info: ProjectFileInfo }>> {
+ let safeName: string;
+ let pid: string;
+ let dir: string;
+ try {
+ safeName = safeFileName(name);
+ pid = safeProjectId(projectId);
+ dir = await ensureProjectFilesDir(userId, pid);
+ } catch (err) { return { ok: false, error: (err as Error).message }; }
+
+ const buf = Buffer.isBuffer(raw) ? raw : Buffer.from(raw || []);
+ const cap = maxBytesFor(safeName);
+ if (buf.length > cap) {
+ return { ok: false, error: t('errors.file_too_large_mb', { mb: Math.round(cap / 1024 / 1024) }) };
+ }
+ if (TEXT_EXTS.has(path.extname(safeName).toLowerCase())) {
+ const s = buf.toString('utf8');
+ if (Buffer.from(s, 'utf8').length !== buf.length) {
+ return { ok: false, error: t('errors.not_utf8') };
+ }
+ }
+
+ const target = uniqueTarget(dir, safeName);
+ try { fs.writeFileSync(target, buf); }
+ catch (err) { return { ok: false, error: (err as Error).message }; }
+
+ const info = infoFor(target);
+ if (!info) return { ok: false, error: 'write failed' };
+ _notifyDirty(pid);
+ log.info(`upload user=${userId} pid=${pid} name=${info.name} kind=${info.kind} bytes=${info.bytes}`);
+ return { ok: true, info };
+}
+
+export async function deleteProjectFile(userId: string, projectId: string, name: string): Promise<Result> {
+ let safeName: string;
+ let pid: string;
+ try {
+ safeName = safeFileName(name);
+ pid = safeProjectId(projectId);
+ if (!await projectExists(userId, pid)) return { ok: false, error: 'not_found' };
+ } catch (err) { return { ok: false, error: (err as Error).message }; }
+
+ const dir = projectFilesDir(userId, pid);
+ const abs = path.resolve(dir, safeName);
+ if (path.relative(path.resolve(dir), abs).startsWith('..')) return { ok: false, error: 'forbidden' };
+ if (!fs.existsSync(abs)) return { ok: false, error: 'not_found' };
+ try { fs.unlinkSync(abs); }
+ catch (err) { return { ok: false, error: (err as Error).message }; }
+ try { invalidateFileCache(userId, abs); }
+ catch (err) { log.warn(`invalidate cache ${abs}: ${(err as Error).message}`); }
+ try {
+ if (fs.existsSync(dir) && fs.readdirSync(dir).length === 0) fs.rmdirSync(dir);
+ } catch { /* best-effort */ }
+ _notifyDirty(pid);
+ return { ok: true };
+}
+
+export async function resolveProjectFileAbsPath(
+ userId: string,
+ projectId: string,
+ name: string,
+): Promise<Result<{ absPath: string; kind: ProjectFileKind }>> {
+ let safeName: string;
+ let pid: string;
+ try {
+ safeName = safeFileName(name);
+ pid = safeProjectId(projectId);
+ if (!await projectExists(userId, pid)) return { ok: false, error: 'not_found' };
+ } catch (err) { return { ok: false, error: (err as Error).message }; }
+
+ const root = path.resolve(projectFilesDir(userId, pid));
+ const abs = path.resolve(root, safeName);
+ const rel = path.relative(root, abs);
+ if (rel.startsWith('..') || path.isAbsolute(rel)) return { ok: false, error: 'forbidden' };
+ let st: fs.Stats;
+ try { st = fs.statSync(abs); }
+ catch { return { ok: false, error: 'not_found' }; }
+ if (!st.isFile()) return { ok: false, error: 'not_found' };
+ return { ok: true, absPath: abs, kind: kindOfName(safeName) };
+}
+
+export async function getProjectFilesRoot(userId: string, projectId: string): Promise<string | null> {
+ try {
+ const pid = safeProjectId(projectId);
+ if (!await projectExists(userId, pid)) return null;
+ return projectFilesDir(userId, pid);
+ } catch { return null; }
+}
+
+export async function isProjectFilePath(userId: string, projectId: string, absPath: string): Promise<boolean> {
+ const root = await getProjectFilesRoot(userId, projectId);
+ if (!root) return false;
+ const rel = path.relative(path.resolve(root), path.resolve(absPath));
+ return rel === '' || (!rel.startsWith('..') && !path.isAbsolute(rel));
+}
+
+function escapeAttr(s: string): string {
+ return s.replace(/&/g, '&amp;').replace(/"/g, '&quot;').replace(/</g, '&lt;').replace(/>/g, '&gt;');
+}
+
+async function listProjectFilePathsForPrompt(userId: string, projectId: string): Promise<string[]> {
+ let dir: string;
+ try {
+ const pid = safeProjectId(projectId);
+ if (!await projectExists(userId, pid)) return [];
+ dir = projectFilesDir(userId, pid);
+ } catch { return []; }
+
+ let items: fs.Dirent[];
+ try { items = fs.readdirSync(dir, { withFileTypes: true }); }
+ catch { return []; }
+
+ return items
+ .filter(e => e.isFile() && !e.name.startsWith('.') && ALLOWED_EXTENSIONS.has(path.extname(e.name).toLowerCase()))
+ .sort((a, b) => a.name.toLowerCase().localeCompare(b.name.toLowerCase()))
+ .map(e => path.join(dir, e.name));
+}
+
+export async function buildProjectFilesManifest(userId: string, projectId: string): Promise<string> {
+ const files = await listProjectFilePathsForPrompt(userId, projectId);
+ if (!files.length) return '';
+ const entries = files.map((p) => `<file path="${escapeAttr(p)}"/>`);
+ return [
+ '<project-files>',
+ 'These files are available to every conversation in this project. Use read_file with the listed path when the task needs content; use stat_file first only when metadata or extraction is needed.',
+ ...entries,
+ '</project-files>',
+ ].join('\n');
+}
+
+export async function buildProjectFilesCliBlock(userId: string, projectId: string): Promise<string> {
+ const files = await listProjectFilePathsForPrompt(userId, projectId);
+ if (!files.length) return '';
+ const lines = files.map(p => `- ${p}`);
+ return `## Project files\n${lines.join('\n')}`;
+}
diff --git a/src/main/features/reflection-digest.ts b/src/main/features/reflection-digest.ts
deleted file mode 100644
index c244a9ae..00000000
--- a/src/main/features/reflection-digest.ts
+++ /dev/null
@@ -1,201 +0,0 @@
-/**
- * Build a compact activity digest for the metacognition reflection.
- *
- * Walks recent core-agent session jsonls bound to a target agent and
- * aggregates tool usage + errors into a markdown summary that gets
- * spliced into the reflection prompt as `conversationDigest`. Without
- * this, startup reflection is blind to what actually happened in the
- * cooldown window — see `reflection-trigger.ts` for the call site.
- *
- * Scope:
- * - `<uid>-gconv-*.jsonl` (commander session) AND
- * - `<uid>-gmember-<cid>-<aid>.jsonl` (agent worker session) for the
- * target agent — added per plan §1.3 P0 because agent worker tool
- * calls / errors are the largest reflection blind spot when only
- * the commander session is read.
- *
- * Output target: ~800 chars, ~250 tokens.
- */
-
-import * as fs from 'node:fs';
-import { userSessionFile } from '../paths';
-import { createLogger } from '../logger';
-import { listConversations, type Conversation } from './chats';
-import { buildGmemberSessionId } from './group_chat/state';
-
-const log = createLogger('reflection-digest');
-
-const ERROR_SAMPLE_MAX = 3;
-const ERROR_SAMPLE_LEN = 160;
-const TOOL_TOP_N = 8;
-const SKILLS_TOP_N = 8;
-
-// ── Aggregator (pure) ───────────────────────────────────────────────────
-
-export interface SessionMetrics {
- sessionsAnalyzed: number;
- toolCalls: Record<string, number>;
- errorCount: number;
- errorSamples: string[];
- skillsLoaded: Record<string, number>;
- earliestTs: number | null;
- latestTs: number | null;
-}
-
-export function emptyMetrics(): SessionMetrics {
- return {
- sessionsAnalyzed: 0,
- toolCalls: {},
- errorCount: 0,
- errorSamples: [],
- skillsLoaded: {},
- earliestTs: null,
- latestTs: null,
- };
-}
-
-/**
- * Aggregate one session's messages into the running totals. Messages are
- * the parsed jsonl lines — same shape as core-agent emits (role + content
- * blocks: text / tool_use / tool_result).
- */
-export function aggregateSession(messages: any[], into: SessionMetrics): void {
- if (!messages.length) return;
- into.sessionsAnalyzed += 1;
-
- for (const m of messages) {
- if (!m || typeof m !== 'object') continue;
- if (typeof m.ts === 'number') {
- into.earliestTs = into.earliestTs === null ? m.ts : Math.min(into.earliestTs, m.ts);
- into.latestTs = into.latestTs === null ? m.ts : Math.max(into.latestTs, m.ts);
- }
- const blocks = Array.isArray(m.content) ? m.content : [];
- for (const block of blocks) {
- if (!block || typeof block !== 'object') continue;
- if (block.type === 'tool_use' && typeof block.name === 'string') {
- into.toolCalls[block.name] = (into.toolCalls[block.name] || 0) + 1;
- if (block.name === 'skill_manage' && block.input?.action === 'read' && typeof block.input?.id === 'string') {
- into.skillsLoaded[block.input.id] = (into.skillsLoaded[block.input.id] || 0) + 1;
- }
- } else if (block.type === 'tool_result' && block.isError === true) {
- into.errorCount += 1;
- if (into.errorSamples.length < ERROR_SAMPLE_MAX) {
- const text = typeof block.content === 'string'
- ? block.content
- : JSON.stringify(block.content ?? '');
- into.errorSamples.push(text.slice(0, ERROR_SAMPLE_LEN).replace(/\s+/g, ' ').trim());
- }
- }
- }
- }
-}
-
-// ── Formatter (pure) ────────────────────────────────────────────────────
-
-export function formatDigest(metrics: SessionMetrics, sinceMs: number, nowMs: number): string {
- if (metrics.sessionsAnalyzed === 0) {
- return `No new conversation activity in the range ${isoDate(sinceMs)} to ${isoDate(nowMs)}.`;
- }
-
- const parts: string[] = [];
- const range = `${isoDate(metrics.earliestTs ?? sinceMs)} to ${isoDate(metrics.latestTs ?? nowMs)}`;
- parts.push(`Based on ${metrics.sessionsAnalyzed} session(s) (${range}):`);
-
- const sortedTools = Object.entries(metrics.toolCalls).sort((a, b) => b[1] - a[1]).slice(0, TOOL_TOP_N);
- if (sortedTools.length) {
- parts.push('');
- parts.push('**Tool calls**:');
- for (const [name, n] of sortedTools) parts.push(`- ${name}: ${n}`);
- }
-
- if (metrics.errorCount > 0) {
- parts.push('');
- parts.push(`**Total errors**: ${metrics.errorCount}`);
- if (metrics.errorSamples.length) {
- parts.push('**Error samples**:');
- for (const s of metrics.errorSamples) parts.push(`- ${s}`);
- }
- }
-
- const sortedSkills = Object.entries(metrics.skillsLoaded).sort((a, b) => b[1] - a[1]).slice(0, SKILLS_TOP_N);
- if (sortedSkills.length) {
- parts.push('');
- parts.push('**Skills loaded**:');
- for (const [id, n] of sortedSkills) parts.push(`- ${id}${n > 1 ? ` (${n}x)` : ''}`);
- }
-
- return parts.join('\n');
-}
-
-function isoDate(ms: number): string {
- return new Date(ms).toISOString().slice(0, 10);
-}
-
-// ── I/O wrapper ─────────────────────────────────────────────────────────
-
-/**
- * Build the digest for a target agent ('_default' or a specific agent_id).
- * Picks conversations whose `agent_id` matches and whose session jsonl
- * was modified at or after `sinceMs`.
- */
-export async function buildAgentReflectionDigest(
- uid: string,
- agentId: string,
- sinceMs: number,
- nowMs: number = Date.now(),
-): Promise<string> {
- const metrics = emptyMetrics();
- let convs: Conversation[] = [];
- try {
- convs = await listConversations(uid);
- } catch (err) {
- log.warn(`listConversations failed for uid ${uid}: ${(err as Error).message}`);
- return formatDigest(metrics, sinceMs, nowMs);
- }
-
- // _default → empty agent_id; specific agent → exact match
- const targetMatches = (convAgentId: string): boolean =>
- agentId === '_default' ? !convAgentId : convAgentId === agentId;
-
- for (const conv of convs) {
- if (!targetMatches(conv.agent_id || '')) continue;
- // Two sessions per matched conv: the commander gconv-* (already on
- // conv.session_id), and — when the target agent is a specific one
- // (not '_default') — the agent worker's gmember-* derived from
- // (cid, agentId). _default = commander-only convs.
- const sessionIds: string[] = [];
- if (conv.session_id) sessionIds.push(conv.session_id);
- if (agentId !== '_default') {
- sessionIds.push(buildGmemberSessionId(conv.conversation_id, agentId));
- }
- for (const sessionId of sessionIds) {
- _scanSessionFile(uid, sessionId, sinceMs, metrics);
- }
- }
-
- return formatDigest(metrics, sinceMs, nowMs);
-}
-
-/** Read one session jsonl (gated by mtime), parse lines, fold into metrics.
- * Logs warn + skips on read errors so a single corrupt file doesn't break
- * the digest. */
-function _scanSessionFile(uid: string, sessionId: string, sinceMs: number, metrics: SessionMetrics): void {
- let file: string;
- try { file = userSessionFile(uid, sessionId); }
- catch { return; }
- let stat: fs.Stats;
- try { stat = fs.statSync(file); }
- catch { return; }
- if (stat.mtimeMs < sinceMs) return;
-
- let raw: string;
- try { raw = fs.readFileSync(file, 'utf8'); }
- catch (err) { log.warn(`read ${file} failed: ${(err as Error).message}`); return; }
- const messages: any[] = [];
- for (const line of raw.split('\n')) {
- if (!line.trim()) continue;
- try { messages.push(JSON.parse(line)); }
- catch { /* skip malformed line */ }
- }
- aggregateSession(messages, metrics);
-}
diff --git a/src/main/features/reflection-orchestrator.ts b/src/main/features/reflection-orchestrator.ts
new file mode 100644
index 00000000..14afe92a
--- /dev/null
+++ b/src/main/features/reflection-orchestrator.ts
@@ -0,0 +1,321 @@
+/**
+ * Reflection orchestrator — the single trigger for metacognitive reflection.
+ *
+ * Replaces the old startup-only `reflection-trigger.ts` and the per-turn
+ * `runner.evaluateReflection` path. Per `docs/plans/reflection-redesign.md`:
+ *
+ * - One scheduler: fire on boot (after STARTUP_DELAY_MS) and every
+ * `CYCLE_INTERVAL_MS` thereafter via a setTimeout chain.
+ * - Per-agent gating: 4h min cooldown, dirty gate (signals.jsonl or
+ * session jsonl newer than lastReflectedAt), 7-day max-gap fallback.
+ * - Per-cycle cap: at most `MAX_AGENTS_PER_CYCLE`; eligible-but-deferred
+ * agents wait for the next cycle (no agents lost).
+ * - Sequential execution; failures isolate per-agent and don't advance
+ * `lastReflectedAt`, so the next cycle retries.
+ *
+ * Background work; never awaited by the boot path. Tests inject `reflect`,
+ * `now`, and timing knobs via `runOneCycle`.
+ */
+
+import * as fs from 'node:fs';
+import { userReflectionStateFile, userLocalConfigDir, userSessionFile } from '../paths';
+import { writeJsonSync } from '../storage';
+import { createLogger } from '../logger';
+import { listAgents } from './agents';
+import { listConversations, type Conversation } from './chats';
+import { buildGmemberSessionId } from './group_chat/state';
+import * as metacognition from './metacognition';
+import { buildTranscript } from './reflection-transcript';
+import { querySignals } from './expert_signals';
+import { buildRunner } from '../model/core-agent/runner';
+
+const log = createLogger('reflection-orchestrator');
+
+// ── Constants (per plan §2.1) ────────────────────────────────────────────
+
+/** Delay after boot before firing the first cycle. */
+export const STARTUP_DELAY_MS = 30_000;
+/** Interval between cycles after the first. */
+export const CYCLE_INTERVAL_MS = 12 * 3600 * 1000;
+/** Minimum gap between reflections for the same agent (anti-thrash). */
+export const MIN_COOLDOWN_MS = 4 * 3600 * 1000;
+/** Maximum gap — force a sanity reflection even when no signals/activity. */
+export const MAX_GAP_MS = 7 * 24 * 3600 * 1000;
+/** Default initial lookback window for never-reflected agents. */
+export const DEFAULT_LOOKBACK_MS = 48 * 3600 * 1000;
+/** Per-cycle agent cap — defer overflow to next cycle. */
+export const MAX_AGENTS_PER_CYCLE = 5;
+/** Sentinel agent id covering all `normal` (no-agent-bound) conversations. */
+export const DEFAULT_AGENT_ID = '_default';
+
+// ── State IO (kept compatible with old reflection-state.json) ────────────
+
+export interface ReflectionState {
+ /** ISO timestamp per agent id (or `_default`). */
+ lastReflectedAt: Record<string, string>;
+}
+
+export function readReflectionState(uid: string): ReflectionState {
+ const file = userReflectionStateFile(uid);
+ if (!fs.existsSync(file)) return { lastReflectedAt: {} };
+ try {
+ const raw = fs.readFileSync(file, 'utf8');
+ const data = JSON.parse(raw);
+ if (!data || typeof data !== 'object' || !data.lastReflectedAt || typeof data.lastReflectedAt !== 'object') {
+ log.warn(`reflection-state.json malformed for uid ${uid}, treating as empty`);
+ return { lastReflectedAt: {} };
+ }
+ const clean: Record<string, string> = {};
+ for (const [k, v] of Object.entries(data.lastReflectedAt as Record<string, unknown>)) {
+ if (typeof v === 'string') clean[k] = v;
+ }
+ return { lastReflectedAt: clean };
+ } catch (err) {
+ log.warn(`reflection-state.json parse failed for uid ${uid}: ${(err as Error).message}, treating as empty`);
+ return { lastReflectedAt: {} };
+ }
+}
+
+export function writeReflectionState(uid: string, state: ReflectionState): void {
+ writeJsonSync(userReflectionStateFile(uid), state);
+}
+
+// ── Eligibility & dirty gate (pure / IO) ────────────────────────────────
+
+interface AgentDecision {
+ agentId: string;
+ /** Lower bound for activity to include in the next reflection's transcript. */
+ sinceMs: number;
+ /** Why we picked this agent (logged when we actually run). */
+ reason: 'dirty' | 'max_gap' | 'never_reflected';
+}
+
+/** Decide which agents are due for reflection in this cycle.
+ * Pure-ish wrt to `state` (read once, ts compared in-memory); the dirty
+ * check is delegated to `isDirty` so tests can stub it. */
+export async function pickAgentsForCycle(
+ uid: string,
+ agentIds: string[],
+ state: ReflectionState,
+ now: number,
+ isDirty: (uid: string, agentId: string, sinceMs: number) => Promise<boolean>,
+): Promise<AgentDecision[]> {
+ const out: AgentDecision[] = [];
+ for (const id of agentIds) {
+ const lastIso = state.lastReflectedAt[id];
+ const lastMs = lastIso ? Date.parse(lastIso) : NaN;
+ const hasLast = !Number.isNaN(lastMs);
+
+ if (hasLast && now - lastMs < MIN_COOLDOWN_MS) continue; // cooldown
+
+ if (hasLast && now - lastMs >= MAX_GAP_MS) {
+ out.push({ agentId: id, sinceMs: lastMs, reason: 'max_gap' });
+ continue;
+ }
+
+ const sinceMs = hasLast ? lastMs : now - DEFAULT_LOOKBACK_MS;
+ if (await isDirty(uid, id, sinceMs)) {
+ out.push({ agentId: id, sinceMs, reason: hasLast ? 'dirty' : 'never_reflected' });
+ }
+ }
+
+ // Apply per-cycle cap: most-stale (earliest lastReflectedAt) first; ties broken
+ // by ordering of agentIds (we already put _default first). Never-reflected
+ // agents sort earliest (Date.parse(undefined) = NaN → treat as 0).
+ out.sort((a, b) => {
+ const aLast = Date.parse(state.lastReflectedAt[a.agentId] || '') || 0;
+ const bLast = Date.parse(state.lastReflectedAt[b.agentId] || '') || 0;
+ return aLast - bLast;
+ });
+
+ if (out.length > MAX_AGENTS_PER_CYCLE) {
+ const deferred = out.length - MAX_AGENTS_PER_CYCLE;
+ log.info(`cycle: ${out.length} eligible, capping at ${MAX_AGENTS_PER_CYCLE}, deferring ${deferred} to next cycle`);
+ return out.slice(0, MAX_AGENTS_PER_CYCLE);
+ }
+ return out;
+}
+
+/** Dirty check: an agent is dirty if signals.jsonl has any entry attributed
+ * to it since `sinceMs`, OR if any of its session jsonl files has a turn
+ * newer than `sinceMs`. */
+export async function isAgentDirty(uid: string, agentId: string, sinceMs: number): Promise<boolean> {
+ const isDefault = agentId === DEFAULT_AGENT_ID;
+
+ // (1) signals.jsonl probe
+ try {
+ const sigs = await querySignals({
+ since: new Date(sinceMs).toISOString(),
+ aid: isDefault ? null : agentId,
+ limit: 1,
+ });
+ if (sigs.length > 0) return true;
+ } catch (err) {
+ log.warn(`isAgentDirty: querySignals failed agent=${agentId}: ${(err as Error).message}`);
+ }
+
+ // (2) session jsonl mtime probe (gconv for _default; gconv + gmember for specific)
+ let convs: Conversation[] = [];
+ try { convs = await listConversations(uid); }
+ catch (err) {
+ log.warn(`isAgentDirty: listConversations failed: ${(err as Error).message}`);
+ return false;
+ }
+ for (const c of convs) {
+ const a = c.agent_id || '';
+ const matches = isDefault ? !a : a === agentId;
+ if (!matches) continue;
+ if (_sessionNewerThan(uid, c.session_id, sinceMs)) return true;
+ if (!isDefault) {
+ const gmemberSid = buildGmemberSessionId(c.conversation_id, agentId);
+ if (_sessionNewerThan(uid, gmemberSid, sinceMs)) return true;
+ }
+ }
+ return false;
+}
+
+function _sessionNewerThan(uid: string, sessionId: string, sinceMs: number): boolean {
+ let file: string;
+ try { file = userSessionFile(uid, sessionId); } catch { return false; }
+ try {
+ const stat = fs.statSync(file);
+ return stat.mtimeMs >= sinceMs;
+ } catch { return false; }
+}
+
+// ── Reflection invocation ────────────────────────────────────────────────
+
+export type ReflectFn = (uid: string, agentId: string, sinceMs: number) => Promise<void>;
+
+/** Build the reflection prompt for one agent and run it. Throws on failure
+ * so the caller can decide whether to stamp `lastReflectedAt` (success)
+ * or leave it (retry next cycle). */
+async function realReflectForAgent(uid: string, agentId: string, sinceMs: number): Promise<void> {
+ const runnerAgentId = agentId === DEFAULT_AGENT_ID ? '' : agentId;
+
+ // Ephemeral session — runReflection uses an in-memory session (not the
+ // jsonl) so this id is just a label for the LLM-archive devtools.
+ const tail = Date.now().toString(36) + Math.random().toString(36).slice(2, 8);
+ const sessionId = `reflect-${tail}`;
+
+ const { runner } = await buildRunner({ sessionId, userId: uid, agentId: runnerAgentId });
+
+ const transcriptResult = await buildTranscript(uid, agentId, sinceMs);
+ if (!transcriptResult.text) {
+ // Dirty gate said yes but transcript came back empty (race / sweep / etc.).
+ // Skip without stamping — next cycle's dirty check decides again.
+ log.info(`reflect ${agentId}: transcript empty (considered=${transcriptResult.stats.convsConsidered}), skipping`);
+ throw new Error('transcript empty after dirty gate');
+ }
+
+ const ca = await import('#core-agent');
+ const comp = metacognition.readContent(agentId, 'competence');
+ const strat = metacognition.readContent(agentId, 'strategies');
+ const prompt = ca.buildReviewPrompt(comp.content || '', strat.content || '', transcriptResult.text);
+
+ // `runReflection` swallows provider/LLM/loop errors and returns ''; an
+ // empty response is treated as a failed reflection (cooldown not stamped).
+ const responseText = await runner.runReflection(prompt);
+ if (!responseText || !responseText.trim()) {
+ throw new Error('reflection returned empty (provider/LLM error or max loops; see core-agent log)');
+ }
+
+ log.info(`reflect ${agentId}: ok (transcript ${transcriptResult.stats.convsIncluded}/${transcriptResult.stats.convsConsidered} convs, ~${transcriptResult.stats.estimatedTokens} tokens)`);
+}
+
+// ── Cycle ───────────────────────────────────────────────────────────────
+
+export interface RunCycleOpts {
+ /** Override reflection invocation (test seam). */
+ reflect?: ReflectFn;
+ /** Override `Date.now()` (test seam). */
+ now?: () => number;
+ /** Override dirty check (test seam). */
+ isDirty?: (uid: string, agentId: string, sinceMs: number) => Promise<boolean>;
+}
+
+/** Run one reflection cycle: enumerate agents, pick eligible (capped),
+ * reflect sequentially. Returns the count actually reflected — useful for
+ * tests and logging. */
+export async function runOneCycle(uid: string, opts: RunCycleOpts = {}): Promise<number> {
+ if (!uid) {
+ log.debug('no active uid, skipping cycle');
+ return 0;
+ }
+ if (!metacognition.isFeatureEnabled()) {
+ log.debug('metacognition disabled, skipping cycle');
+ return 0;
+ }
+ fs.mkdirSync(userLocalConfigDir(uid), { recursive: true });
+
+ const now = (opts.now ?? Date.now)();
+ const reflect = opts.reflect ?? realReflectForAgent;
+ const dirtyFn = opts.isDirty ?? isAgentDirty;
+
+ let agents: Awaited<ReturnType<typeof listAgents>> = [];
+ try { agents = await listAgents(); }
+ catch (err) { log.warn(`listAgents failed: ${(err as Error).message}`); }
+
+ // `_default` first so the most common bucket gets attention even if a
+ // long agent list later in the loop hits issues.
+ const agentIds = [DEFAULT_AGENT_ID, ...agents.map((a) => a.agent_id)];
+ const state = readReflectionState(uid);
+ const eligible = await pickAgentsForCycle(uid, agentIds, state, now, dirtyFn);
+
+ if (eligible.length === 0) {
+ log.debug(`cycle: nothing eligible (${agentIds.length} agents scanned)`);
+ return 0;
+ }
+ log.info(`cycle start: ${eligible.length}/${agentIds.length} agent(s) eligible`);
+
+ let completed = 0;
+ for (const { agentId, sinceMs, reason } of eligible) {
+ try {
+ await reflect(uid, agentId, sinceMs);
+ // Stamp success — read fresh state in case another writer touched it.
+ const next = readReflectionState(uid);
+ next.lastReflectedAt[agentId] = new Date(now).toISOString();
+ writeReflectionState(uid, next);
+ completed += 1;
+ log.info(`reflect ${agentId}: completed (${reason})`);
+ } catch (err) {
+ log.warn(`reflect ${agentId}: failed (${reason}): ${(err as Error).message}`);
+ }
+ }
+ log.info(`cycle end: ${completed}/${eligible.length} agent(s) completed`);
+ return completed;
+}
+
+// ── Loop control ─────────────────────────────────────────────────────────
+
+export interface LoopHandle {
+ /** Cancel the next scheduled cycle. Any in-flight cycle continues to
+ * completion (no preemption). */
+ stop(): void;
+}
+
+/** Start the reflection loop: run one cycle now, then every
+ * `CYCLE_INTERVAL_MS` until stopped. The boot path calls this once after
+ * STARTUP_DELAY_MS and stores the handle so `before-quit` can cancel. */
+export function startReflectionLoop(uid: string, opts: RunCycleOpts = {}): LoopHandle {
+ let timer: NodeJS.Timeout | null = null;
+ let stopped = false;
+
+ const cycle = async () => {
+ if (stopped) return;
+ try { await runOneCycle(uid, opts); }
+ catch (err) { log.error(`cycle threw: ${(err as Error).message}`); }
+ if (stopped) return;
+ timer = setTimeout(cycle, CYCLE_INTERVAL_MS);
+ };
+
+ // Fire-and-forget; the caller has already delayed boot by STARTUP_DELAY_MS.
+ cycle();
+
+ return {
+ stop() {
+ stopped = true;
+ if (timer) { clearTimeout(timer); timer = null; }
+ },
+ };
+}
diff --git a/src/main/features/reflection-transcript.ts b/src/main/features/reflection-transcript.ts
new file mode 100644
index 00000000..659f62fb
--- /dev/null
+++ b/src/main/features/reflection-transcript.ts
@@ -0,0 +1,344 @@
+/**
+ * Build a chronological activity transcript for reflection.
+ *
+ * Replaces `reflection-digest.ts` per `docs/plans/reflection-redesign.md`.
+ * Reads the target agent's recent conversations, time-interleaves user
+ * voice (from gconv) with agent replies (from gmember), injects four
+ * system-event signal types (retry / skip / form_left_blank / silence),
+ * and applies a double cap (≤5 conversations AND ≤16K tokens).
+ *
+ * Output is a plain string for the reflection LLM prompt — no aggregation,
+ * no structured fields, just light annotation. Reflection LLM judges raw
+ * patterns from the text. T0/T1 signals other than the 4 inlined kinds are
+ * NOT consumed here (those serve future critic / weekly-review consumers).
+ *
+ * Pure side-effect-free; safe to call from anywhere with active uid.
+ */
+
+import * as fs from 'node:fs';
+import { userSessionFile } from '../paths';
+import { createLogger } from '../logger';
+import { listConversations, type Conversation } from './chats';
+import { buildGmemberSessionId } from './group_chat/state';
+import { querySignals, type Signal, type SignalType } from './expert_signals';
+
+const log = createLogger('reflection-transcript');
+
+// ── Caps (per plan §2.2) ────────────────────────────────────────────────
+
+export const MAX_CONVS = 5;
+export const MAX_TOKENS = 16_000;
+export const MAX_AGENT_REPLY_CHARS = 800;
+const SYSTEM_EVENT_TYPES: SignalType[] = ['retry', 'skip', 'form_left_blank', 'silence'];
+
+// ── Token estimation (CJK-aware) ────────────────────────────────────────
+
+/** Estimate token count. CJK chars count as ~0.7 token each; other chars
+ * as ~0.25 token (≈4 chars/token English heuristic). Empirically within
+ * ~10% of model tokenizers for mixed Chinese / English text. */
+export function estimateTokens(text: string): number {
+ // CJK Unified Ideographs (U+4E00–U+9FFF) + Hiragana (U+3040–U+309F)
+ // + Katakana (U+30A0–U+30FF) + Hangul Syllables (U+AC00–U+D7AF).
+ const cjkChars = (text.match(/[一-鿿぀-ヿ가-힯]/g) || []).length;
+ const otherChars = text.length - cjkChars;
+ return Math.ceil(cjkChars * 0.7 + otherChars / 4);
+}
+
+// ── Internal types ──────────────────────────────────────────────────────
+
+interface TranscriptEntry {
+ ts: number;
+ kind: 'user' | 'agent' | 'system';
+ text: string;
+}
+
+interface ConvSection {
+ conv: Conversation;
+ entries: TranscriptEntry[];
+}
+
+// ── Message parsing ─────────────────────────────────────────────────────
+
+/** Parse `<msg from="X" to="Y">...</msg>` wrapper used by group_chat. The
+ * wrapper carries the actual sender identity — `from="user"` is the human,
+ * `from="commander"` is commander dispatch, `from="<aid>"` is another agent.
+ * If no wrapper is detected (e.g. legacy edit chats), treat as a user
+ * message with the raw text. */
+function parseMsgWrapper(raw: string): { from: string; inner: string } {
+ const m = raw.match(/^<msg\s+from="([^"]+)"\s+to="[^"]*"[^>]*>([\s\S]*?)<\/msg>\s*$/);
+ if (!m) return { from: 'user', inner: raw };
+ return { from: m[1], inner: m[2].trim() };
+}
+
+/** Extract user-voice entries from a session jsonl. Filters to role=user
+ * text blocks where the `<msg>` wrapper says `from="user"` — drops
+ * commander dispatch / agent system messages. Skips tool_result blocks. */
+function extractUserEntries(messages: any[]): TranscriptEntry[] {
+ const out: TranscriptEntry[] = [];
+ for (const m of messages) {
+ if (!m || m.role !== 'user') continue;
+ if (typeof m.ts !== 'number') continue;
+ const blocks = Array.isArray(m.content) ? m.content : [];
+ const texts: string[] = [];
+ for (const b of blocks) {
+ if (!b || b.type !== 'text' || typeof b.text !== 'string') continue;
+ const { from, inner } = parseMsgWrapper(b.text);
+ if (from !== 'user') continue;
+ const trimmed = inner.trim();
+ if (trimmed) texts.push(trimmed);
+ }
+ if (texts.length) out.push({ ts: m.ts, kind: 'user', text: texts.join('\n') });
+ }
+ return out;
+}
+
+/** Extract agent-reply entries from a session jsonl. Filters to role=assistant
+ * text blocks only — drops thinking / tool_use / tool_result (those are
+ * intermediate work, not the final user-facing output). Caps each entry
+ * text at MAX_AGENT_REPLY_CHARS. */
+function extractAgentEntries(messages: any[]): TranscriptEntry[] {
+ const out: TranscriptEntry[] = [];
+ for (const m of messages) {
+ if (!m || m.role !== 'assistant') continue;
+ if (typeof m.ts !== 'number') continue;
+ const blocks = Array.isArray(m.content) ? m.content : [];
+ const texts: string[] = [];
+ for (const b of blocks) {
+ if (!b || b.type !== 'text' || typeof b.text !== 'string') continue;
+ const trimmed = b.text.trim();
+ if (trimmed) texts.push(trimmed);
+ }
+ if (!texts.length) continue;
+ let text = texts.join('\n');
+ if (text.length > MAX_AGENT_REPLY_CHARS) {
+ text = text.slice(0, MAX_AGENT_REPLY_CHARS) + '…(truncated)';
+ }
+ out.push({ ts: m.ts, kind: 'agent', text });
+ }
+ return out;
+}
+
+/** Render one signal as a synthetic system-event transcript entry. Returns
+ * null when the signal type isn't one of the 4 inlined kinds (defensive —
+ * caller already filters by type). */
+function renderSignalEntry(sig: Signal): TranscriptEntry | null {
+ const ts = Date.parse(sig.ts);
+ if (Number.isNaN(ts)) return null;
+ const meta = (sig.metadata || {}) as Record<string, unknown>;
+ switch (sig.type) {
+ case 'retry':
+ return { ts, kind: 'system', text: `user clicked retry on step #${meta.step_index ?? '?'}` };
+ case 'skip':
+ return { ts, kind: 'system', text: `user clicked skip on step #${meta.step_index ?? '?'}` };
+ case 'form_left_blank': {
+ const reqLabel = meta.was_required ? 'required field' : 'field';
+ return { ts, kind: 'system', text: `user left ${reqLabel} "${meta.input_id ?? '?'}" blank on form submit` };
+ }
+ case 'silence':
+ return { ts, kind: 'system', text: `agent message received no user response (silence threshold reached)` };
+ default:
+ return null;
+ }
+}
+
+// ── Session IO ──────────────────────────────────────────────────────────
+
+function readSessionJsonl(uid: string, sessionId: string): any[] {
+ let file: string;
+ try { file = userSessionFile(uid, sessionId); } catch { return []; }
+ let raw: string;
+ try { raw = fs.readFileSync(file, 'utf8'); }
+ catch { return []; }
+ const out: any[] = [];
+ for (const line of raw.split('\n')) {
+ if (!line.trim()) continue;
+ try { out.push(JSON.parse(line)); } catch { /* skip malformed line */ }
+ }
+ return out;
+}
+
+// ── Rendering ───────────────────────────────────────────────────────────
+
+function pad2(n: number): string { return String(n).padStart(2, '0'); }
+
+function formatTs(ts: number): string {
+ const d = new Date(ts);
+ return `${d.getFullYear()}-${pad2(d.getMonth() + 1)}-${pad2(d.getDate())} ${pad2(d.getHours())}:${pad2(d.getMinutes())}`;
+}
+
+function renderEntry(e: TranscriptEntry): string {
+ const ts = formatTs(e.ts);
+ if (e.kind === 'system') return `[${ts} system event] ${e.text}`;
+ return `[${ts} ${e.kind}]\n${e.text}`;
+}
+
+function formatConvSection(section: ConvSection): string {
+ const c = section.conv;
+ const dateBase = c.created_at ? Date.parse(c.created_at) : section.entries[0]?.ts || Date.now();
+ const datePart = new Date(dateBase).toISOString().slice(0, 10);
+ const lines: string[] = [`### ${c.conversation_id} — ${c.title || '(untitled)'} (${datePart})`, ''];
+ for (const e of section.entries) {
+ lines.push(renderEntry(e));
+ lines.push('');
+ }
+ return lines.join('\n').trimEnd();
+}
+
+// ── Top-level build ─────────────────────────────────────────────────────
+
+export interface TranscriptResult {
+ /** Markdown-ish transcript string. Empty when no activity matched. */
+ text: string;
+ /** Sanity-check counters for callers / observability. */
+ stats: {
+ convsConsidered: number;
+ convsIncluded: number;
+ convsTruncated: number;
+ estimatedTokens: number;
+ };
+}
+
+/**
+ * Build the activity transcript for one agent over a lookback window.
+ * Returns `text: ''` (with `stats.convsIncluded === 0`) when nothing matches.
+ *
+ * @param uid active user id
+ * @param agentId `_default` (no-agent / commander-only conversations) or a specific agent_id
+ * @param sinceMs lower bound for activity inclusion (typically lastReflectedAt epoch ms)
+ */
+export async function buildTranscript(
+ uid: string,
+ agentId: string,
+ sinceMs: number,
+): Promise<TranscriptResult> {
+ const isDefault = agentId === '_default';
+
+ let convs: Conversation[] = [];
+ try { convs = await listConversations(uid); }
+ catch (err) {
+ log.warn(`listConversations failed uid=${uid}: ${(err as Error).message}`);
+ return _empty();
+ }
+
+ const matched = convs.filter((c) => {
+ const a = c.agent_id || '';
+ return isDefault ? !a : a === agentId;
+ });
+ if (!matched.length) return _empty();
+
+ // Single signal query covers the whole window; we partition by cid below.
+ // For `_default` we pass `aid: null` (commander-scope signals like
+ // agent_dispatched are filtered out by the SYSTEM_EVENT_TYPES whitelist
+ // anyway, so a permissive aid filter for _default is fine).
+ let windowSignals: Signal[] = [];
+ try {
+ windowSignals = await querySignals({
+ since: new Date(sinceMs).toISOString(),
+ types: SYSTEM_EVENT_TYPES,
+ ...(isDefault ? {} : { aid: agentId }),
+ });
+ } catch (err) {
+ log.warn(`querySignals failed: ${(err as Error).message}`);
+ }
+
+ const sections: ConvSection[] = [];
+ for (const conv of matched) {
+ const gconvMsgs = readSessionJsonl(uid, conv.session_id);
+ if (!gconvMsgs.length) continue;
+
+ // Activity gate: skip conv whose newest msg predates the window.
+ const maxTs = gconvMsgs.reduce((acc, m) => Math.max(acc, typeof m.ts === 'number' ? m.ts : 0), 0);
+ if (maxTs < sinceMs) continue;
+
+ const userEntries = extractUserEntries(gconvMsgs).filter((e) => e.ts >= sinceMs);
+
+ let agentEntries: TranscriptEntry[];
+ if (isDefault) {
+ // No agent worker — commander's reply IS the response shown to user.
+ agentEntries = extractAgentEntries(gconvMsgs).filter((e) => e.ts >= sinceMs);
+ } else {
+ const gmemberSid = buildGmemberSessionId(conv.conversation_id, agentId);
+ const gmemberMsgs = readSessionJsonl(uid, gmemberSid);
+ agentEntries = extractAgentEntries(gmemberMsgs).filter((e) => e.ts >= sinceMs);
+ }
+
+ const sigEntries = windowSignals
+ .filter((s) => s.cid === conv.conversation_id)
+ .map(renderSignalEntry)
+ .filter((e): e is TranscriptEntry => e !== null)
+ .filter((e) => e.ts >= sinceMs);
+
+ const entries = [...userEntries, ...agentEntries, ...sigEntries]
+ .sort((a, b) => a.ts - b.ts);
+
+ if (!entries.length) continue;
+ sections.push({ conv, entries });
+ }
+
+ if (!sections.length) return _empty();
+
+ // Sort by most-recent-activity desc and apply MAX_CONVS cap.
+ sections.sort((a, b) => _lastTs(b) - _lastTs(a));
+ const convsConsidered = sections.length;
+ let convsTruncated = Math.max(0, sections.length - MAX_CONVS);
+ let kept = sections.slice(0, MAX_CONVS);
+
+ // Render in chronological order (oldest first) and drop oldest sections
+ // until token budget is met. Always keep at least one section (the most
+ // recent) so a single oversized conv still produces output rather than ''.
+ kept.sort((a, b) => _lastTs(a) - _lastTs(b));
+
+ const sinceLabel = _fmtSinceLabel(sinceMs);
+ const header = `## Activity since ${sinceLabel}`;
+
+ while (kept.length > 1) {
+ const full = `${header}\n\n${kept.map(formatConvSection).join('\n\n')}`;
+ if (estimateTokens(full) <= MAX_TOKENS) {
+ return _result(full, convsConsidered, kept.length, convsTruncated);
+ }
+ kept.shift();
+ convsTruncated += 1;
+ }
+
+ // Single section remaining — emit even if it exceeds budget (logged).
+ const full = `${header}\n\n${formatConvSection(kept[0])}`;
+ const tokens = estimateTokens(full);
+ if (tokens > MAX_TOKENS) {
+ log.warn(`transcript exceeds token cap: ${tokens} > ${MAX_TOKENS} (single conv ${kept[0].conv.conversation_id} kept)`);
+ }
+ return _result(full, convsConsidered, kept.length, convsTruncated);
+}
+
+// ── Helpers ─────────────────────────────────────────────────────────────
+
+function _lastTs(s: ConvSection): number {
+ return s.entries[s.entries.length - 1].ts;
+}
+
+function _fmtSinceLabel(ms: number): string {
+ return new Date(ms).toISOString().slice(0, 16).replace('T', ' ');
+}
+
+function _empty(): TranscriptResult {
+ return { text: '', stats: { convsConsidered: 0, convsIncluded: 0, convsTruncated: 0, estimatedTokens: 0 } };
+}
+
+function _result(text: string, considered: number, included: number, truncated: number): TranscriptResult {
+ const estimatedTokens = estimateTokens(text);
+ if (truncated > 0) {
+ log.info(`transcript: ${included}/${considered} convs included, ${truncated} dropped (caps), ~${estimatedTokens} tokens`);
+ }
+ return { text, stats: { convsConsidered: considered, convsIncluded: included, convsTruncated: truncated, estimatedTokens } };
+}
+
+// ── Test seam ───────────────────────────────────────────────────────────
+
+export const _internals = {
+ parseMsgWrapper,
+ extractUserEntries,
+ extractAgentEntries,
+ renderSignalEntry,
+ formatConvSection,
+ renderEntry,
+ formatTs,
+};
diff --git a/src/main/features/reflection-trigger.ts b/src/main/features/reflection-trigger.ts
deleted file mode 100644
index 2be6d833..00000000
--- a/src/main/features/reflection-trigger.ts
+++ /dev/null
@@ -1,239 +0,0 @@
-/**
- * Startup-triggered metacognitive reflection.
- *
- * Replaces the old per-turn trigger in `model/core-agent/client.ts`. Runs
- * once shortly after app boot, with a 48-hour per-agent cooldown so we
- * don't reflect on the same agent twice in a session.
- *
- * Flow:
- * 1. Enumerate `_default` (catches all `normal` conversations) + every
- * custom and builtin agent for the active user.
- * 2. Read `<uid>/local/config/reflection-state.json` for cooldown gates.
- * 3. For each eligible agent (cooldown passed), build a runner bound to
- * that agent and call `runner.runReflection(periodicReviewPrompt)`.
- * 4. On success, stamp `lastReflectedAt[agentId]` and persist.
- *
- * Iteration is sequential — no point hammering the API in parallel for
- * background work, and reflection per agent already takes single-digit
- * seconds at most.
- */
-
-import * as fs from 'node:fs';
-import { userReflectionStateFile, userLocalConfigDir } from '../paths';
-import { writeJsonSync } from '../storage';
-import { createLogger } from '../logger';
-import { listAgents } from './agents';
-import * as metacognition from './metacognition';
-import { buildAgentReflectionDigest } from './reflection-digest';
-import { buildRunner } from '../model/core-agent/runner';
-
-const log = createLogger('reflection-trigger');
-
-export const COOLDOWN_HOURS = 48;
-export const STARTUP_DELAY_MS = 30_000;
-/** Sentinel agent id covering all `normal` (no-agent) conversations. */
-export const DEFAULT_AGENT_ID = '_default';
-/** Default digest lookback window for first-ever reflection (no prior stamp). */
-export const DEFAULT_LOOKBACK_HOURS = 7 * 24;
-
-export interface ReflectionState {
- /** ISO timestamp per agent id (or `_default`). */
- lastReflectedAt: Record<string, string>;
-}
-
-const EMPTY_STATE: ReflectionState = { lastReflectedAt: {} };
-
-// ── State IO ─────────────────────────────────────────────────────────────
-
-export function readReflectionState(uid: string): ReflectionState {
- const file = userReflectionStateFile(uid);
- if (!fs.existsSync(file)) return { lastReflectedAt: {} };
- try {
- const raw = fs.readFileSync(file, 'utf8');
- const data = JSON.parse(raw);
- if (!data || typeof data !== 'object' || !data.lastReflectedAt || typeof data.lastReflectedAt !== 'object') {
- log.warn(`reflection-state.json malformed for uid ${uid}, treating as empty`);
- return { lastReflectedAt: {} };
- }
- // Filter out non-string values defensively.
- const clean: Record<string, string> = {};
- for (const [k, v] of Object.entries(data.lastReflectedAt as Record<string, unknown>)) {
- if (typeof v === 'string') clean[k] = v;
- }
- return { lastReflectedAt: clean };
- } catch (err) {
- log.warn(`reflection-state.json parse failed for uid ${uid}: ${(err as Error).message}, treating as empty`);
- return { lastReflectedAt: {} };
- }
-}
-
-export function writeReflectionState(uid: string, state: ReflectionState): void {
- const file = userReflectionStateFile(uid);
- writeJsonSync(file, state);
-}
-
-// ── Eligibility (pure) ───────────────────────────────────────────────────
-
-/**
- * Filter `agentIds` down to those past their cooldown.
- *
- * - Agent never reflected → eligible.
- * - Agent reflected within `cooldownHours` of `now` → skipped.
- * - Stale/invalid timestamp → eligible (defensive).
- */
-export function pickEligibleAgents(
- agentIds: string[],
- state: ReflectionState,
- now: number,
- cooldownHours: number = COOLDOWN_HOURS,
-): string[] {
- const cooldownMs = cooldownHours * 3600 * 1000;
- const eligible: string[] = [];
- for (const id of agentIds) {
- const lastIso = state.lastReflectedAt[id];
- if (!lastIso) {
- eligible.push(id);
- continue;
- }
- const lastMs = Date.parse(lastIso);
- if (Number.isNaN(lastMs)) {
- eligible.push(id);
- continue;
- }
- if (now - lastMs >= cooldownMs) {
- eligible.push(id);
- }
- }
- return eligible;
-}
-
-// ── Reflection invocation ────────────────────────────────────────────────
-
-export type ReflectFn = (uid: string, agentId: string, sinceMs: number) => Promise<void>;
-
-/**
- * Build a runner bound to `agentId` and run a periodic-review reflection.
- * `sinceMs` is the start of the activity window the digest aggregates over
- * (typically the last reflection's timestamp; 7 days back for first-ever).
- * Throws on auth/provider/LLM failure — orchestrator catches per-agent.
- */
-async function realReflectForAgent(uid: string, agentId: string, sinceMs: number): Promise<void> {
- // `_default` corresponds to "no agent bound" in the chat layer.
- const runnerAgentId = agentId === DEFAULT_AGENT_ID ? '' : agentId;
-
- // Disposable session id — the reflection runs in an in-memory Session
- // (created by AgentRunner.runReflection), so the persistent jsonl
- // created by buildRunner stays empty. Acceptable: same pattern as
- // organizer/extract-img kinds (§5).
- const tail = Date.now().toString(36) + Math.random().toString(36).slice(2, 8);
- const sessionId = `reflect-${tail}`;
-
- const { runner } = await buildRunner({
- sessionId,
- userId: uid,
- agentId: runnerAgentId,
- });
-
- // Aggregate recent activity into a markdown digest so the LLM reflects
- // on actual events instead of just rereading its own self-assessment.
- const digest = await buildAgentReflectionDigest(uid, agentId, sinceMs);
-
- // Build a periodic-review prompt directly — bypasses evaluateReflection's
- // signal-based gating (we're triggered by time, not by metrics).
- const ca = await import('#core-agent');
- const comp = metacognition.readContent(agentId, 'competence');
- const strat = metacognition.readContent(agentId, 'strategies');
- const prompt = ca.buildAdaptiveReviewPrompt(
- 'periodic_review',
- comp.content || '',
- strat.content || '',
- undefined,
- digest,
- );
-
- // `runReflection` swallows provider/LLM/loop-exhaustion errors and returns
- // `''` instead of throwing (see core-agent runner.ts: line 687/745/749).
- // Empty string = nothing actually succeeded, so it must not burn the
- // 48h cooldown. Even when the LLM concludes "no change needed", it
- // returns at least one sentence of explanation (non-empty), so a
- // post-trim blank is treated as failure. The specific failure reason
- // is in the core-agent logs (log.error / log.warn both fire); we
- // throw a generic error here.
- const responseText = await runner.runReflection(prompt);
- if (!responseText || !responseText.trim()) {
- throw new Error('reflection returned empty (provider/LLM error or max loops; see core-agent log)');
- }
-}
-
-// ── Orchestrator ─────────────────────────────────────────────────────────
-
-export interface RunStartupReflectionsOpts {
- /** Override the per-agent reflection call (test seam). */
- reflect?: ReflectFn;
- /** Override `Date.now()` for tests. */
- now?: () => number;
- /** Override the cooldown window for tests. */
- cooldownHours?: number;
-}
-
-/**
- * Top-level entry. Called once after boot completes (with delay). Safe to
- * call concurrently — each agent's reflection awaits in sequence so the
- * function as a whole serializes naturally.
- */
-export async function runStartupReflections(uid: string, opts: RunStartupReflectionsOpts = {}): Promise<void> {
- if (!uid) {
- log.debug('no active uid, skipping startup reflection');
- return;
- }
- if (!metacognition.isFeatureEnabled()) {
- log.debug('metacognition disabled, skipping startup reflection');
- return;
- }
- // Make sure local/config exists for first-boot users so writeReflectionState doesn't race.
- fs.mkdirSync(userLocalConfigDir(uid), { recursive: true });
-
- const now = (opts.now ?? Date.now)();
- const reflect = opts.reflect ?? realReflectForAgent;
- const cooldownHours = opts.cooldownHours ?? COOLDOWN_HOURS;
-
- let agents: Awaited<ReturnType<typeof listAgents>> = [];
- try {
- agents = await listAgents();
- } catch (err) {
- log.warn(`listAgents failed: ${(err as Error).message}`);
- }
- // `_default` first so the most common bucket gets attention even if a
- // long agent list later in the loop fails.
- const agentIds = [DEFAULT_AGENT_ID, ...agents.map((a) => a.agent_id)];
-
- const state = readReflectionState(uid);
- const eligible = pickEligibleAgents(agentIds, state, now, cooldownHours);
-
- if (eligible.length === 0) {
- log.info(`startup reflection: nothing eligible (${agentIds.length} agents, all within ${cooldownHours}h cooldown)`);
- return;
- }
- log.info(`startup reflection: ${eligible.length}/${agentIds.length} agent(s) eligible`);
-
- const defaultLookbackMs = DEFAULT_LOOKBACK_HOURS * 3600 * 1000;
- for (const agentId of eligible) {
- // Window for the activity digest: since last reflection, or default
- // lookback for first-ever (so we don't drag in ancient sessions).
- const lastIso = state.lastReflectedAt[agentId];
- const lastMs = lastIso ? Date.parse(lastIso) : NaN;
- const sinceMs = Number.isNaN(lastMs) ? now - defaultLookbackMs : lastMs;
-
- try {
- await reflect(uid, agentId, sinceMs);
- // Stamp success — read fresh state in case another writer touched it.
- const next = readReflectionState(uid);
- next.lastReflectedAt[agentId] = new Date(now).toISOString();
- writeReflectionState(uid, next);
- log.info(`reflection completed for agent ${agentId}`);
- } catch (err) {
- log.error(`reflection failed for agent ${agentId}: ${(err as Error).message}`);
- // Don't update lastReflectedAt — retry on next startup.
- }
- }
-}
diff --git a/src/main/features/skills.ts b/src/main/features/skills.ts
index bf9cf1d4..95af30c3 100644
--- a/src/main/features/skills.ts
+++ b/src/main/features/skills.ts
@@ -8,8 +8,8 @@
* 4. Skill inline edit chat (per user × skill, prefix injection, file-block extraction)
*
* Two skill sources:
- * builtin — data/shared/skills/builtin/<id>/ (seeded from PC/builtin/skills/)
- * custom — data/shared/skills/custom/<id>/ (user-created, editable)
+ * marketplace — <uid>/local/marketplace/skills/<id>/
+ * custom — <uid>/cloud/skills/<id>/
*
* core-agent integration: after any custom-skill mutation we call
* `invalidateSkills()` from `model/core-agent/skill-registry` so the next
@@ -83,7 +83,8 @@ function _parseSkillFileAttrs(raw: string): { path?: string } {
return out;
}
-export type SkillSource = 'builtin' | 'custom';
+export type SkillSource = 'marketplace' | 'custom';
+type SkillSourceInput = SkillSource | 'builtin';
export interface SkillListing {
id: string;
@@ -101,7 +102,7 @@ export interface SkillListing {
/** Marketplace author uid. Kept optional for install/reconcile compatibility; global UI
* surfaces must not render it. */
create_uid?: string;
- /** Marketplace install version for `source==='builtin'`. Read from `_install.json` so the
+ /** Marketplace install version for `source==='marketplace'`. Read from `_install.json` so the
* skills-tab card can render a `v1.0.0` chip. Undefined for custom skills. */
version?: string;
}
@@ -247,8 +248,16 @@ function _unescapeDoubleQuoted(s: string): string {
});
}
-function skillBaseDir(source: SkillSource): string {
- return source === 'custom' ? CUSTOM_SKILLS_DIR() : userMarketplaceSkillsDir(getActiveUserId());
+function normalizeSkillSource(source: SkillSourceInput): SkillSource {
+ return source === 'builtin' ? 'marketplace' : source;
+}
+
+function isMarketplaceSource(source: SkillSourceInput): boolean {
+ return normalizeSkillSource(source) === 'marketplace';
+}
+
+function skillBaseDir(source: SkillSourceInput): string {
+ return normalizeSkillSource(source) === 'custom' ? CUSTOM_SKILLS_DIR() : userMarketplaceSkillsDir(getActiveUserId());
}
// Module-level cache for `listSkills`.
@@ -298,7 +307,7 @@ export async function getSkillForEdit(skillId: string): Promise<SkillForEdit | n
if (!skillId) return null;
const sources: Array<[SkillSource, string]> = [
['custom', CUSTOM_SKILLS_DIR()],
- ['builtin', userMarketplaceSkillsDir(getActiveUserId())],
+ ['marketplace', userMarketplaceSkillsDir(getActiveUserId())],
];
for (const [source, base] of sources) {
const d = path.join(base, skillId);
@@ -336,7 +345,7 @@ export async function listSkills(): Promise<SkillListing[]> {
const out: SkillListing[] = [];
const seen = new Set<string>();
// Custom first so it wins on id collision (matches openclaw symlink rule).
- const sources: Array<[SkillSource, string]> = [['custom', CUSTOM_SKILLS_DIR()], ['builtin', userMarketplaceSkillsDir(getActiveUserId())]];
+ const sources: Array<[SkillSource, string]> = [['custom', CUSTOM_SKILLS_DIR()], ['marketplace', userMarketplaceSkillsDir(getActiveUserId())]];
for (const [source, baseDir] of sources) {
if (!fs.existsSync(baseDir)) continue;
const names = fs.readdirSync(baseDir, { withFileTypes: true })
@@ -345,8 +354,8 @@ export async function listSkills(): Promise<SkillListing[]> {
.sort();
for (const name of names) {
if (seen.has(name)) {
- if (source === 'builtin') {
- log.warn(`id conflict: custom and builtin both define "${name}" — custom wins, rename one`);
+ if (isMarketplaceSource(source)) {
+ log.warn(`id conflict: custom and marketplace both define "${name}" — custom wins, rename one`);
}
continue;
}
@@ -367,7 +376,7 @@ export async function listSkills(): Promise<SkillListing[]> {
// Marketplace-installed skills carry `_install.json` with `version`. Author uid may also
// be present for install/reconcile compatibility, but the global UI intentionally does
// not surface it.
- if (source === 'builtin') {
+ if (isMarketplaceSource(source)) {
try {
const metaFile = path.join(baseDir, name, '_install.json');
if (fs.existsSync(metaFile)) {
@@ -376,7 +385,7 @@ export async function listSkills(): Promise<SkillListing[]> {
}
} catch { /* corrupt _install.json — leave fields undefined */ }
}
- out.push({ id: name, name: displayName, source, ...descPair, category, enabled: true, create_uid: undefined, version });
+ out.push({ id: name, name: displayName, source: normalizeSkillSource(source), ...descPair, category, enabled: true, create_uid: undefined, version });
}
}
_skillListCache = { stamp, data: out };
@@ -399,7 +408,7 @@ export function setSkillEnabledForActiveUser(skillId: string, enabled: boolean):
export type Result<T = Record<string, unknown>> = ({ ok: true } & T) | { ok: false; error: string };
export async function readSkillFile(
- source: SkillSource, skillId: string, filepath = 'SKILL.md',
+ source: SkillSourceInput, skillId: string, filepath = 'SKILL.md',
): Promise<Result<{ content: string; ext: string; path: string }>> {
const base = skillBaseDir(source);
const skillDir = path.resolve(base, skillId);
@@ -420,7 +429,7 @@ export async function readSkillFile(
}
export async function listSkillTree(
- source: SkillSource, skillId: string,
+ source: SkillSourceInput, skillId: string,
): Promise<Result<{ tree: SkillTreeNode[] }>> {
const base = skillBaseDir(source);
const skillDir = path.resolve(base, skillId);
@@ -1441,10 +1450,13 @@ async function _applySkillContainerEdit(skillId: string, files: SkillFileBlock[]
}
return { ok: false, error: t('skills.errors.skill_not_found', { id: skillId }) };
}
- // Built-in skills are dev-mode-only via the detail-panel edit chat; the
- // commander flow always rejects regardless of dev mode (mirrors the agent
- // edit policy at `bus.ts` post-stream parsing).
- if (skill.source !== 'custom') {
+ // Marketplace skills are dev-mode-only from any write path (commander
+ // here + inline edit chat at sendToSkillChat). Mirrors the agent edit
+ // policy at `bus.ts` post-stream parsing. Downstream
+ // `writeSkillFileForEditChecked` routes the dev branch through
+ // `skills_dev.writeBuiltinSkillFile`, which writes the local marketplace
+ // install dir.
+ if (skill.source !== 'custom' && !false) {
return { ok: false, error: t('errors.builtin_skill_not_editable') };
}
const written: string[] = [];
@@ -1507,6 +1519,14 @@ export async function buildSkillEditSystemPrompt(skill: {
dir?: string;
}): Promise<string> {
const files = await listCustomSkillFiles(skill.id || '');
+ /** Picks the file-listing path: `marketplace` reads files from `dir`
+ * directly (per-machine install root), `custom` resolves via id under
+ * the user's custom skills dir. */
+ source?: SkillSource;
+}): Promise<string> {
+ const files = isMarketplaceSource(skill.source || 'custom') && skill.dir
+ ? await _listSkillFilesAt(skill.dir)
+ : await listCustomSkillFiles(skill.id || '');
// Resolve all three forms into a single legacy `$skill_description` for the
// current template, plus the bilingual pair for forward-compat. Phase 2
// splits the template; this keeps existing prompt rendering working.
@@ -1524,6 +1544,8 @@ export async function buildSkillEditSystemPrompt(skill: {
skill_files: skillFilesBlock(files),
});
return `${body}\n\n---\n\n${buildLanguageDirective()}`;
+ const tail = buildLanguageDirective();
+ return `${body}\n\n---\n\n${tail}`;
}
export interface SkillChatResult {
@@ -1539,6 +1561,9 @@ export interface SkillChatResult {
export async function sendToSkillChat(userId: string, skillId: string, content: string): Promise<SkillChatResult> {
const skill = await getCustomSkill(skillId);
if (!skill) return { ok: false, error: 'skill not found' };
+ if (isMarketplaceSource(skill.source) && !false) {
+ return { ok: false, error: t('errors.builtin_skill_not_editable') };
+ }
const meta = await loadSkillChatMeta(userId, skillId);
const sessionId = meta.session_id || defaultSkillSessionId(skillId);
@@ -1622,6 +1647,11 @@ export async function* streamSendToSkillChat(
yield { type: 'done' };
return;
}
+ if (isMarketplaceSource(skill.source) && !false) {
+ yield { type: 'error', text: t('errors.builtin_skill_not_editable') };
+ yield { type: 'done' };
+ return;
+ }
const meta = await loadSkillChatMeta(userId, skillId);
const sessionId = meta.session_id || defaultSkillSessionId(skillId);
diff --git a/src/main/index.ts b/src/main/index.ts
index 5fc07908..d6eff295 100644
--- a/src/main/index.ts
+++ b/src/main/index.ts
@@ -130,7 +130,7 @@ import * as searchFeature from './features/search';
import * as authFeature from './features/auth';
import * as appConfig from './features/config';
import { getRendererTables } from './i18n';
-import * as reflectionTrigger from './features/reflection-trigger';
+import * as reflectionOrchestrator from './features/reflection-orchestrator';
import * as scheduledTasks from './features/scheduled_tasks';
import * as chatAttachments from './features/chat_attachments';
import * as chatArtifacts from './features/chat_artifacts';
@@ -138,11 +138,11 @@ import * as chatArtifacts from './features/chat_artifacts';
import * as connectorsFeature from './features/connectors';
import * as windowState from './features/window_state';
// (sync + relay both depend on account; connectors depends on the Server OAuth bridge).
+// (both depend on the Orkas account/Server surface). Connectors, including Google, are synced.
function createWindow(): BrowserWindow {
const dev = !app.isPackaged;
- const dev = false;
const restored = windowState.restoreWindowState();
const win = new BrowserWindow({
width: 1280,
@@ -256,6 +256,7 @@ function registerIpc(): void {
builtin_skills_dir: 'X', custom_skills_dir: 'X',
agents_index: '', plan_state: '',
os: 'X', working_dir: 'X', local_exec_state: 'X',
+ project_files_block: '',
});
const tplLen = tplNormal.length;
const skills = await skillsFeature.listSkills();
@@ -276,7 +277,7 @@ function registerIpc(): void {
},
skills: {
total: skills.length,
- builtin: skills.filter((s) => s.source === 'builtin').length,
+ marketplace: skills.filter((s) => s.source === 'marketplace').length,
custom: skills.filter((s) => s.source === 'custom').length,
ids: skills.map((s) => `${s.source}:${s.id}`),
},
@@ -754,16 +755,15 @@ if (!gotLock) {
}
});
- // Background: per-agent metacognitive reflection. Time-triggered (48h
- // cooldown per agent), not per-turn. Delay so it doesn't race the UI's
- // first paint or the user's first action.
+ // Background: per-agent metacognitive reflection. Single 12h-chained
+ // loop replaces the old 48h-cooldown startup-only trigger
+ // (docs/plans/reflection-redesign.md). Delay the first cycle so it
+ // doesn't race the UI's first paint or the user's first action; the
+ // loop self-schedules thereafter.
setTimeout(() => {
const uid = users.getActiveUserId();
- const reflectLog = createLogger('reflection-trigger');
- reflectionTrigger.runStartupReflections(uid).catch((err) => {
- reflectLog.error('startup reflection batch failed', { error: err?.message || String(err) });
- });
- }, reflectionTrigger.STARTUP_DELAY_MS);
+ reflectionOrchestrator.startReflectionLoop(uid);
+ }, reflectionOrchestrator.STARTUP_DELAY_MS);
// Background: scheduled-agent-tasks tick (every 30s). Reads
// <uid>/cloud/config/scheduled_tasks.json on each tick and dispatches
diff --git a/src/main/ipc/index.ts b/src/main/ipc/index.ts
index e11ed467..5b36ec18 100644
--- a/src/main/ipc/index.ts
+++ b/src/main/ipc/index.ts
@@ -19,6 +19,7 @@ import { ipcMain, dialog, BrowserWindow, type WebContents } from 'electron';
import * as users from '../features/users';
import * as chats from '../features/chats';
import * as projects from '../features/projects';
+import * as projectFiles from '../features/project_files';
import * as groupChat from '../features/group_chat';
import type { GroupEvent } from '../features/group_chat/bus';
import * as agents from '../features/agents';
@@ -35,6 +36,7 @@ import * as kbVector from '../features/kb_vector';
import * as kbIndexer from '../features/kb_indexer';
import * as chatAttachments from '../features/chat_attachments';
import * as chatArtifacts from '../features/chat_artifacts';
+import * as conversationFiles from '../features/conversation_files';
import * as savedApps from '../features/saved_apps';
import * as search from '../features/search';
import * as auth from '../features/auth';
@@ -44,6 +46,8 @@ import * as permissions from '../features/permissions';
import * as appConfig from '../features/config';
import * as avatars from '../features/avatars';
import { getRendererTables, isLang, t } from '../i18n';
+import { isPathAllowed } from '../util/path-sandbox';
+import * as devtools from '../features/devtools';
import * as userWorkspace from '../features/user_workspace';
import { invokeHandlers as localAgentsHandlers } from './local_agents';
import { invokeHandlers as qualityHandlers } from './quality';
@@ -53,7 +57,9 @@ import { createLogger, logFromRenderer } from '../logger';
import * as path from 'node:path';
import * as fs from 'node:fs';
import { shell } from 'electron';
-import { WS_ROOT, chatAttachmentDir } from '../paths';
+import { WS_ROOT, chatAttachmentDir, projectFilesDir } from '../paths';
+import { readState as readGroupChatState } from '../features/group_chat/state';
+import { readPlan as readGroupChatPlan } from '../features/group_chat/plan';
const log = createLogger('ipc');
@@ -98,6 +104,72 @@ function _attachmentScopeForPayload(userId: string, payload: any): string | null
return path.resolve(chatAttachmentDir(userId, payload.cid));
}
+// Project-file scope for sandbox checks. Takes the already-resolved projectId
+// (computed by `_resolveWorkspaceScope`, which is cid-authoritative) rather
+// than reading payload.projectId directly — this enforces that a caller
+// passing `{cid, projectId}` where conv-cid.project_id !== projectId cannot
+// reach a foreign project's files (the cid wins, the claimed projectId
+// silently drops). When no cid is in payload, `_resolveWorkspaceScope`
+// already falls back to payload.projectId, so the commander-tab path
+// (project chip click before any conversation exists) continues to work.
+function _projectFileScopeForUser(userId: string, projectId: string | undefined): string | null {
+ if (!projectId || !safeId(projectId)) return null;
+ return path.resolve(projectFilesDir(userId, projectId));
+}
+
+/** Build the allowed-roots list for the file-class IPC sandbox: workspace ∪
+ * current cid's attachment dir ∪ payload's project-file dir. The actual
+ * containment check happens via `util/path-sandbox.isPathAllowed`, which
+ * realpath-resolves both candidate and roots so a symlink planted inside
+ * any allowed root cannot exfiltrate to a path outside.
+ *
+ * Centralised for `conversations.attachments.import` / `workspace.revealPath`
+ * / `produced.readText` / `produced.writeText` so the scope union stays in
+ * sync. Previously each handler did its own `path.resolve(target).startsWith(
+ * scope + path.sep)` triplet — lexical only, which let a symlink target
+ * outside the scope quietly slip through under a `<uid>` workspace path
+ * that itself contained one (the realistic attack: LLM-assisted skill
+ * drops a symlink into an attachment dir; user later writes to the
+ * apparent path through `produced.writeText` and the bytes land at
+ * attacker-chosen target). */
+async function _ipcFileSandboxAllowedRoots(userId: string, payload: any): Promise<string[]> {
+ const projectId = await _resolveWorkspaceScope(userId, payload);
+ const roots: string[] = [userWorkspace.getWorkspacePath(userId, projectId)];
+ const att = _attachmentScopeForPayload(userId, payload);
+ if (att) roots.push(att);
+ const pf = _projectFileScopeForUser(userId, projectId);
+ if (pf) roots.push(pf);
+ return roots;
+}
+
+/** Test-only export — see `test/main/ipc/{produced-readText,workspace-reveal}.test.ts`. */
+export const _ipcFileSandboxAllowedRootsForTest = _ipcFileSandboxAllowedRoots;
+
+async function _isConversationRecordedFile(userId: string, cid: string, absPath: string): Promise<boolean> {
+ if (!safeId(cid)) return false;
+ const target = path.resolve(absPath);
+ const matches = (value: unknown): boolean =>
+ typeof value === 'string' && !!value && path.resolve(value) === target;
+
+ try {
+ const messages = await chats.getMessages(userId, cid, 2000);
+ for (const msg of messages as any[]) {
+ const produced = Array.isArray(msg?.produced) ? msg.produced : [];
+ if (produced.some(matches)) return true;
+ }
+ } catch { /* best-effort allow-list */ }
+
+ try {
+ const plan = await readGroupChatPlan(userId, cid);
+ for (const step of plan?.steps || []) {
+ const files = Array.isArray(step.output_files) ? step.output_files : [];
+ if (files.some(matches)) return true;
+ }
+ } catch { /* best-effort allow-list */ }
+
+ return false;
+}
+
// ── Invoke handlers ──────────────────────────────────────────────────────
// Contract: `(payload, { userId, sender }) => result` where result is
// merged into a `{ ok: true, ...result }` response. Throw to signal error.
@@ -134,6 +206,17 @@ const invokeHandlers: Record<string, InvokeHandler> = {
};
},
+ 'conversations.files.list': async ({ cid }, ctx) => {
+ if (!safeId(cid)) throw new Error('invalid cid');
+ const projectId = await userWorkspace.resolveProjectIdForCid(ctx.userId, cid);
+ const workspaceRoot = userWorkspace.getWorkspacePath(ctx.userId, projectId);
+ const state = await readGroupChatState(ctx.userId, cid);
+ const root = state.workspace_dir
+ ? path.join(workspaceRoot, state.workspace_dir)
+ : workspaceRoot;
+ return conversationFiles.listWorkspaceFiles(root);
+ },
+
'conversations.create': async ({ title = '', projectId = '' } = {}, ctx) => {
// Validate the projectId belongs to this user before persisting it on
// the conv record. Unknown / invalid projectIds are dropped silently
@@ -203,6 +286,33 @@ const invokeHandlers: Record<string, InvokeHandler> = {
return { project };
},
+ 'projects.files.list': async ({ projectId }, ctx) => {
+ if (!safeId(projectId)) throw new Error('invalid projectId');
+ if (!await projects.projectExists(ctx.userId, projectId)) throw new Error('not_found');
+ return { files: await projectFiles.listProjectFiles(ctx.userId, projectId) };
+ },
+
+ 'projects.files.upload': async ({ projectId, name, data }, ctx) => {
+ if (!safeId(projectId)) throw new Error('invalid projectId');
+ if (typeof data !== 'string') throw new Error('missing data');
+ const buf = Buffer.from(data, 'base64');
+ return projectFiles.uploadProjectFile(ctx.userId, projectId, name || '', buf);
+ },
+
+ 'projects.files.delete': async ({ projectId, name }, ctx) => {
+ if (!safeId(projectId)) throw new Error('invalid projectId');
+ if (typeof name !== 'string' || !name) throw new Error('invalid name');
+ return projectFiles.deleteProjectFile(ctx.userId, projectId, name);
+ },
+
+ 'projects.files.absPath': async ({ projectId, name }, ctx) => {
+ if (!safeId(projectId)) throw new Error('invalid projectId');
+ if (typeof name !== 'string' || !name) throw new Error('invalid name');
+ const r = await projectFiles.resolveProjectFileAbsPath(ctx.userId, projectId, name);
+ if (!r.ok) return { ok: false, error: (r as { error?: string }).error || 'failed' };
+ return { ok: true, path: r.absPath, kind: r.kind };
+ },
+
// ── Project bindings (the strict scope of agents/skills visible inside
// a project conversation; see CLAUDE.md §6 outer-intersection rule) ──
// `bindings.list` returns the bound ids JOINED with name/description so
@@ -425,6 +535,22 @@ const invokeHandlers: Record<string, InvokeHandler> = {
return chatAttachments.uploadAttachment(ctx.userId, cid, name || '', buf);
},
+ 'conversations.attachments.import': async (payload, ctx) => {
+ const cid = payload?.cid;
+ const sourcePath = payload?.path;
+ if (!safeId(cid)) throw new Error('invalid cid');
+ if (typeof sourcePath !== 'string' || !sourcePath) throw new Error('missing path');
+
+ const norm = path.resolve(sourcePath);
+ const allowedRoots = await _ipcFileSandboxAllowedRoots(ctx.userId, payload);
+ const inSandbox = isPathAllowed(norm, allowedRoots);
+ const inRecordedFile = !inSandbox && await _isConversationRecordedFile(ctx.userId, cid, norm);
+ if (!inSandbox && !inRecordedFile) {
+ throw new Error('path is outside the user workspace');
+ }
+ return chatAttachments.importAttachmentFromPath(ctx.userId, cid, norm);
+ },
+
'conversations.attachments.delete': async ({ cid, name }, ctx) => {
if (!safeId(cid)) throw new Error('invalid cid');
return chatAttachments.deleteAttachment(ctx.userId, cid, name || '');
@@ -569,7 +695,7 @@ const invokeHandlers: Record<string, InvokeHandler> = {
'skills.list': async () => ({ skills: await skills.listSkills() }),
'skills.read': async ({ source, id, file = 'SKILL.md' }) => {
- if (source !== 'builtin' && source !== 'custom') throw new Error('invalid source');
+ if (source !== 'marketplace' && source !== 'builtin' && source !== 'custom') throw new Error('invalid source');
if (!skills.isValidSkillId(id)) throw new Error('invalid skill id');
return skills.readSkillFile(source, id, file);
},
@@ -583,7 +709,7 @@ const invokeHandlers: Record<string, InvokeHandler> = {
},
'skills.tree': async ({ source, id }) => {
- if (source !== 'builtin' && source !== 'custom') throw new Error('invalid source');
+ if (source !== 'marketplace' && source !== 'builtin' && source !== 'custom') throw new Error('invalid source');
if (!skills.isValidSkillId(id)) throw new Error('invalid skill id');
return skills.listSkillTree(source, id);
},
@@ -1043,14 +1169,8 @@ const invokeHandlers: Record<string, InvokeHandler> = {
if (typeof target !== 'string' || !target) {
throw new Error('missing path');
}
- const projectId = await _resolveWorkspaceScope(ctx.userId, payload);
- const ws = userWorkspace.getWorkspacePath(ctx.userId, projectId);
const norm = path.resolve(target);
- const wsNorm = path.resolve(ws);
- const attachmentScope = _attachmentScopeForPayload(ctx.userId, payload);
- const inAttachment = !!attachmentScope &&
- (norm === attachmentScope || norm.startsWith(attachmentScope + path.sep));
- if (!inAttachment && !norm.startsWith(wsNorm + path.sep) && norm !== wsNorm) {
+ if (!isPathAllowed(norm, await _ipcFileSandboxAllowedRoots(ctx.userId, payload))) {
throw new Error('path is outside the user workspace');
}
if (!fs.existsSync(norm)) throw new Error('file not found');
@@ -1088,15 +1208,8 @@ const invokeHandlers: Record<string, InvokeHandler> = {
if (typeof target !== 'string' || !target) {
throw new Error('missing path');
}
- const projectId = await _resolveWorkspaceScope(ctx.userId, payload);
- const ws = userWorkspace.getWorkspacePath(ctx.userId, projectId);
const norm = path.resolve(target);
- const wsNorm = path.resolve(ws);
- const attachmentScope = _attachmentScopeForPayload(ctx.userId, payload);
- const inAttachment = !!attachmentScope &&
- (norm === attachmentScope || norm.startsWith(attachmentScope + path.sep));
- const inWorkspace = norm === wsNorm || norm.startsWith(wsNorm + path.sep);
- if (!inWorkspace && !inAttachment) {
+ if (!isPathAllowed(norm, await _ipcFileSandboxAllowedRoots(ctx.userId, payload))) {
throw new Error('path is outside the user workspace');
}
let st: fs.Stats;
@@ -1134,22 +1247,30 @@ const invokeHandlers: Record<string, InvokeHandler> = {
if (bytes > MAX_TEXT_BYTES) {
return { ok: false, error: 'too_large', size: bytes, cap: MAX_TEXT_BYTES };
}
- const projectId = await _resolveWorkspaceScope(ctx.userId, payload);
- const ws = userWorkspace.getWorkspacePath(ctx.userId, projectId);
const norm = path.resolve(target);
- const wsNorm = path.resolve(ws);
- const attachmentScope = _attachmentScopeForPayload(ctx.userId, payload);
- const inAttachment = !!attachmentScope &&
- (norm === attachmentScope || norm.startsWith(attachmentScope + path.sep));
- const inWorkspace = norm === wsNorm || norm.startsWith(wsNorm + path.sep);
- if (!inWorkspace && !inAttachment) {
+ if (!isPathAllowed(norm, await _ipcFileSandboxAllowedRoots(ctx.userId, payload))) {
throw new Error('path is outside the user workspace');
}
+ // Used below to gate sync.markDirty: we want to bump the project-files
+ // sync domain ONLY when the write actually landed inside the
+ // project-files scope (so a workspace-only write doesn't tickle the
+ // project-sync engine). Computed via isPathAllowed on the narrow scope
+ // so the symlink-safe semantics match the outer gate.
+ const projectId = await _resolveWorkspaceScope(ctx.userId, payload);
+ const projectFileScope = _projectFileScopeForUser(ctx.userId, projectId);
+ const inProjectFile = !!projectFileScope && isPathAllowed(norm, [projectFileScope]);
try {
fs.writeFileSync(norm, content, 'utf8');
} catch (err) {
return { ok: false, error: String((err as Error).message || 'write failed') };
}
+ if (inProjectFile && payload?.projectId) {
+ try {
+ // eslint-disable-next-line @typescript-eslint/no-require-imports, global-require
+ const sync = require('../features/sync') as { markDirty?: (domain: string, relPath: string) => void };
+ sync.markDirty?.('projects', `cloud/projects/${payload.projectId}/files`);
+ } catch { /* sync stripped */ }
+ }
return { ok: true, size: bytes };
},
@@ -1226,15 +1347,13 @@ const streamHandlers: Record<string, StreamHandler> = {
}
const atts = Array.isArray(attachments) ? attachments.filter((n: any) => typeof n === 'string') : [];
// Legacy `conversations.stream` is now a thin wrapper around the
- // group_chat bus. **Subscribe to the bus directly BEFORE calling
- // `groupChat.send`** — `send` internally wakes the recipient worker
+ // group_chat bus. Subscribe to the bus directly BEFORE calling
+ // `groupChat.send` — `send` internally wakes the recipient worker
// synchronously, and that worker's first state_changed / process events
- // can fire on the same microtask cycle as `send` returns. If we waited
- // and then opened the subscription via `streamEvents`, those first
- // events would arrive at the bus listener list before our listener
- // attached and be dropped. The previous "send + then for-await
- // streamEvents" pattern was the source of "agent reply doesn't appear
- // until I refresh".
+ // can fire on the same microtask cycle as `send` returns. We also drain
+ // the subscription while `send` is still in flight: plan-triggered runs
+ // can spend real time dispatching/reconciling before `send` resolves,
+ // but the bus is already carrying the agent's process/delta stream.
//
// We relay events until the bus is fully quiescent (no worker running
// AND every actor's queue empty) — checked via the in-memory bus
@@ -1243,34 +1362,69 @@ const streamHandlers: Record<string, StreamHandler> = {
const buf: GroupEvent[] = [];
let wake: (() => void) | null = null;
let cancelled = signal.aborted;
+ const notify = () => {
+ const w = wake; wake = null; w?.();
+ };
const onAbort = () => {
cancelled = true;
- const w = wake; wake = null; w?.();
+ notify();
};
if (!cancelled) signal.addEventListener('abort', onAbort, { once: true });
const unsub = groupChat.subscribeBus(ctx.userId, cid, (ev) => {
buf.push(ev);
- const w = wake; wake = null; w?.();
+ notify();
});
- try {
- const sendRes = await groupChat.send({
- userId: ctx.userId, cid, text,
- ...(atts.length ? { attachments: atts } : {}),
- });
- if (!sendRes.ok) {
- yield { type: 'error', text: sendRes.error || 'send failed' };
- return;
+ let relayCount = 0;
+ let processCount = 0;
+ let firstProcessLogged = false;
+ let sendDone = false;
+ let sendRes: Awaited<ReturnType<typeof groupChat.send>> | null = null;
+ let sendErr: unknown = null;
+ const sendPromise = (async () => {
+ try {
+ sendRes = await groupChat.send({
+ userId: ctx.userId, cid, text,
+ ...(atts.length ? { attachments: atts } : {}),
+ });
+ } catch (err) {
+ sendErr = err;
+ } finally {
+ sendDone = true;
+ notify();
}
+ })();
+ void sendPromise;
+ try {
drainLoop: while (!cancelled) {
while (buf.length) {
const ev = buf.shift()!;
+ relayCount += 1;
+ if (ev.type === 'process') {
+ processCount += 1;
+ if (!firstProcessLogged) {
+ firstProcessLogged = true;
+ log.info(`sendStream first process cid=${cid} actor=${(ev as any).actor || ''} kind=${(ev as any).data?.type || ''}`);
+ }
+ }
yield { type: 'event', event: { stream: 'group', data: ev } };
}
- if (groupChat.busIsQuiescent(ctx.userId, cid)) break drainLoop;
+ if (sendDone) {
+ if (sendErr) {
+ const errText = sendErr instanceof Error ? sendErr.message : String(sendErr || 'send failed');
+ yield { type: 'error', text: errText };
+ return;
+ }
+ if (!sendRes?.ok) {
+ yield { type: 'error', text: sendRes?.error || 'send failed' };
+ return;
+ }
+ if (groupChat.busIsQuiescent(ctx.userId, cid)) break drainLoop;
+ }
if (cancelled) break;
await new Promise<void>((resolve) => { wake = resolve; });
}
} finally {
+ log.info(`sendStream closed cid=${cid} relayed=${relayCount} process=${processCount} sendDone=${sendDone} cancelled=${cancelled}`);
try { unsub(); } catch { /* ignore */ }
try { signal.removeEventListener?.('abort', onAbort); } catch { /* ignore */ }
}
@@ -1286,6 +1440,9 @@ const streamHandlers: Record<string, StreamHandler> = {
let wake: (() => void) | null = null;
let cancelled = signal.aborted;
let sawWorkActivity = !groupChat.busIsQuiescent(ctx.userId, cid);
+ let relayCount = 0;
+ let processCount = 0;
+ let firstProcessLogged = false;
const onAbort = () => { cancelled = true; const w = wake; wake = null; w?.(); };
if (!cancelled) signal.addEventListener('abort', onAbort, { once: true });
const unsub = groupChat.subscribeBus(ctx.userId, cid, (ev) => {
@@ -1309,6 +1466,14 @@ const streamHandlers: Record<string, StreamHandler> = {
sawWorkActivity = true;
}
}
+ relayCount += 1;
+ if (ev.type === 'process') {
+ processCount += 1;
+ if (!firstProcessLogged) {
+ firstProcessLogged = true;
+ log.info(`groupEvents first process cid=${cid} actor=${(ev as any).actor || ''} kind=${(ev as any).data?.type || ''}`);
+ }
+ }
yield ev;
if (sawWorkActivity && groupChat.busIsQuiescent(ctx.userId, cid)) return;
}
@@ -1317,6 +1482,7 @@ const streamHandlers: Record<string, StreamHandler> = {
await new Promise<void>((resolve) => { wake = resolve; });
}
} finally {
+ log.info(`groupEvents closed cid=${cid} relayed=${relayCount} process=${processCount} cancelled=${cancelled}`);
try { unsub(); } catch { /* ignore */ }
try { signal.removeEventListener?.('abort', onAbort); } catch { /* ignore */ }
}
@@ -1478,6 +1644,7 @@ export function register(): void {
const controller = new AbortController();
const state: StreamState = { cancelled: false, controller };
activeStreams.set(requestId, state);
+ log.info(`streamStart channel=${channel} requestId=${requestId} cid=${payload?.cid || payload?.id || payload?.agent_id || ''}`);
try {
const ctx = await resolveContext(event.sender);
for await (const ev of handler(payload || {}, ctx, controller.signal)) {
@@ -1490,6 +1657,7 @@ export function register(): void {
out({ type: 'error', text: (err as Error).message || String(err) });
} finally {
activeStreams.delete(requestId);
+ log.info(`streamDone channel=${channel} requestId=${requestId} cancelled=${state.cancelled}`);
out({ type: 'done' });
}
});
diff --git a/src/main/locales/en.json b/src/main/locales/en.json
index 5bcdc7e8..736ac454 100644
--- a/src/main/locales/en.json
+++ b/src/main/locales/en.json
@@ -9,6 +9,8 @@
"errors.kb_duplicate_sha1": "A file with identical content already exists in the knowledge base: {existingPath}",
"errors.kb_name_too_long": "Filename too long: \"{name}\". Keep it within ~100 ASCII chars or ~50 CJK chars.",
"errors.builtin_agent_not_editable": "Built-in agents cannot be edited through chat",
+ "errors.builtin_agent_not_editable": "Marketplace agents cannot be edited through chat",
+ "errors.builtin_skill_not_editable": "Marketplace skills cannot be edited through chat",
"errors.agent_disabled": "The agent bound to this conversation is disabled — re-enable it or switch conversation.",
"errors.no_files_to_organize": "No files to organize",
"errors.skill_write_failed": "Write failed (invalid path or skill not found)",
@@ -41,7 +43,7 @@
"skills.errors.name_too_long": "Skill name is too long (max 64 characters)",
"skills.errors.name_invalid": "Skill name must start with a letter and may contain English letters, digits, underscores (_), and hyphens (-); single spaces between letters are allowed",
"skills.errors.skill_exists": "Skill \"{name}\" already exists. Please choose another name.",
- "skills.errors.builtin_conflict": "Skill name \"{name}\" conflicts with a built-in skill. Please choose another name.",
+ "skills.errors.builtin_conflict": "Skill name \"{name}\" conflicts with a marketplace skill. Please choose another name.",
"skills.errors.url_scheme": "URL must start with http(s)://",
"skills.errors.create_failed": "Failed to create skill",
"skills.errors.validation_blocked": "Skill content failed quality validation — creation aborted",
diff --git a/src/main/locales/ja.json b/src/main/locales/ja.json
index 7b98e56b..5d8f4ddf 100644
--- a/src/main/locales/ja.json
+++ b/src/main/locales/ja.json
@@ -8,8 +8,8 @@
"errors.file_already_exists": "同じ名前のファイルがすでに存在します",
"errors.kb_duplicate_sha1": "同一内容のファイルがナレッジベースにすでに存在します: {existingPath}",
"errors.kb_name_too_long": "ファイル名が長すぎます: 「{name}」。約 100 ASCII 文字または約 50 CJK 文字以内にしてください。",
- "errors.builtin_agent_not_editable": "組み込みエージェントはチャットから編集できません",
- "errors.builtin_skill_not_editable": "組み込みスキルはチャットから編集できません",
+ "errors.builtin_agent_not_editable": "マーケットプレイスのエージェントはチャットから編集できません",
+ "errors.builtin_skill_not_editable": "マーケットプレイスのスキルはチャットから編集できません",
"errors.agent_disabled": "この会話に紐づくエージェントは無効です。再度有効にするか、会話を切り替えてください。",
"errors.no_files_to_organize": "整理するファイルがありません",
"errors.skill_write_failed": "書き込みに失敗しました(パスが無効、またはスキルが見つかりません)",
@@ -42,7 +42,7 @@
"skills.errors.name_too_long": "スキル名が長すぎます(最大 64 文字)",
"skills.errors.name_invalid": "スキル名は英字で始め、英字・数字・アンダースコア(_)・ハイフン(-)を使用できます。英字間の単一スペースも使用できます",
"skills.errors.skill_exists": "スキル「{name}」はすでに存在します。別の名前を選んでください。",
- "skills.errors.builtin_conflict": "スキル名「{name}」は組み込みスキルと競合します。別の名前を選んでください。",
+ "skills.errors.builtin_conflict": "スキル名「{name}」はマーケットプレイスのスキルと競合します。別の名前を選んでください。",
"skills.errors.url_scheme": "URL は http(s):// で始まる必要があります",
"skills.errors.create_failed": "スキルの作成に失敗しました",
"skills.errors.validation_blocked": "スキル内容が品質検証に失敗したため、作成を中止しました",
diff --git a/src/main/locales/zh.json b/src/main/locales/zh.json
index f6d5ca28..84c9d71d 100644
--- a/src/main/locales/zh.json
+++ b/src/main/locales/zh.json
@@ -9,6 +9,8 @@
"errors.kb_duplicate_sha1": "相同内容的文件已存在于知识库:{existingPath}",
"errors.kb_name_too_long": "文件名太长:「{name}」。请缩短到中文 50 字 / 英文 100 字符以内。",
"errors.builtin_agent_not_editable": "内置智能体不可通过对话编辑",
+ "errors.builtin_agent_not_editable": "平台智能体不可通过对话编辑",
+ "errors.builtin_skill_not_editable": "平台技能不可通过对话编辑",
"errors.agent_disabled": "本对话绑定的智能体已被禁用,请重新启用或换一个对话",
"errors.no_files_to_organize": "无可整理的文件",
"errors.skill_write_failed": "写入失败(路径无效或技能不存在)",
@@ -41,7 +43,7 @@
"skills.errors.name_too_long": "技能名称过长(最多 64 个字符)",
"skills.errors.name_invalid": "技能名称需以字母开头,只能包含英文字母、数字、下划线(_)、连字符(-),字母之间可以用单个空格分隔",
"skills.errors.skill_exists": "技能「{name}」已存在,请换一个名称",
- "skills.errors.builtin_conflict": "技能名「{name}」与内置技能冲突,请换一个名称",
+ "skills.errors.builtin_conflict": "技能名「{name}」与平台技能冲突,请换一个名称",
"skills.errors.url_scheme": "URL 必须以 http(s):// 开头",
"skills.errors.create_failed": "创建技能失败",
"skills.errors.validation_blocked": "技能内容未通过质量校验,已取消创建",
diff --git a/src/main/logger.ts b/src/main/logger.ts
index 5dfc5731..890efec4 100644
--- a/src/main/logger.ts
+++ b/src/main/logger.ts
@@ -62,8 +62,10 @@ function datePrefixOf(name: string): string | null {
// ── Sensitive-field redaction ────────────────────────────────────────────
-// Keys that may carry secrets — redacted regardless of nesting depth.
-// Matched case-insensitively against property names.
+// Keys that may carry secrets or PII — redacted regardless of nesting depth.
+// Matched case-insensitively against property names. `name` is intentionally
+// NOT here (too broad: agent.name / connector.name / project.name / filename
+// are all legit business values).
const REDACT_KEYS = new Set([
'key', 'apikey', 'api_key',
'access', 'refresh',
@@ -73,6 +75,11 @@ const REDACT_KEYS = new Set([
'password', 'passwd', 'pwd',
'authorization',
'cookie', 'setcookie', 'set-cookie',
+ // PII (defense-in-depth — no current call site logs these unmasked, but
+ // the rule shields against future regressions).
+ 'phone', 'mobile',
+ 'email',
+ 'username',
]);
const MASK = '***REDACTED***';
diff --git a/src/main/model/client.ts b/src/main/model/client.ts
index d2833473..a38a6adc 100644
--- a/src/main/model/client.ts
+++ b/src/main/model/client.ts
@@ -17,6 +17,8 @@
import type { AgentTool } from '#core-agent';
import {
+ abortActiveSession as _abortActiveSession,
+ abortActiveSessionsForConversation as _abortActiveSessionsForConversation,
chatWithModel as _chatWithModel,
streamChatWithModel as _streamChatWithModel,
} from './core-agent/client';
@@ -153,3 +155,5 @@ export interface ChatOptions {
export const chatWithModel = _chatWithModel;
export const streamChatWithModel = _streamChatWithModel;
+export const abortActiveSession = _abortActiveSession;
+export const abortActiveSessionsForConversation = _abortActiveSessionsForConversation;
diff --git a/src/main/model/core-agent/client.ts b/src/main/model/core-agent/client.ts
index 83da51e0..6d6e49db 100644
--- a/src/main/model/core-agent/client.ts
+++ b/src/main/model/core-agent/client.ts
@@ -68,6 +68,54 @@ function buildSkillSandboxEnv(): Record<string, string> {
return _skillSandboxEnv;
}
+type ActiveSessionAbort = {
+ abort: () => void;
+};
+
+const activeSessionAborts = new Map<string, Set<ActiveSessionAbort>>();
+
+function addActiveSessionAbort(sessionId: string, entry: ActiveSessionAbort): void {
+ let set = activeSessionAborts.get(sessionId);
+ if (!set) {
+ set = new Set();
+ activeSessionAborts.set(sessionId, set);
+ }
+ set.add(entry);
+}
+
+function removeActiveSessionAbort(sessionId: string, entry: ActiveSessionAbort): void {
+ const set = activeSessionAborts.get(sessionId);
+ if (!set) return;
+ set.delete(entry);
+ if (set.size === 0) activeSessionAborts.delete(sessionId);
+}
+
+export function abortActiveSession(sessionId: string): number {
+ const set = activeSessionAborts.get(sessionId);
+ if (!set || set.size === 0) return 0;
+ let count = 0;
+ for (const entry of Array.from(set)) {
+ try {
+ entry.abort();
+ count += 1;
+ } catch { /* already aborted */ }
+ }
+ return count;
+}
+
+export function abortActiveSessionsForConversation(cid: string): number {
+ if (!cid) return 0;
+ let count = 0;
+ const commanderSession = `gconv-${cid}`;
+ const memberPrefix = `gmember-${cid}-`;
+ for (const sessionId of Array.from(activeSessionAborts.keys())) {
+ if (sessionId === commanderSession || sessionId.startsWith(memberPrefix)) {
+ count += abortActiveSession(sessionId);
+ }
+ }
+ return count;
+}
+
/**
* Stream chat using core-agent. Yields the same events as the openclaw
* client so existing consumers don't care which backend is live.
@@ -147,6 +195,17 @@ export async function* streamChatWithModel(opts: ChatOptions): AsyncGenerator<St
let idleTimer: NodeJS.Timeout | null = null;
let idleHit = false;
let externalAbort = false;
+ let directSessionAbort = false;
+ const activeAbortEntry: ActiveSessionAbort = {
+ abort: () => {
+ directSessionAbort = true;
+ log.info(`direct session abort ${turnTag} — releasing locks immediately`);
+ controller.abort();
+ releaseSlotOnce('session-abort');
+ releaseSessionOnce('session-abort');
+ },
+ };
+ addActiveSessionAbort(sessionId, activeAbortEntry);
const resetIdle = () => {
if (idleTimer) clearTimeout(idleTimer);
@@ -284,9 +343,12 @@ export async function* streamChatWithModel(opts: ChatOptions): AsyncGenerator<St
// Metacognitive reflection is no longer triggered per-turn — it now
// runs once at app startup with a per-agent cooldown. See
// `features/reflection-trigger.ts`.
+ // runs from the background orchestrator on a 12h cycle. See
+ // `features/reflection-orchestrator.ts`. Keeping `agentRunResult`
+ // captured above for the recorder/archive payload.
void agentRunResult;
- if (externalAbort) {
+ if (externalAbort || directSessionAbort) {
// mapCoreAgentEvents may have already yielded 'error: empty response'
// for the short-circuit; tag the stream as aborted for the client.
yield { type: 'error', text: 'aborted', aborted: true };
@@ -295,10 +357,18 @@ export async function* streamChatWithModel(opts: ChatOptions): AsyncGenerator<St
yield { type: 'error', text: errText };
}
} catch (err) {
- errText = (err as Error).message || String(err);
- log.error('stream error:', err);
- yield { type: 'error', text: errText };
+ const wasAborted = externalAbort || directSessionAbort || (controller.signal.aborted && !idleHit);
+ errText = wasAborted ? 'aborted' : ((err as Error).message || String(err));
+ if (wasAborted) {
+ abortedFlag = true;
+ log.info(`stream aborted ${turnTag}`);
+ yield { type: 'error', text: errText, aborted: true };
+ } else {
+ log.error('stream error:', err);
+ yield { type: 'error', text: errText };
+ }
} finally {
+ removeActiveSessionAbort(sessionId, activeAbortEntry);
if (idleTimer) clearTimeout(idleTimer);
if (abortSignal) abortSignal.removeEventListener?.('abort', onExternalAbort);
// Heal orphan tool_use in the cached session before releasing the
diff --git a/src/main/model/core-agent/file-tools.ts b/src/main/model/core-agent/file-tools.ts
index a9c79764..7da22b84 100644
--- a/src/main/model/core-agent/file-tools.ts
+++ b/src/main/model/core-agent/file-tools.ts
@@ -14,13 +14,14 @@
* conversation's attachment dir + active workspace.
* Never triggers extract; `total_chars` is included
* only when the cache already has it.
- * - `grep_files` — cross-file text search in the same scope. text/md/
- * code → direct; pdf/docx → extract (cached); image
- * skipped.
+ * - `grep_files` — cross-file text search in that same scanned scope.
+ * text/md/code → direct; pdf/docx → extract (cached);
+ * image skipped.
*
- * Scope is enforced via `util/path-sandbox.isPathAllowed`: every tool's
- * first step is to verify the target falls under
- * [ active workspace dir, chat_attachments/<cid>/ ].
+ * Scope is enforced via `util/path-sandbox.isPathAllowed`: path-taking tools
+ * first verify the target falls under
+ * [ active workspace dir, chat_attachments/<cid>/, project files /
+ * caller-provided extra roots ].
* Paths outside that set return an explicit E_PATH_OUT_OF_SCOPE error.
*
* These tools do NOT require localExec permission — they only read from
@@ -77,8 +78,8 @@ export interface FileToolsOpts {
* Read AND write are permitted under these roots — used by per-skill edit
* chats to expose the skill dir for the `<<<skill-file>>>` tooling. */
extraRoots?: readonly string[];
- /** Read-only extra roots: file-tools (read_file / search_files / grep_files
- * / stat_file) can see these, but write-side tools (edit_file / write_file
+ /** Read-only extra roots: path-taking file tools (read_file / stat_file)
+ * can see these, but write-side tools (edit_file / write_file
* / bash / markdown_to_pdf / html_to_pdf / generate_image) cannot mutate
* paths inside. Used by the group-chat commander to inspect agent.json /
* built-in agents / skill specs without giving direct-write access — the
@@ -131,7 +132,7 @@ function guardPath(opts: FileToolsOpts, abs: string): string | null {
if (!isPathAllowed(abs, allowedRoots(opts))) {
return errText(
'E_PATH_OUT_OF_SCOPE',
- `path is outside the current conversation's visible scope (workspace + attachments): ${abs}`,
+ `path is outside the current conversation's visible scope (workspace + attachments + project files): ${abs}`,
);
}
return null;
@@ -146,7 +147,7 @@ function createReadFileTool(opts: FileToolsOpts): AgentTool {
'Read a slice of a file\'s text by absolute path.\n'
+ '\n'
+ 'Parameters:\n'
- + ' path — required. Must be inside the current workspace or this conversation\'s attachment dir.\n'
+ + ' path — required. Must be inside the current workspace, this conversation\'s attachment dir, or a listed project file.\n'
+ ' charStart — 0-based inclusive start offset. Default 0.\n'
+ ' charEnd — 0-based exclusive end offset. Default = total_chars (end of file).\n'
+ '\n'
@@ -169,7 +170,7 @@ function createReadFileTool(opts: FileToolsOpts): AgentTool {
inputSchema: {
type: 'object',
properties: {
- path: { type: 'string', description: 'Absolute path. Must be inside workspace or current attachment dir.' },
+ path: { type: 'string', description: 'Absolute path. Must be inside workspace, current attachment dir, or project files.' },
charStart: { type: 'number', description: '0-based start char (inclusive). Default 0.' },
charEnd: { type: 'number', description: '0-based end char (exclusive). Default total_chars.' },
},
@@ -278,7 +279,7 @@ function createStatFileTool(opts: FileToolsOpts): AgentTool {
+ '`total_chars` for the file — typically for a pdf/docx that has never been read.\n'
+ '\n'
+ 'Parameters:\n'
- + ' path — required. Absolute path inside workspace or current attachment dir.\n'
+ + ' path — required. Absolute path inside workspace, current attachment dir, or project files.\n'
+ '\n'
+ 'Response:\n'
+ ' <file path="..." kind="text|pdf|docx" total_chars="N"/>\n'
@@ -291,7 +292,7 @@ function createStatFileTool(opts: FileToolsOpts): AgentTool {
inputSchema: {
type: 'object',
properties: {
- path: { type: 'string', description: 'Absolute path. Must be inside workspace or current attachment dir.' },
+ path: { type: 'string', description: 'Absolute path. Must be inside workspace, current attachment dir, or project files.' },
},
required: ['path'],
},
diff --git a/src/main/paths.ts b/src/main/paths.ts
index 81c81795..b879d2d9 100644
--- a/src/main/paths.ts
+++ b/src/main/paths.ts
@@ -216,11 +216,12 @@ export const userProjectsDir = (uid: string) => path.join(userCloudRoot(ui
export const projectDir = (uid: string, pid: string) => path.join(userProjectsDir(uid), pid);
export const projectMetaFile = (uid: string, pid: string) => path.join(projectDir(uid, pid), 'project.json');
export const projectBindingsFile = (uid: string, pid: string) => path.join(projectDir(uid, pid), 'bindings.json');
+export const projectFilesDir = (uid: string, pid: string) => path.join(projectDir(uid, pid), 'files');
export const agentDir = (uid: string, agentId: string) => path.join(userAgentsDir(uid), agentId || '_default');
export const agentDefinitionFile = (uid: string, agentId: string) => path.join(agentDir(uid, agentId), 'agent.json');
// Agent metacognition (per-agent self-assessment + learning strategies).
-// Writes go through features/metacognition.ts; reflection-trigger fires
+// Writes go through features/metacognition.ts; reflection-orchestrator fires
// automatically and the `metacognition` tool lets the agent maintain it
// itself.
export const agentMetaDir = (uid: string, agentId: string) => path.join(agentDir(uid, agentId), 'meta');
diff --git a/src/main/prompts/chat_agent_in_group.md b/src/main/prompts/chat_agent_in_group.md
index b5eec0eb..566235a6 100644
--- a/src/main/prompts/chat_agent_in_group.md
+++ b/src/main/prompts/chat_agent_in_group.md
@@ -8,7 +8,7 @@ Follow the "workflow" below to complete the work the commander / user has dispat
**Two hard constraints**:
- **Stay concise**: replies center on facts and conclusions, **no filler** (drop wording like "OK, I'll help you..." or "Thank you so much for trusting me..."). The user is reading the result, not the attitude.
-- **If a dependency is missing, stop and report immediately**: skill unavailable / credentials missing / tool call failed / required input missing — **stop right away**, state clearly "what's missing + how far you got"; **do not force through** and do not bolt on some side path as a fallback.
+- **When a dependency is missing, stop and report**: credentials missing / required input missing / non-recoverable tool failure / referenced skill unavailable — **stop right away**, state clearly "what's missing + how far you got"; **do not force through** and do not bolt on a side path. **Exception**: installable external deps declared in a skill (pip / npm / CLI) follow the rule in Shared rules — install first, stop only on failure.
---
@@ -32,7 +32,7 @@ Follow the "workflow" below to complete the work the commander / user has dispat
- You can **only see** what is written into the inbound message. Tool-call results from the dispatcher, conversations of other agents, and other history of the user — **you cannot see any of those**.
- On your first wake-up, the system prepends a `<group-chat-history>` block containing the visible history; afterwards the LLM session accumulates naturally and there's no further prepending.
-- If the dispatcher wants you to use particular material, it **must** put the content into the inbound message text (file paths, summaries from the previous step, references — all carried in the inbound message text).
+- If the dispatcher wants you to use particular material, it **must** put the content into the inbound message text (file paths, summaries from the previous step, references — all carried in the inbound message text). Shared project files are the exception: when present, they are listed in the Runtime injection block below and may be read by path.
- When information is missing, **don't guess**: send a form to ask the user to fill in the key info, or write a body saying "the available info is only enough to do X; recommend filling in Y / Z to continue" — and stop. **You don't need** to find the commander; the bus handles your paused state automatically.
---
@@ -143,4 +143,6 @@ $inputs_schema
### Working directory
$working_dir
+$project_files_block
+
$output_format_hint
diff --git a/src/main/prompts/chat_commander.md b/src/main/prompts/chat_commander.md
index c7fe8ca8..faf323fd 100644
--- a/src/main/prompts/chat_commander.md
+++ b/src/main/prompts/chat_commander.md
@@ -253,11 +253,11 @@ When the system prompt has a `## Connectors` block, those services are reachable
### Attachments and files
-When a user message has an `<attachments>` prefix, each `<file name=... path=... kind=... [total_chars=...]/>` entry's `path` is the **authoritative absolute path**.
+When a user message has an `<attachments>` prefix, or the runtime context includes a `<project-files>` block, each `<file path=.../>` entry's `path` is the **authoritative absolute path**. Project-file entries intentionally include only `path`; call `stat_file(path)` only when you need metadata or extraction.
**Locating**:
-- Files in the manifest → call `read_file(path=...)` **directly**; don't `search_files` first.
-- Files NOT in the manifest → **first `search_files`** (scope = `$working_dir` + this conversation's attachment dir); not being in the manifest does not mean "invisible" — the file may be in the workspace.
+- Files in `<attachments>` or `<project-files>` → call `read_file(path=...)` **directly**; don't `search_files` first.
+- Files NOT in those blocks → **first `search_files`** (scope = `$working_dir` + this conversation's attachment dir); not being listed does not mean "invisible" — the file may be in the workspace.
- If neither has it → ask the user where the file is, or to upload it.
**`read_file` / `stat_file` semantics**:
@@ -265,7 +265,7 @@ When a user message has an `<attachments>` prefix, each `<file name=... path=...
- For pdf / docx not yet extracted, `read_file` returns `E_NEED_STAT`; call `stat_file(path)` first to trigger extraction. text has no such issue.
- image does NOT take a range; the response is a real-time compressed grayscale JPEG fed to the vision model (**you see it; the user does NOT**). Manifest entries marked `attached="inline"` already ride this turn's vision input — answer from what you see, do NOT call `read_file` on them.
-**`search_files` / `grep_files`**: scope = `$working_dir` ∪ the current conversation's attachment dir. `search_files` finds paths by filename / glob; `grep_files` searches text across files (auto-extracts pdf/docx on hit).
+**`search_files` / `grep_files`**: scope = `$working_dir` ∪ the current conversation's attachment dir. Project files are already listed explicitly in `<project-files>`; use those paths directly. `search_files` finds paths by filename / glob; `grep_files` searches text across files (auto-extracts pdf/docx on hit).
### Resource path constants
@@ -289,6 +289,8 @@ $local_exec_state
$plan_state
+$project_files_block
+
### Agents list
> Each entry shows `name / source / id / description` plus optional `inputs_schema: [...]` (structured input params — when dispatching via `dispatch_to`, write the field values into `message` in natural language). The block header lists the `read_file(<ROOT>/<id>/agent.json)` pattern + resolved ROOT values per Source.
diff --git a/src/main/prompts/chat_shared_rules.md b/src/main/prompts/chat_shared_rules.md
index d4d6baa5..a6503717 100644
--- a/src/main/prompts/chat_shared_rules.md
+++ b/src/main/prompts/chat_shared_rules.md
@@ -9,6 +9,13 @@ Switch automatically by availability (paid API / model native / built-in `web_se
- Regardless of source, **do not** stitch together a "trend summary" purely from search snippets.
3. **Failures keep going**: skip to the next URL on a fetch failure; on empty search results or `isError`, **try at least two different strategies** (current UI language ↔ English / different keywords / `site:`) before giving up — a single empty result is not a reason to give up. When everything fails, state the actual cause (empty results / preview text), not vague "API error" wording.
+## Skill external dependencies
+
+When a skill's `SKILL.md` lists external runtime requirements (pip / npm packages, CLI binaries, credentials, ...), resolve at invocation time **before** stopping:
+
+1. **Self-resolvable** (pip / npm / CLI installable via a package manager) → install once using the skill's stated command; on success continue, on failure stop and report what you tried (a `bash`-unauthorized error here means the user must enable "Settings → Local execution" and install themselves).
+2. **User-required** (API keys, OAuth tokens, paid credentials, sudo) → stop and report what the user must do; never invent a credential or call the tool with a placeholder.
+
## PDF toolchain rules
To generate a PDF you **must** use `markdown_to_pdf` (pure markdown) or `html_to_pdf` (tables / custom styles), both based on Electron `printToPDF` + system fonts. **Do not** call reportlab / pypdf / pdfkit / wkhtmltopdf / LaTeX from `bash` — CJK fonts will render as squares. **Even when the built-in PDF tools error, do not fall back** to those low-level libraries — report the error truthfully, do not silently swap paths to "patch over it".
diff --git a/src/main/smoke.ts b/src/main/smoke.ts
index 70ca44ed..b6c66fec 100644
--- a/src/main/smoke.ts
+++ b/src/main/smoke.ts
@@ -79,6 +79,7 @@ async function section(name: string, fn: () => void | Promise<void>): Promise<vo
builtin_skills_dir: '/b/s', custom_skills_dir: '/c/s',
agents_index: '- foo', plan_state: '',
os: 'macOS', working_dir: '/tmp', local_exec_state: 'x',
+ project_files_block: '',
});
assert.ok(typeof rendered === 'string' && rendered.length > 0);
});
diff --git a/src/main/util/auto-title.ts b/src/main/util/auto-title.ts
new file mode 100644
index 00000000..92510f9d
--- /dev/null
+++ b/src/main/util/auto-title.ts
@@ -0,0 +1,22 @@
+// Canonical regexes + thresholds for the heuristic conversation auto-title
+// ladder. Defined here so `src/main/features/chats.ts` (backend persistence
+// path) can import them directly. The renderer mirrors these in
+// `src/renderer/modules/auto-title.js` because §8 forbids cross-layer
+// imports from renderer to main; the parity is asserted in
+// `test/renderer/auto-title-parity.test.ts`, so adding a filler word to one
+// side without updating the other fails the test loudly.
+//
+// Filler regex notes:
+// - Both ZH and EN lists are ordered LONGEST-FIRST so the alternation
+// doesn't match a shorter prefix when a longer one is also valid
+// ("请帮我看一下..." must strip "请帮我看一下", not "请帮我" then leave "看一下").
+// - Single-character ZH fillers (bare "请") are intentionally NOT listed
+// — too likely to clip real content like "请教...".
+// - The clause regex catches the first ZH/EN sentence terminator so a
+// long input gets clipped at the first clause boundary.
+// - 30-char truncate matches sidebar visual budget.
+
+export const ZH_FILLER_RE = /^(帮我看一下|可不可以|帮我看下|帮我看看|麻烦你|帮我看|我想要|想问问|能不能|可以不|可以吗|请帮我|看一下|麻烦|帮我|我想|想问|请问|看下|看看)\s*/;
+export const EN_FILLER_RE = /^(could you|would you|can you|help me|i'?d like to|i want to|please)\s+/i;
+export const CLAUSE_RE = /[,。?!;;,.?!]/;
+export const TITLE_MAX = 30;
diff --git a/src/main/util/log-sanitize.ts b/src/main/util/log-sanitize.ts
new file mode 100644
index 00000000..c6ef53dd
--- /dev/null
+++ b/src/main/util/log-sanitize.ts
@@ -0,0 +1,38 @@
+/**
+ * Best-effort positional redaction for log text being sent off-device.
+ *
+ * The logger's `redact()` hook covers NAMED object fields (token / api_key
+ * / email / ...) but cannot see values stitched into a positional string —
+ * `log.warn(\`token=${val}\`)` slips through. This util runs AFTER a log
+ * file is assembled (e.g. before `feedback.ts` uploads it), applying narrow
+ * regex masks for shapes that are unambiguously sensitive in plain text:
+ *
+ * - "Bearer <token>" → "Bearer ***"
+ * - JWT (3 base64url segs) → "***JWT***"
+ * - email [email protected] → "a***@b.c" (matches Server mask_email)
+ * - CN mobile 13800138000 → "138****8000" (matches Server mask_phone)
+ *
+ * Why these four and not more: each additional pattern risks destroying
+ * useful diagnostic data. The four above are
+ * (a) unambiguous in plain text — a literal "Bearer X" / "eyJ.eyJ.X" /
+ * "@example.com" / "13xxxxxxxxx" rarely has a non-secret meaning,
+ * (b) consistent with the Server-side masking convention
+ * (`Server/utils/util.py::mask_phone` / `mask_email`).
+ *
+ * Opaque hex/SHA tokens (32+ chars of [a-f0-9]) are deliberately NOT
+ * masked — too many false positives (commit shas, file hashes, paths).
+ */
+export function sanitizeLogTextForUpload(text: string): string {
+ if (!text) return text;
+ return text
+ // Bearer in Authorization header — covers raw "Authorization: Bearer X"
+ // and JSON-stringified "\"authorization\":\"Bearer X\"". Case-insensitive
+ // because RFC 6750 declares the scheme name case-insensitive.
+ .replace(/Bearer\s+[A-Za-z0-9._\-~+/=]+/gi, 'Bearer ***')
+ // JWT — three base64url segments, first two starting "eyJ" (base64 of "{").
+ .replace(/\beyJ[A-Za-z0-9_\-]+\.eyJ[A-Za-z0-9_\-]+\.[A-Za-z0-9_\-]+\b/g, '***JWT***')
+ // Email — keep first local char + full domain for diagnostic value.
+ .replace(/\b([A-Za-z0-9._%+-])[A-Za-z0-9._%+-]*(@[A-Za-z0-9.-]+\.[A-Za-z]{2,})\b/g, '$1***$2')
+ // CN mobile (11 digits starting 1[3-9]) — keep first 3 + last 4.
+ .replace(/\b(1[3-9]\d)\d{4}(\d{4})\b/g, '$1****$2');
+}
diff --git a/src/renderer/index.html b/src/renderer/index.html
index dca54216..6e5beec6 100644
--- a/src/renderer/index.html
+++ b/src/renderer/index.html
@@ -124,7 +124,7 @@
</button>
<div id="conv-project-empty-banner" class="conv-project-empty-banner" style="display:none">
<span class="muted" data-i18n="project.bindings.empty_in_chat">This project has no agents or skills bound. Open the project page to add some.</span>
- <button type="button" class="btn btn-sm" id="conv-project-empty-banner-open" data-i18n="project.bindings.open_settings">Open project page</button>
+ <button type="button" class="btn btn-sm" id="conv-project-empty-banner-open" data-i18n="project.bindings.open_settings">Open</button>
</div>
<div id="chat-history" class="chat-history">
<div class="empty" data-i18n="chat.empty">No messages yet...</div>
@@ -247,7 +247,7 @@
<div class="skills-grid" id="skills-grid-custom"></div>
</div>
<div class="skills-group" id="skills-grid-builtin-group" style="display:none">
- <div class="skills-group-label" data-i18n="skills.builtin_group">Built-in</div>
+ <div class="skills-group-label" data-i18n="skills.builtin_group">Marketplace</div>
<div class="skills-grid" id="skills-grid-builtin"></div>
</div>
<div class="skills-empty" id="skills-empty" data-i18n="skills.empty">No skills yet</div>
@@ -342,7 +342,7 @@
<div class="agents-grid" id="agents-grid-custom"></div>
</div>
<div class="agents-group" id="agents-grid-builtin-group" style="display:none">
- <div class="agents-group-label" data-i18n="agents.builtin_group">Built-in</div>
+ <div class="agents-group-label" data-i18n="agents.builtin_group">Marketplace</div>
<div class="agents-grid" id="agents-grid-builtin"></div>
</div>
<div class="agents-empty" id="agents-empty" data-i18n="agents.empty">No agents yet</div>
@@ -464,34 +464,81 @@
<div class="project-detail-view">
<div class="project-detail-content" id="project-detail-content">
<div class="project-detail-header">
- <h2 class="project-detail-title" id="project-detail-title"></h2>
- <input type="text" class="project-detail-title-input" id="project-detail-title-input"
- style="display:none" autocomplete="off" spellcheck="false" />
+ <div class="project-detail-title-wrap">
+ <h2 class="project-detail-title" id="project-detail-title"></h2>
+ <input type="text" class="project-detail-title-input" id="project-detail-title-input"
+ style="display:none" autocomplete="off" spellcheck="false" />
+ </div>
<div class="project-detail-actions">
- <button type="button" class="btn btn-sm" id="project-action-new-conv" data-i18n="project.action.new_conv">+ New chat</button>
<button type="button" class="btn btn-sm" id="project-action-rename" data-i18n="project.action.rename">Rename</button>
<button type="button" class="btn btn-sm btn-danger" id="project-action-delete" data-i18n="project.action.delete">Delete</button>
</div>
</div>
- <section class="project-detail-card" id="project-detail-agents-section">
- <div class="project-detail-card-head">
- <span class="project-detail-card-title" data-i18n="project.bindings.agents_title">Agents</span>
- <button type="button" class="btn btn-sm" id="project-add-agent-btn" data-i18n="project.bindings.add_agent">Add</button>
- </div>
- <div class="project-bindings-list" id="project-agents-list"></div>
- </section>
- <!-- Skills section intentionally hidden this round: skills are
- globally available regardless of project membership. The
- storage + IPC + runner plumbing for skill bindings stays in
- place so this card can be re-introduced without a schema
- change later. -->
- <section class="project-detail-card" id="project-detail-skills-section" style="display:none">
- <div class="project-detail-card-head">
- <span class="project-detail-card-title" data-i18n="project.bindings.skills_title">Skills</span>
- <button type="button" class="btn btn-sm" id="project-add-skill-btn" data-i18n="project.bindings.add_skill">Add</button>
+ <div class="project-detail-grid">
+ <div class="project-detail-main">
+ <section class="project-composer-card">
+ <div class="project-composer-input-area">
+ <div class="chat-input-overlay">
+ <div class="chat-attachments" id="project-chat-attachments" style="display:none"></div>
+ </div>
+ <textarea id="project-chat-input" data-i18n-placeholder="project.chat.placeholder" placeholder="Type @ to choose agents, skills, connectors." rows="2"></textarea>
+ <div class="chat-bottom-bar project-composer-bottom">
+ <button type="button" class="chat-attach-btn" id="project-chat-attach-btn" data-i18n-title="project.files.add" title="Add file">+</button>
+ <button type="button" class="chat-attach-btn chat-mic-btn" id="project-chat-mic-btn" data-i18n-title="chat.mic_title" title="Voice input">
+ <span data-ui-icon="mic" data-ui-icon-class="chat-mic-icon"></span>
+ </button>
+ <button type="button" class="chat-recipient-chip" id="project-chat-recipient-chip" data-i18n-title="chat.recipient_picker_title" title="Choose agent, skill, or connector">
+ <span class="chat-recipient-prefix" data-i18n="chat.recipient_label">To: </span>
+ <span class="chat-recipient-name" id="project-chat-recipient-name" data-i18n="chat.recipient_commander">Commander</span>
+ </button>
+ <div class="chat-skill-chip" id="project-chat-skill-chip" data-target="project" style="display:none">
+ <span class="chip-label"></span>
+ <span class="chip-close" data-i18n-title="chat.chip_remove_title" title="Remove">×</span>
+ </div>
+ <button type="button" class="chat-send-btn project-chat-send-btn" id="project-chat-send-btn" data-i18n-title="chat.send_title" title="Send">
+ <span data-ui-icon="send" data-ui-icon-class="send-icon"></span>
+ </button>
+ </div>
+ </div>
+ </section>
</div>
- <div class="project-bindings-list" id="project-skills-list"></div>
- </section>
+
+ <aside class="project-detail-side">
+ <section class="project-detail-card" id="project-detail-agents-section">
+ <div class="project-detail-card-head">
+ <span class="project-detail-card-title" data-i18n="project.bindings.agents_title">Agents</span>
+ <button type="button" class="btn btn-sm" id="project-add-agent-btn" data-i18n="project.bindings.add_agent">Add</button>
+ </div>
+ <div class="project-bindings-list" id="project-agents-list"></div>
+ </section>
+
+ <section class="project-detail-card" id="project-detail-files-section">
+ <div class="project-detail-card-head">
+ <span class="project-detail-card-title" data-i18n="project.files.title">Files</span>
+ <button type="button" class="project-card-icon-btn" id="project-add-file-btn"
+ data-i18n-title="project.files.add" title="Add file" aria-label="Add file">+</button>
+ <input type="file" id="project-file-input" multiple
+ accept=".md,.markdown,.txt,.csv,.tsv,.json,.yaml,.yml,.log,.pdf,.docx,.png,.jpg,.jpeg,.webp,.gif,.mp4,.webm,.mov,.m4v,.ogv,text/markdown,text/plain,text/csv,application/json,application/x-yaml,application/pdf,application/vnd.openxmlformats-officedocument.wordprocessingml.document,image/png,image/jpeg,image/webp,image/gif,video/mp4,video/webm,video/quicktime"
+ style="display:none" />
+ </div>
+ <div class="project-files-status muted" id="project-files-status"></div>
+ <div class="project-files-list" id="project-files-list"></div>
+ </section>
+
+ <!-- Skills section intentionally hidden this round: skills are
+ globally available regardless of project membership. The
+ storage + IPC + runner plumbing for skill bindings stays in
+ place so this card can be re-introduced without a schema
+ change later. -->
+ <section class="project-detail-card" id="project-detail-skills-section" style="display:none">
+ <div class="project-detail-card-head">
+ <span class="project-detail-card-title" data-i18n="project.bindings.skills_title">Skills</span>
+ <button type="button" class="btn btn-sm" id="project-add-skill-btn" data-i18n="project.bindings.add_skill">Add</button>
+ </div>
+ <div class="project-bindings-list" id="project-skills-list"></div>
+ </section>
+ </aside>
+ </div>
</div>
</div>
</section>
@@ -1247,6 +1294,7 @@
<script src="./modules/icons.js"></script>
<script src="./modules/chat-file-viewer.js"></script>
<script src="./modules/md-view-edit.js"></script>
+ <script src="./modules/text-view-edit.js"></script>
<script src="./modules/context-menu.js"></script>
<script src="./modules/chat-md-drawer.js"></script>
<script src="./modules/strip-structural-blocks.js"></script>
@@ -1270,6 +1318,7 @@
<script src="./modules/agents.js"></script>
<script src="./modules/saved-apps.js"></script>
<script src="./modules/agent-schedule.js"></script>
+ <script src="./modules/auto-title.js"></script>
<script src="./modules/utils.js"></script>
<script src="./modules/validation-report-view.js"></script>
<script src="./modules/marketplace.js"></script>
diff --git a/src/renderer/locales/en.json b/src/renderer/locales/en.json
index 24ca6c11..f62173e8 100644
--- a/src/renderer/locales/en.json
+++ b/src/renderer/locales/en.json
@@ -149,6 +149,7 @@
"chat.new_conv_title": "New conversation",
"chat.conv_pin_title": "Pin",
"chat.conv_unpin_title": "Unpin",
+ "chat.pin_failed": "Failed to update pin state. Try again or refresh the conversation list.",
"chat.conv_del_title": "Delete",
"chat.conv_del_confirm": "Delete this conversation?",
"chat.loading": "Loading...",
@@ -195,7 +196,8 @@
"conversation_info.load_failed": "Could not load conversation info: {reason}",
"conversation_info.empty_tasks": "No tasks yet",
"conversation_info.empty_steps": "No steps",
- "conversation_info.empty_files": "No generated files yet",
+ "conversation_info.empty_files": "No files yet",
+ "conversation_info.files_truncated": "Showing first {count} files",
"conversation_info.empty_attachments": "No attachments",
"conversation_info.task_plan_fallback": "Execution plan",
"conversation_info.step_files": "Files",
@@ -317,14 +319,12 @@
"chat.created_skill_chip": "Open skill: {name}",
"chat.create_agent_failed": "Failed to create agent: {reason}",
"chat.create_agent_streaming_placeholder": "Assembling agent config…",
- "chat.create_skill_streaming_placeholder": "Assembling skill files…",
- "chat.skill_file_streaming_placeholder": "Writing {path}…",
- "chat.skill_file_streaming_placeholder_unknown": "Writing file…",
+ "chat.create_skill_streaming_placeholder": "Organizing skill config…",
"skills.title": "Skills",
"skills.add": "+ Add",
"skills.more_btn": "More",
"skills.custom_group": "Custom",
- "skills.builtin_group": "Platform",
+ "skills.builtin_group": "Marketplace",
"skills.empty": "No skills yet",
"skills.use": "Use",
"skills.edit": "Edit",
@@ -344,8 +344,8 @@
"skills.use_label": "Skill: {skill}",
"skills.use_prefix": "Use {skill} skill: {content}",
"skills.rename_title": "Click to rename",
- "skills.source_custom": "Custom",
- "skills.source_builtin": "Built-in",
+ "skills.source_custom": "custom",
+ "skills.source_marketplace": "marketplace",
"skills.name_invalid": "Name must start with a letter; allowed: letters, digits, underscore, hyphen, single internal spaces",
"skills.desc_unset": "(no description)",
"skills.no_match": "No matching skills",
@@ -374,15 +374,12 @@
"skills.delete_failed_with": "Delete failed: {reason}",
"skills.delete_confirm": "Uninstall the skill \"{name}\" and all its files?",
"skills.delete_failed_with": "Uninstall failed: {reason}",
- "skills.promote_to_builtin": "Make built-in (dev)",
- "skills.promote_confirm": "Promote the custom skill \"{name}\" to built-in? This writes to both src/builtin and data/builtin and removes the custom copy.",
- "skills.promote_failed_with": "Promote to built-in failed: {reason}",
"agents.title": "Agents",
"agents.create_btn": "+ Create",
"agents.more_btn": "More",
"agents.empty": "No agents yet",
"agents.custom_group": "Custom",
- "agents.builtin_group": "Platform",
+ "agents.builtin_group": "Marketplace",
"agents.back": "Back to agents",
"agents.update_failed": "Update failed",
"agents.use": "Use",
@@ -398,8 +395,8 @@
"agents.clear_chat_title": "Clear the edit chat for this agent",
"agents.chat_input_placeholder": "Describe what to adjust...",
"agents.unnamed": "Untitled",
- "agents.source_builtin": "Built-in",
- "agents.source_custom": "Custom",
+ "agents.source_custom": "custom",
+ "agents.source_marketplace": "marketplace",
"agents.placeholder_unset": "(unset)",
"agents.edit_btn_done": "Done",
"agents.edit_btn_edit": "Edit",
@@ -419,9 +416,6 @@
"agents.delete_confirm": "Uninstall this agent? Related conversations will also be removed.",
"agents.delete_failed": "Uninstall failed",
"agents.delete_failed_with": "Uninstall failed: {reason}",
- "agents.promote_to_builtin": "Make built-in (dev)",
- "agents.promote_confirm": "Promote the custom agent \"{name}\" to built-in? Only agent.json is copied to src/builtin and data/builtin (meta/ and self-evolved skills/ are dropped); the custom copy is removed.",
- "agents.promote_failed_with": "Promote to built-in failed: {reason}",
"agents.edit_chat_empty": "Tell me how you'd like to adjust this agent",
"agents.clear_confirm": "Clear the edit chat for this agent?",
"agents.agent_not_found": "Agent not found",
@@ -628,7 +622,7 @@
"settings.section.tools_note": "Search and image-gen API keys are stored locally only — never synced or sent off-device.",
"settings.delete": "Delete",
"settings.search.title": "Search API Key",
- "settings.search.sub": "Priority: paid API > model-native search > built-in keyless scraper.",
+ "settings.search.sub": "Priority: paid API > model-native search > bundled keyless scraper.",
"settings.search.provider": "Provider",
"settings.search.api_key": "API Key",
"settings.search.add": "+ Add search key",
@@ -638,7 +632,7 @@
"settings.search.adding": "Adding…",
"settings.search.add_ok": "Added",
"settings.search.add_failed": "Add failed",
- "settings.search.empty": "No search API key configured yet (built-in keyless scraper will be used).",
+ "settings.search.empty": "No search API key configured yet (bundled keyless scraper will be used).",
"settings.search.active_tag": "Active",
"settings.search.confirm_delete": "Delete the {provider} search API key?",
"settings.image.title": "Image generation API Key",
@@ -873,7 +867,7 @@
"devtools.archives.tools_empty": "(no tools — or pre-upgrade archive)",
"devtools.archives.tools_source_extra": "Caller-injected (e.g. group_chat commander's plan_set / agent management tools)",
"devtools.archives.tools_source_orkas": "Orkas-injected (override / extension)",
- "devtools.archives.tools_source_core_agent": "core-agent builtin",
+ "devtools.archives.tools_source_core_agent": "core-agent bundled",
"agent_modal.title": "New agent",
"agent_modal.tab_create": "Create",
"agent_modal.tab_external": "External",
@@ -972,6 +966,16 @@
"project.action.new_conv": "New chat",
"project.action.rename": "Rename",
"project.action.delete": "Delete",
+ "project.chat.placeholder": "Type @ to choose agents, skills, connectors.",
+ "project.chat.hint": "Project tasks stay independent while sharing agents and files.",
+ "project.files.title": "Files",
+ "project.files.add": "Add file",
+ "project.files.empty": "No project files yet.",
+ "project.files.remove": "Remove file",
+ "project.files.uploading": "Uploading...",
+ "project.files.upload_failed": "{name} failed to upload: {reason}",
+ "project.files.upload_failed_list": "Some files failed to upload:\n{list}",
+ "project.files.delete_failed": "Failed to delete project file, please try again",
"project.bindings.picker_search_placeholder": "Search by name or description",
"project.bindings.picker_no_match": "No matches found.",
"agents.no_project_agents": "This project has no agents bound. Open the project page to add some.",
@@ -991,7 +995,7 @@
"project.bindings.candidates_empty_skills": "No skills available to add.",
"project.bindings.stale": "Removed",
"project.bindings.empty_in_chat": "This project has no agents bound. Open the project page to add one.",
- "project.bindings.open_settings": "Open project page",
+ "project.bindings.open_settings": "Open",
"search.history_title": "Recent searches",
"search.history_clear": "Clear",
"search.failed": "Search failed: {msg}",
@@ -1010,7 +1014,7 @@
"search.scope.agent": "Agent",
"search.scope.skill": "Skill",
"search.source.custom": "Custom",
- "search.source.builtin": "Built-in",
+ "search.source.marketplace": "marketplace",
"search.no_results": "No results for \"{query}\"",
"search.role.user": "Me",
"search.role.ai": "AI",
diff --git a/src/renderer/locales/ja.json b/src/renderer/locales/ja.json
index 0404fc2d..6be76847 100644
--- a/src/renderer/locales/ja.json
+++ b/src/renderer/locales/ja.json
@@ -144,6 +144,7 @@
"chat.new_conv_title": "新規会話",
"chat.conv_pin_title": "固定",
"chat.conv_unpin_title": "固定解除",
+ "chat.pin_failed": "固定状態の更新に失敗しました。もう一度試すか、会話一覧を再読み込みしてください。",
"chat.conv_del_title": "削除",
"chat.conv_del_confirm": "この会話を削除しますか?",
"chat.loading": "読み込み中...",
@@ -190,7 +191,8 @@
"conversation_info.load_failed": "会話情報を読み込めませんでした: {reason}",
"conversation_info.empty_tasks": "タスクはまだありません",
"conversation_info.empty_steps": "ステップはありません",
- "conversation_info.empty_files": "生成されたファイルはまだありません",
+ "conversation_info.empty_files": "ファイルはまだありません",
+ "conversation_info.files_truncated": "先頭 {count} 件のファイルを表示しています",
"conversation_info.empty_attachments": "添付ファイルはありません",
"conversation_info.task_plan_fallback": "実行計画",
"conversation_info.step_files": "ファイル",
@@ -312,14 +314,12 @@
"chat.created_skill_chip": "スキルを開く: {name}",
"chat.create_agent_failed": "エージェントの作成に失敗しました: {reason}",
"chat.create_agent_streaming_placeholder": "エージェント設定を組み立てています…",
- "chat.create_skill_streaming_placeholder": "スキルファイルを組み立てています…",
- "chat.skill_file_streaming_placeholder": "{path} を書き込み中…",
- "chat.skill_file_streaming_placeholder_unknown": "ファイルを書き込み中…",
+ "chat.create_skill_streaming_placeholder": "スキル設定を整理しています…",
"skills.title": "スキル",
"skills.add": "+ 追加",
"skills.more_btn": "その他",
"skills.custom_group": "カスタム",
- "skills.builtin_group": "プラットフォーム",
+ "skills.builtin_group": "マーケットプレイス",
"skills.empty": "スキルはまだありません",
"skills.use": "使用",
"skills.edit": "編集",
@@ -340,7 +340,7 @@
"skills.use_prefix": "{skill} スキルを使用: {content}",
"skills.rename_title": "クリックして名前を変更",
"skills.source_custom": "カスタム",
- "skills.source_builtin": "組み込み",
+ "skills.source_marketplace": "マーケットプレイス",
"skills.name_invalid": "名前は英字で始めてください。使用可能: 英数字、アンダースコア、ハイフン、単一の内部スペース",
"skills.desc_unset": "(説明なし)",
"skills.no_match": "一致するスキルはありません",
@@ -367,15 +367,12 @@
"skills.save_failed": "保存に失敗しました",
"skills.delete_confirm": "スキル「{name}」とその全ファイルをアンインストールしますか?",
"skills.delete_failed_with": "アンインストールに失敗しました: {reason}",
- "skills.promote_to_builtin": "組み込みにする(開発)",
- "skills.promote_confirm": "カスタムスキル「{name}」を組み込みに昇格しますか?src/builtin と data/builtin の両方に書き込み、カスタムコピーは削除されます。",
- "skills.promote_failed_with": "組み込み化に失敗しました: {reason}",
"agents.title": "エージェント",
"agents.create_btn": "+ 作成",
"agents.more_btn": "その他",
"agents.empty": "エージェントはまだありません",
"agents.custom_group": "カスタム",
- "agents.builtin_group": "プラットフォーム",
+ "agents.builtin_group": "マーケットプレイス",
"agents.back": "エージェントに戻る",
"agents.update_failed": "更新に失敗しました",
"agents.use": "使用",
@@ -396,8 +393,8 @@
"agents.clear_chat_title": "このエージェントの編集チャットをクリア",
"agents.chat_input_placeholder": "調整内容を説明...",
"agents.unnamed": "無題",
- "agents.source_builtin": "組み込み",
"agents.source_custom": "カスタム",
+ "agents.source_marketplace": "マーケットプレイス",
"agents.placeholder_unset": "(未設定)",
"agents.edit_btn_done": "完了",
"agents.edit_btn_edit": "編集",
@@ -414,9 +411,6 @@
"agents.delete_confirm": "このエージェントをアンインストールしますか?関連する会話も削除されます。",
"agents.delete_failed": "アンインストールに失敗しました",
"agents.delete_failed_with": "アンインストールに失敗しました: {reason}",
- "agents.promote_to_builtin": "組み込みにする(開発)",
- "agents.promote_confirm": "カスタムエージェント「{name}」を組み込みに昇格しますか?agent.json のみが src/builtin と data/builtin にコピーされます(meta/ と自己進化スキルは除外)。カスタムコピーは削除されます。",
- "agents.promote_failed_with": "組み込み化に失敗しました: {reason}",
"agents.edit_chat_empty": "このエージェントをどう調整したいか教えてください",
"agents.clear_confirm": "このエージェントの編集チャットをクリアしますか?",
"agents.agent_not_found": "エージェントが見つかりません",
@@ -623,7 +617,7 @@
"settings.section.tools_note": "検索と画像生成の API キーはローカルにのみ保存され、同期や外部送信はされません。",
"settings.delete": "削除",
"settings.search.title": "検索 API キー",
- "settings.search.sub": "優先順位: 有料 API > モデルネイティブ検索 > 組み込みのキー不要スクレイパー。",
+ "settings.search.sub": "優先順位: 有料 API > モデルネイティブ検索 > 同梱のキー不要スクレイパー。",
"settings.search.provider": "プロバイダー",
"settings.search.api_key": "API キー",
"settings.search.add": "+ 検索キーを追加",
@@ -633,7 +627,7 @@
"settings.search.adding": "追加中…",
"settings.search.add_ok": "追加しました",
"settings.search.add_failed": "追加に失敗しました",
- "settings.search.empty": "検索 API キーはまだ設定されていません(組み込みのキー不要スクレイパーを使用します)。",
+ "settings.search.empty": "検索 API キーはまだ設定されていません(同梱のキー不要スクレイパーを使用します)。",
"settings.search.active_tag": "有効",
"settings.search.confirm_delete": "{provider} の検索 API キーを削除しますか?",
"settings.image.title": "画像生成 API キー",
@@ -865,7 +859,7 @@
"devtools.archives.tools_empty": "(ツールなし、またはアップグレード前のアーカイブ)",
"devtools.archives.tools_source_extra": "呼び出し元注入(例: group_chat commander の plan_set / agent 管理ツール)",
"devtools.archives.tools_source_orkas": "Orkas 注入(上書き / 拡張)",
- "devtools.archives.tools_source_core_agent": "core-agent 組み込み",
+ "devtools.archives.tools_source_core_agent": "core-agent 同梱",
"agent_modal.title": "新規エージェント",
"agent_modal.tab_create": "作成",
"agent_modal.tab_external": "外部",
@@ -964,6 +958,16 @@
"project.action.new_conv": "新規チャット",
"project.action.rename": "名前を変更",
"project.action.delete": "削除",
+ "project.chat.placeholder": "@ を入力してエージェント、スキル、コネクタを選択",
+ "project.chat.hint": "プロジェクト内のタスクは独立し、エージェントとファイルを共有します。",
+ "project.files.title": "ファイル",
+ "project.files.add": "ファイルを追加",
+ "project.files.empty": "プロジェクトファイルはまだありません。",
+ "project.files.remove": "ファイルを削除",
+ "project.files.uploading": "アップロード中...",
+ "project.files.upload_failed": "{name} のアップロードに失敗しました: {reason}",
+ "project.files.upload_failed_list": "一部のファイルをアップロードできませんでした:\n{list}",
+ "project.files.delete_failed": "プロジェクトファイルを削除できませんでした。もう一度お試しください",
"project.bindings.picker_search_placeholder": "名前または説明で検索",
"project.bindings.picker_no_match": "一致するものがありません。",
"agents.no_project_agents": "このプロジェクトに紐づくエージェントはありません。プロジェクトページを開いて追加してください。",
@@ -983,7 +987,7 @@
"project.bindings.candidates_empty_skills": "追加できるスキルはありません。",
"project.bindings.stale": "削除済み",
"project.bindings.empty_in_chat": "このプロジェクトに紐づくエージェントはありません。プロジェクトページを開いて追加してください。",
- "project.bindings.open_settings": "プロジェクトページを開く",
+ "project.bindings.open_settings": "開く",
"search.history_title": "最近の検索",
"search.history_clear": "クリア",
"search.failed": "検索に失敗しました: {msg}",
@@ -1002,7 +1006,7 @@
"search.scope.agent": "エージェント",
"search.scope.skill": "スキル",
"search.source.custom": "カスタム",
- "search.source.builtin": "組み込み",
+ "search.source.marketplace": "マーケットプレイス",
"search.no_results": "「{query}」の結果はありません",
"search.role.user": "自分",
"search.role.ai": "AI",
diff --git a/src/renderer/locales/zh.json b/src/renderer/locales/zh.json
index 27cfad68..6d39a25a 100644
--- a/src/renderer/locales/zh.json
+++ b/src/renderer/locales/zh.json
@@ -149,6 +149,7 @@
"chat.new_conv_title": "新对话",
"chat.conv_pin_title": "置顶",
"chat.conv_unpin_title": "取消置顶",
+ "chat.pin_failed": "置顶状态更新失败,请重试或刷新对话列表。",
"chat.conv_del_title": "删除",
"chat.conv_del_confirm": "删除这个对话?",
"chat.loading": "加载中...",
@@ -195,7 +196,8 @@
"conversation_info.load_failed": "对话信息加载失败:{reason}",
"conversation_info.empty_tasks": "暂无任务",
"conversation_info.empty_steps": "暂无步骤",
- "conversation_info.empty_files": "暂无生成文件",
+ "conversation_info.empty_files": "暂无文件",
+ "conversation_info.files_truncated": "仅显示前 {count} 个文件",
"conversation_info.empty_attachments": "暂无附件",
"conversation_info.task_plan_fallback": "执行计划",
"conversation_info.step_files": "文件",
@@ -318,8 +320,6 @@
"chat.create_agent_failed": "智能体创建失败:{reason}",
"chat.create_agent_streaming_placeholder": "正在整理智能体配置…",
"chat.create_skill_streaming_placeholder": "正在整理技能配置…",
- "chat.skill_file_streaming_placeholder": "正在写入 {path}…",
- "chat.skill_file_streaming_placeholder_unknown": "正在写入文件…",
"skills.title": "技能库",
"skills.add": "+ 添加",
"skills.more_btn": "更多",
@@ -345,7 +345,7 @@
"skills.use_prefix": "使用 {skill} 技能:{content}",
"skills.rename_title": "点击重命名",
"skills.source_custom": "自定义",
- "skills.source_builtin": "内置",
+ "skills.source_marketplace": "平台",
"skills.name_invalid": "名称需以字母开头,可含字母 / 数字 / 下划线 / 横线,词之间允许单个空格",
"skills.desc_unset": "(未设置描述)",
"skills.no_match": "无匹配技能",
@@ -374,9 +374,6 @@
"skills.delete_failed_with": "删除失败:{reason}",
"skills.delete_confirm": "确认卸载技能 \"{name}\" 及其所有文件?",
"skills.delete_failed_with": "卸载失败:{reason}",
- "skills.promote_to_builtin": "设为内置(开发)",
- "skills.promote_confirm": "确认把自定义技能 \"{name}\" 提升为内置?会同步写入 src/builtin 与 data/builtin,并删除原自定义目录。",
- "skills.promote_failed_with": "提升为内置失败:{reason}",
"agents.title": "智能体",
"agents.create_btn": "+ 创建",
"agents.more_btn": "更多",
@@ -398,8 +395,8 @@
"agents.clear_chat_title": "清空当前智能体的对话记录",
"agents.chat_input_placeholder": "描述需要调整的内容...",
"agents.unnamed": "未命名",
- "agents.source_builtin": "内置",
"agents.source_custom": "自定义",
+ "agents.source_marketplace": "平台",
"agents.placeholder_unset": "(未填写)",
"agents.edit_btn_done": "完成",
"agents.edit_btn_edit": "编辑",
@@ -419,9 +416,6 @@
"agents.delete_confirm": "确认卸载这个智能体?相关对话会一并删除。",
"agents.delete_failed": "卸载失败",
"agents.delete_failed_with": "卸载失败:{reason}",
- "agents.promote_to_builtin": "设为内置(开发)",
- "agents.promote_confirm": "确认把自定义智能体 \"{name}\" 提升为内置?只复制 agent.json 到 src/builtin 与 data/builtin(meta/ 与自演化 skills/ 会丢弃),并删除原自定义目录。",
- "agents.promote_failed_with": "提升为内置失败:{reason}",
"agents.edit_chat_empty": "跟我说你想怎么调整这个智能体",
"agents.clear_confirm": "确认清空当前智能体的编辑对话记录?",
"agents.agent_not_found": "智能体不存在",
@@ -628,7 +622,7 @@
"settings.section.tools_note": "搜索与图片 API Key 只保存在本机,不上云、不外发。",
"settings.delete": "删除",
"settings.search.title": "搜索 API Key",
- "settings.search.sub": "优先级:搜索 API > 模型原生搜索 > 内置免费搜索。",
+ "settings.search.sub": "优先级:搜索 API > 模型原生搜索 > 本地免费搜索。",
"settings.search.provider": "服务商",
"settings.search.api_key": "API Key",
"settings.search.add": "+ 添加搜索 Key",
@@ -638,7 +632,7 @@
"settings.search.adding": "添加中…",
"settings.search.add_ok": "已添加",
"settings.search.add_failed": "添加失败",
- "settings.search.empty": "尚未配置搜索 API Key(将使用内置免费搜索)。",
+ "settings.search.empty": "尚未配置搜索 API Key(将使用本地免费搜索)。",
"settings.search.active_tag": "当前生效",
"settings.search.confirm_delete": "确认删除 {provider} 的搜索 API Key?",
"settings.image.title": "图片生成 API Key",
@@ -873,7 +867,7 @@
"devtools.archives.tools_empty": "(无工具或非新版本存档)",
"devtools.archives.tools_source_extra": "调用方注入(如群聊 commander 的 plan_set / agent 管理工具)",
"devtools.archives.tools_source_orkas": "Orkas 注入(覆盖 / 扩展)",
- "devtools.archives.tools_source_core_agent": "core-agent 内置",
+ "devtools.archives.tools_source_core_agent": "core-agent 随附",
"agent_modal.title": "新建智能体",
"agent_modal.tab_create": "创建",
"agent_modal.tab_external": "外接",
@@ -972,6 +966,16 @@
"project.action.new_conv": "新对话",
"project.action.rename": "重命名",
"project.action.delete": "删除",
+ "project.chat.placeholder": "输入 @ 选择智能体、技能、连接器",
+ "project.chat.hint": "项目内任务彼此独立,共用项目智能体和文件。",
+ "project.files.title": "文件",
+ "project.files.add": "添加文件",
+ "project.files.empty": "暂无项目文件。",
+ "project.files.remove": "移除文件",
+ "project.files.uploading": "正在上传...",
+ "project.files.upload_failed": "{name} 上传失败:{reason}",
+ "project.files.upload_failed_list": "部分文件上传失败:\n{list}",
+ "project.files.delete_failed": "删除项目文件失败,请重试",
"project.bindings.picker_search_placeholder": "按名称或描述搜索",
"project.bindings.picker_no_match": "没有匹配结果。",
"agents.no_project_agents": "本项目尚未添加任何智能体。请打开项目页面进行配置。",
@@ -991,7 +995,7 @@
"project.bindings.candidates_empty_skills": "没有可添加的技能。",
"project.bindings.stale": "已删除",
"project.bindings.empty_in_chat": "本项目尚未添加任何智能体。请打开项目页面进行配置。",
- "project.bindings.open_settings": "打开项目页面",
+ "project.bindings.open_settings": "打开",
"search.history_title": "搜索历史",
"search.history_clear": "清空",
"search.failed": "搜索失败: {msg}",
@@ -1010,7 +1014,7 @@
"search.scope.agent": "智能体",
"search.scope.skill": "技能",
"search.source.custom": "自定义",
- "search.source.builtin": "内置",
+ "search.source.marketplace": "平台",
"search.no_results": "未找到「{query}」相关内容",
"search.role.user": "我",
"search.role.ai": "AI",
diff --git a/src/renderer/modules/agents.js b/src/renderer/modules/agents.js
index 21d4c6cf..122afabe 100644
--- a/src/renderer/modules/agents.js
+++ b/src/renderer/modules/agents.js
@@ -41,6 +41,18 @@ function _isReservedAgentName(name) {
return _RESERVED_AGENT_NAMES.has(key);
}
+function _agentSource(source) {
+ return (typeof normalizeCatalogSource === 'function')
+ ? normalizeCatalogSource(source)
+ : String(source || '');
+}
+
+function _isAgentPlatformSource(source) {
+ return (typeof isMarketplaceCatalogSource === 'function')
+ ? isMarketplaceCatalogSource(source)
+ : _agentSource(source) === 'marketplace';
+}
+
/** Version + category chips for a marketplace-installed agent. Mirrors the
* marketplace card footer so users see the same metadata in the agents grid.
* `_mpCategoriesCache` is a module-level variable in `marketplace.js` (flat top-level
@@ -106,7 +118,10 @@ async function loadAgents(forceRefresh) {
// < orkas < 全(q)'). Without this, the backend's listAgents
// internal sort (by agent_id, a 12-char nanoid) feels random to
// users. The override here fixes that.
- _agentsCache = (data.agents || []).slice().sort((a, b) => {
+ _agentsCache = (data.agents || []).map((a) => ({
+ ...a,
+ source: _agentSource(a.source),
+ })).sort((a, b) => {
if (a.source !== b.source) return a.source === 'custom' ? -1 : 1;
const ka = pinyinSortKey(a.name || a.agent_id || '');
const kb = pinyinSortKey(b.name || b.agent_id || '');
@@ -163,7 +178,7 @@ function renderAgentsList(agents) { renderAgentsGrid(agents); }
function renderAgentsGrid(agents) {
const custom = agents.filter(a => a.source === 'custom');
- const builtin = agents.filter(a => a.source === 'builtin');
+ const marketplace = agents.filter(a => _isAgentPlatformSource(a.source));
const emptyEl = document.getElementById('agents-empty');
const customGroup = document.getElementById('agents-grid-custom-group');
const builtinGroup = document.getElementById('agents-grid-builtin-group');
@@ -198,7 +213,7 @@ function renderAgentsGrid(agents) {
// Marketplace-installed (`builtin`) agents carry version + category from _install.json /
// agent.json — show them as inline chips so users see "where this came from" without
// opening the detail page. Custom agents skip — they have no published version.
- const platformChips = a.source === 'builtin' ? _agentPlatformChipsHtml(a) : '';
+ const platformChips = _isAgentPlatformSource(a.source) ? _agentPlatformChipsHtml(a) : '';
return `
<div class="agent-card${enabled ? '' : ' is-disabled'}" data-id="${escapeHtml(a.agent_id)}" data-source="${a.source}">
<div class="agent-card-header">
@@ -223,7 +238,7 @@ function renderAgentsGrid(agents) {
gridEl.innerHTML = list.map(cardHtml).join('');
};
renderGroup(custom, customGroup, customGrid);
- renderGroup(builtin, builtinGroup, builtinGrid);
+ renderGroup(marketplace, builtinGroup, builtinGrid);
for (const card of document.querySelectorAll('.agent-card')) {
const id = card.dataset.id;
@@ -379,7 +394,7 @@ window.addEventListener('i18n-change', () => {
async function refreshAgentsAfterMarketplaceReconcile() {
await loadAgents(true);
if (_agentEditing) return;
- if (_selectedAgent?.id && _selectedAgent.source === 'builtin') {
+ if (_selectedAgent?.id && _isAgentPlatformSource(_selectedAgent.source)) {
await selectAgent(_selectedAgent.id);
}
}
@@ -398,6 +413,11 @@ function _renderAgentRowMenuItems(menu, agentId) {
// Dev-only entry on builtin: tag the label so the user knows this isn't a
// normal user capability (mirrors marketplace.upload's "(dev)" treatment).
const editLabelSuffix = (a?.source === 'builtin' && false) ? t('common.dev_suffix') : '';
+ // Dev mode lifts the source guard for marketplace edit / delete.
+ const canEdit = isCustom || (_isAgentPlatformSource(a?.source) && false);
+ // Dev-only entry on builtin: tag the label so the user knows this isn't a
+ // normal user capability (mirrors marketplace.upload's "(dev)" treatment).
+ const editLabelSuffix = (_isAgentPlatformSource(a?.source) && false) ? t('common.dev_suffix') : '';
const toggleLabel = enabled ? t('component.disable') : t('component.enable');
const items = [];
if (canEdit) {
@@ -468,6 +488,7 @@ async function selectAgent(agentId) {
const res = await apiFetch(`/api/agents/${encodeURIComponent(agentId)}`);
const data = await res.json();
if (!data.ok || !data.agent) return;
+ data.agent.source = _agentSource(data.agent.source);
_selectedAgent = { id: data.agent.agent_id, name: data.agent.name, source: data.agent.source };
_renderAgentDetail(data.agent, false);
// Reset every nested scroll container — `.agents-detail-content` and
@@ -490,7 +511,7 @@ async function selectAgent(agentId) {
// `agents.js::_renderAgentDetail` and `skills.js::useSkill`. Two forms:
// - custom item: single "自定义" chip (kept for symmetry with prior chrome)
// - marketplace-installed item: category chip, without author uid / official tags
-// `item.source` distinguishes custom vs marketplace-installed; `item.category` (code) maps
+ // `item.source` distinguishes custom vs marketplace-installed; `item.category` (code) maps
// to display name via `_mpCategoriesCache` (loaded at marketplace boot, hot in localStorage).
function _renderSourceMetaHtml(item) {
if (!item) return '';
@@ -509,6 +530,7 @@ function _renderSourceMetaHtml(item) {
}
function _renderAgentDetail(agent, editing) {
+ agent = { ...agent, source: _agentSource(agent.source) };
document.getElementById('agents-detail-content').style.display = '';
const nameEl = document.getElementById('agents-detail-name');
@@ -518,8 +540,7 @@ function _renderAgentDetail(agent, editing) {
const editBtn = document.getElementById('agent-edit-btn');
nameEl.textContent = agent.name || '';
- // Header chips: custom = single "自定义" tag; marketplace-installed (source=builtin) drops
- // the "内置" wording entirely and shows category only, without author uid / official tags.
+ // Header chips: custom = single "自定义" tag; marketplace-installed items show category only.
sourceEl.innerHTML = _renderSourceMetaHtml(agent);
// Runtime slot lives at the top of the body now (not the header):
// an always-editable dropdown so the user can flip Orkas ↔ local CLI
@@ -550,6 +571,8 @@ function _renderAgentDetail(agent, editing) {
if (!editing && !localizedDesc) descEl.innerHTML = unsetHtml;
// Detail header actions, fixed order: use (icon) / edit / enable-disable / delete.
+ // Detail header actions, fixed order:
+ // use (icon) / edit / enable-disable / delete
// Edit mode hides everything except the "done" button (the relabeled
// "edit" button).
const useBtn = document.getElementById('agent-use-btn');
@@ -564,15 +587,22 @@ function _renderAgentDetail(agent, editing) {
if (enableBtn) enableBtn.style.display = editing ? 'none' : '';
if (promoteBtn) promoteBtn.style.display = (isCustom && false && !editing) ? '' : 'none';
+ const uploadBtn = document.getElementById('agent-upload-marketplace-btn');
+ const delBtn = document.getElementById('agent-delete-btn');
+ const isCustom = agent.source === 'custom';
+ const canEdit = isCustom || (_isAgentPlatformSource(agent.source) && false);
+ if (useBtn) useBtn.style.display = editing ? 'none' : '';
+ if (enableBtn) enableBtn.style.display = editing ? 'none' : '';
// Upload button visibility: gated by marketplace_dev.js's presence (OrkasOpen lacks it).
if (uploadBtn) uploadBtn.style.display = (typeof openMarketplaceUpload === 'function' && !editing) ? '' : 'none';
if (delBtn) delBtn.style.display = (canEdit && !editing) ? '' : 'none';
if (editBtn) {
editBtn.style.display = canEdit ? '' : 'none';
- // Tag the "Edit" label on builtin agents (dev-only entry); "Done" stays
+ // Tag the "Edit" label on marketplace agents (dev-only entry); "Done" stays
// bare because the user is already in edit mode and the marker would be
// redundant noise.
const editSuffix = (!editing && agent.source === 'builtin' && false) ? t('common.dev_suffix') : '';
+ const editSuffix = (!editing && _isAgentPlatformSource(agent.source) && false) ? t('common.dev_suffix') : '';
editBtn.textContent = editing ? t('agents.edit_btn_done') : (t('agents.edit_btn_edit') + editSuffix);
}
_renderAgentEnabledButton({ id: agent.agent_id, enabled: agent.enabled !== false });
@@ -602,6 +632,7 @@ function _renderAgentOutputFormatSection(agent) {
const canEdit = agent.source === 'custom'
|| (agent.source === 'builtin' && typeof isDevMode === 'function' && false);
|| (agent.source === 'builtin' && typeof isDevMode === 'function' && false);
+ || (_isAgentPlatformSource(agent.source) && typeof isDevMode === 'function' && false);
slot.innerHTML = '';
const mount = document.createElement('div');
@@ -862,6 +893,7 @@ async function _renderAgentDetailProjectDir(agent) {
const canEdit = agent.source === 'custom'
|| (agent.source === 'builtin' && typeof isDevMode === 'function' && false);
+ || (_isAgentPlatformSource(agent.source) && typeof isDevMode === 'function' && false);
const renderInfo = (info) => {
if (_selectedAgent?.id !== agent.agent_id || slot.dataset.agentId !== agent.agent_id) return;
@@ -1021,6 +1053,8 @@ function _toggleAgentFieldEditable(on) {
async function toggleAgentEditMode() {
if (!_selectedAgent) return;
if (_selectedAgent.source === 'builtin') return;
+ // Marketplace editing is dev-only; lift the source guard accordingly.
+ if (_isAgentPlatformSource(_selectedAgent.source) && !false) return;
if (_agentEditing) {
await _exitAgentEditMode();
} else {
@@ -1087,7 +1121,7 @@ function _bindAgentFieldSave() {
}
function _scheduleAgentFieldSave(field, value) {
- if (!_selectedAgent || _selectedAgent.source === 'builtin') return;
+ if (!_selectedAgent || _isAgentPlatformSource(_selectedAgent.source)) return;
_pendingAgentField = { field, value };
clearTimeout(_agentFieldSaveTimer);
_agentFieldSaveTimer = setTimeout(_flushAgentFieldSave, 800);
@@ -1482,12 +1516,18 @@ async function _autoSendAgentChat(content) {
async function deleteSelectedAgent() {
if (!_selectedAgent || _selectedAgent.source === 'builtin') return;
+ if (!_selectedAgent) return;
+ const isMarketplace = _isAgentPlatformSource(_selectedAgent.source);
+ if (isMarketplace && !false) return;
if (!(await uiConfirm(t('agents.delete_confirm')))) return;
const agentId = _selectedAgent.id;
try {
const res = await apiFetch(`/api/agents/${encodeURIComponent(agentId)}`,
{ method: 'DELETE' });
const data = await res.json();
+ const data = isMarketplace
+ ? await window.orkas.invoke('agents.builtin.delete', { agent_id: agentId })
+ : await (await apiFetch(`/api/agents/${encodeURIComponent(agentId)}`, { method: 'DELETE' })).json();
if (!data.ok) throw new Error(data.error || t('agents.delete_failed'));
_selectedAgent = null; _agentEditing = false;
document.getElementById('agents-chat-col').style.display = 'none';
@@ -1760,6 +1800,9 @@ function _resolveActiveProjectId(anchorId) {
if (anchorId === 'new-chat-recipient-chip') {
return (typeof getCommanderProjectId === 'function') ? getCommanderProjectId() : '';
}
+ if (anchorId === 'project-chat-recipient-chip') {
+ return (typeof _projectDetailPid !== 'undefined') ? (_projectDetailPid || '') : '';
+ }
if (anchorId === 'chat-recipient-chip') {
if (typeof currentCid !== 'undefined' && currentCid
&& typeof conversations !== 'undefined' && Array.isArray(conversations)) {
@@ -1867,15 +1910,17 @@ function _renderAgentPickerList(filterText) {
: agents;
// Recipient chip exposes "commander" as a virtual top entry so the user can
// switch back without an empty-state. Other anchors keep agent-only listing.
- const isRecipientPicker = anchorId === 'chat-recipient-chip' || anchorId === 'new-chat-recipient-chip';
+ const isRecipientPicker = anchorId === 'chat-recipient-chip'
+ || anchorId === 'new-chat-recipient-chip'
+ || anchorId === 'project-chat-recipient-chip';
const commanderName = t('chat.recipient_commander');
const commanderMatchesFilter = !q || commanderName.toLowerCase().includes(q);
if (!filtered.length && !(isRecipientPicker && commanderMatchesFilter)) {
listEl.innerHTML = `<div class="skill-picker-empty">${escapeHtml(t('agents.no_match'))}</div>`;
return;
}
- const groups = { custom: [], builtin: [] };
- for (const a of filtered) (groups[a.source] || groups.custom).push(a);
+ const groups = { custom: [], marketplace: [] };
+ for (const a of filtered) (groups[_agentSource(a.source)] || groups.custom).push(a);
const groupHtml = (label, list) => {
if (!list.length) return '';
return `<div class="skill-picker-group-label">${escapeHtml(label)}</div>` +
@@ -1902,7 +1947,9 @@ function _renderAgentPickerList(filterText) {
const projectEmptyHint = (!q && _pickerBoundAgentIds && _pickerBoundAgentIds.size === 0)
? `<div class="skill-picker-empty-hint">${escapeHtml(t('agents.no_project_agents'))}</div>`
: '';
- listEl.innerHTML = projectEmptyHint + commanderHtml + groupHtml(t('agents.source_custom'), groups.custom) + groupHtml(t('agents.source_builtin'), groups.builtin);
+ listEl.innerHTML = projectEmptyHint + commanderHtml
+ + groupHtml(t('agents.source_custom'), groups.custom)
+ + groupHtml(t('agents.source_marketplace'), groups.marketplace);
_bindAgentPickerListItems(listEl, anchorId);
}
@@ -1935,8 +1982,11 @@ function _renderSkillPickerList(listEl, filterText, anchorId) {
listEl.innerHTML = `<div class="skill-picker-empty">${escapeHtml(t('skills.no_match'))}</div>`;
return;
}
- const groups = { custom: [], builtin: [] };
- for (const s of filtered) (groups[s.source] || groups.custom).push(s);
+ const groups = { custom: [], marketplace: [] };
+ for (const s of filtered) {
+ const source = (typeof normalizeCatalogSource === 'function') ? normalizeCatalogSource(s.source) : s.source;
+ (groups[source] || groups.custom).push(s);
+ }
const groupHtml = (label, list) => {
if (!list.length) return '';
return `<div class="skill-picker-group-label">${escapeHtml(label)}</div>` +
@@ -1950,7 +2000,8 @@ function _renderSkillPickerList(listEl, filterText, anchorId) {
</div>`;
}).join('');
};
- listEl.innerHTML = groupHtml(t('skills.source_custom'), groups.custom) + groupHtml(t('skills.source_builtin'), groups.builtin);
+ listEl.innerHTML = groupHtml(t('skills.source_custom'), groups.custom)
+ + groupHtml(t('skills.source_marketplace'), groups.marketplace);
_bindAgentPickerListItems(listEl, anchorId);
}
@@ -2021,7 +2072,9 @@ function _moveAgentPickerActive(delta) {
}
function _targetFromPickerAnchor(anchorId) {
- return anchorId === 'new-chat-recipient-chip' ? 'new-chat' : 'conversation';
+ if (anchorId === 'new-chat-recipient-chip') return 'new-chat';
+ if (anchorId === 'project-chat-recipient-chip') return 'project';
+ return 'conversation';
}
async function _triggerPickerItem(kind, itemId, itemName, anchorId) {
@@ -2029,14 +2082,14 @@ async function _triggerPickerItem(kind, itemId, itemName, anchorId) {
if (kind === 'skill') {
setChatSkill(target, itemName || itemId);
_consumeAtKeyChar();
- const inputId = target === 'new-chat' ? 'new-chat-input' : 'chat-input';
+ const inputId = target === 'new-chat' ? 'new-chat-input' : (target === 'project' ? 'project-chat-input' : 'chat-input');
_focusInput(document.getElementById(inputId));
return;
}
if (kind === 'connector') {
setChatConnector(target, itemId, itemName || itemId);
_consumeAtKeyChar();
- const inputId = target === 'new-chat' ? 'new-chat-input' : 'chat-input';
+ const inputId = target === 'new-chat' ? 'new-chat-input' : (target === 'project' ? 'project-chat-input' : 'chat-input');
_focusInput(document.getElementById(inputId));
return;
}
@@ -2048,9 +2101,11 @@ async function _triggerPickerItem(kind, itemId, itemName, anchorId) {
// - recipient chip / @ picker → update the persistent recipient chip.
// - other anchors → spin up a fresh conversation for that agent.
async function _triggerAgent(agentId, agentName, anchorId) {
- const isRecipientAnchor = anchorId === 'chat-recipient-chip' || anchorId === 'new-chat-recipient-chip';
+ const isRecipientAnchor = anchorId === 'chat-recipient-chip'
+ || anchorId === 'new-chat-recipient-chip'
+ || anchorId === 'project-chat-recipient-chip';
if (isRecipientAnchor) {
- const target = anchorId === 'new-chat-recipient-chip' ? 'new-chat' : 'conversation';
+ const target = _targetFromPickerAnchor(anchorId);
if (agentId === '__commander__') {
setChatRecipient(target, { kind: 'commander' });
} else {
@@ -2060,7 +2115,7 @@ async function _triggerAgent(agentId, agentName, anchorId) {
// `@` is now redundant (the chip carries the recipient) and would also
// leak into the sent text — strip it.
_consumeAtKeyChar();
- const inputId = target === 'new-chat' ? 'new-chat-input' : 'chat-input';
+ const inputId = target === 'new-chat' ? 'new-chat-input' : (target === 'project' ? 'project-chat-input' : 'chat-input');
_focusInput(document.getElementById(inputId));
return;
}
@@ -2125,7 +2180,7 @@ function bindAgentPickers() {
if (active) { active.click(); e.preventDefault(); }
}
});
- for (const id of ['chat-recipient-chip', 'new-chat-recipient-chip']) {
+ for (const id of ['chat-recipient-chip', 'new-chat-recipient-chip', 'project-chat-recipient-chip']) {
const btn = document.getElementById(id);
if (!btn || btn.dataset.bound === '1') continue;
btn.dataset.bound = '1';
@@ -2198,6 +2253,12 @@ function bindAgentPickers() {
chatInput.addEventListener('keydown', onAtKey('chat-recipient-chip'));
chatInput.addEventListener('keydown', onBackspaceMention);
}
+ const projectChatInput = document.getElementById('project-chat-input');
+ if (projectChatInput && !projectChatInput.dataset.atBound) {
+ projectChatInput.dataset.atBound = '1';
+ projectChatInput.addEventListener('keydown', onAtKey('project-chat-recipient-chip'));
+ projectChatInput.addEventListener('keydown', onBackspaceMention);
+ }
// One global click handler is enough — guard with a flag on the picker.
const picker = document.getElementById('agent-picker');
if (picker && !picker.dataset.outsideBound) {
diff --git a/src/renderer/modules/auto-title.js b/src/renderer/modules/auto-title.js
new file mode 100644
index 00000000..7119ad96
--- /dev/null
+++ b/src/renderer/modules/auto-title.js
@@ -0,0 +1,50 @@
+// Heuristic conversation auto-title — mirror of `src/main/util/auto-title.ts`.
+//
+// Defined as a top-level renderer module so `_autoTitle` is a global available
+// to consumers in `agents.js` / `conversation.js` / `project-detail.js` once
+// `<script src="./modules/auto-title.js">` runs. Renderer rules (PC/CLAUDE.md
+// §8) forbid `import`/`export` in renderer files; the regex/constant pair
+// stays duplicated on this side by design.
+//
+// Drift between this file and `src/main/util/auto-title.ts` is caught by
+// `test/renderer/auto-title-parity.test.ts` (asserts regex.source/flags +
+// TITLE_MAX). Add a new filler? Update BOTH sides + the parity fixture set.
+//
+// CJS bridge at the bottom: §9 escape so the parity test can `require()` the
+// constants directly. No-op in the browser (`module` is undefined).
+
+const _AUTO_TITLE_ZH_FILLER = /^(帮我看一下|可不可以|帮我看下|帮我看看|麻烦你|帮我看|我想要|想问问|能不能|可以不|可以吗|请帮我|看一下|麻烦|帮我|我想|想问|请问|看下|看看)\s*/;
+const _AUTO_TITLE_EN_FILLER = /^(could you|would you|can you|help me|i'?d like to|i want to|please)\s+/i;
+const _AUTO_TITLE_CLAUSE = /[,。?!;;,.?!]/;
+const _AUTO_TITLE_MAX = 30;
+
+/** Returns the auto-derived sidebar title for `text`. Empty input → ''
+ * (caller is expected to fall back to its own placeholder, typically
+ * `t('chat.default_title')`). Backend equivalent: `chats.ts::autoTitle`. */
+function _autoTitle(text) {
+ const raw = String(text == null ? '' : text).trim().replace(/\s+/g, ' ');
+ if (!raw) return '';
+ let s = raw;
+ for (let i = 0; i < 5; i++) {
+ const before = s;
+ s = s.replace(_AUTO_TITLE_ZH_FILLER, '').replace(_AUTO_TITLE_EN_FILLER, '');
+ if (s === before) break;
+ }
+ s = s.trim();
+ const clauseIdx = s.search(_AUTO_TITLE_CLAUSE);
+ if (clauseIdx >= 4) s = s.slice(0, clauseIdx);
+ s = s.trim();
+ if (!s) s = raw;
+ if (s.length > _AUTO_TITLE_MAX) s = s.slice(0, _AUTO_TITLE_MAX) + '…';
+ return s;
+}
+
+if (typeof module !== 'undefined' && typeof module.exports === 'object') {
+ module.exports = {
+ _autoTitle,
+ _AUTO_TITLE_ZH_FILLER,
+ _AUTO_TITLE_EN_FILLER,
+ _AUTO_TITLE_CLAUSE,
+ _AUTO_TITLE_MAX,
+ };
+}
diff --git a/src/renderer/modules/boot.js b/src/renderer/modules/boot.js
index f32751a7..cc2e6df0 100644
--- a/src/renderer/modules/boot.js
+++ b/src/renderer/modules/boot.js
@@ -256,4 +256,5 @@ function setView(view, cid, opts = {}) {
} else {
currentCid = null;
}
+ if (typeof renderProjectsSection === 'function') renderProjectsSection();
}
diff --git a/src/renderer/modules/chat-file-viewer.js b/src/renderer/modules/chat-file-viewer.js
index 20e2871a..375b4b70 100644
--- a/src/renderer/modules/chat-file-viewer.js
+++ b/src/renderer/modules/chat-file-viewer.js
@@ -10,8 +10,12 @@
// video (.mp4/.webm/.mov/.m4v/.ogv) → <video controls>
// pdf (.pdf) → <iframe> + Chromium PDFium
// html (.html/.htm) → sandbox="allow-scripts" iframe
-// md (.md/.markdown) → renderMarkdown(text) in a <div>
-// text (.txt/.json/.csv/.code…) → <pre> with the source
+// md (.md/.markdown) → mountMdViewEdit (markdown toolbar
+// only fires here — that's the
+// contract the markdown editor
+// carries that text files don't)
+// text (.txt/.json/.csv/.code…) → mountTextViewEdit (<pre> view +
+// plain textarea edit, no toolbar)
// else → uiConfirm "Open the folder?"
//
// chat-media://local/<abs> serves pdf/html bytes (streamed via
@@ -34,14 +38,24 @@
// opts.cid — pass through when the file is a per-conv attachment, so
// main can include the cid's attachment dir in the reveal /
// read scope. Workspace-only paths can omit it.
+// opts.projectId — pass through when the file belongs to a project file
+// pool so reveal / text preview can include that scope.
let _viewerEl = null;
let _viewerBody = null;
let _viewerTitle = null;
let _viewerRevealBtn = null;
+let _viewerMdActions = null;
let _viewerKeyHandler = null;
let _viewerCurrentPath = null;
let _viewerCurrentCid = null;
+let _viewerCurrentProjectId = null;
+// Active view/edit controller — md (mountMdViewEdit) or text
+// (mountTextViewEdit). Both expose the same shape (`destroy / isDirty /
+// getMode / setMode`), so the close path can teardown without branching.
+let _viewerEditController = null;
+let _viewerDirty = false;
+let _viewerRenderSeq = 0;
const _viewerLog = (typeof createLogger === 'function')
? createLogger('chat-file-viewer')
@@ -132,6 +146,7 @@ function _ensureViewer() {
<div class="chat-file-viewer-header">
<span class="chat-file-viewer-title"></span>
<div class="chat-file-viewer-actions">
+ <div class="chat-file-viewer-md-actions"></div>
<button type="button" class="chat-file-viewer-reveal" aria-label="${revealLabel}" title="${revealLabel}">${folderIcon}</button>
<button type="button" class="chat-file-viewer-close" aria-label="${closeLabel}" title="${closeLabel}">×</button>
</div>
@@ -145,6 +160,7 @@ function _ensureViewer() {
_viewerBody = root.querySelector('.chat-file-viewer-body');
_viewerTitle = root.querySelector('.chat-file-viewer-title');
_viewerRevealBtn = root.querySelector('.chat-file-viewer-reveal');
+ _viewerMdActions = root.querySelector('.chat-file-viewer-md-actions');
// i18n change → re-label the icon-only buttons. Same lazy listener pattern
// as chat-lightbox; the singleton is created on first open so the
@@ -166,12 +182,35 @@ function _ensureViewer() {
return root;
}
+function _confirmDiscardViewerEdits() {
+ if (!_viewerDirty) return true;
+ return window.confirm(_viewerLabel('chat.md_drawer.close_confirm', 'Discard unsaved changes?'));
+}
+
+function _teardownViewerContent() {
+ if (_viewerEditController) {
+ try { _viewerEditController.destroy(); }
+ catch (err) { _viewerLog.warn('edit controller destroy threw', err); }
+ }
+ _viewerEditController = null;
+ _viewerDirty = false;
+ if (_viewerBody) _viewerBody.innerHTML = '';
+ if (_viewerMdActions) _viewerMdActions.innerHTML = '';
+ if (_viewerEl) _viewerEl.classList.remove('is-markdown', 'is-text');
+}
+
+function _typesetViewerMarkdown() {
+ if (typeof typesetMath !== 'function' || !_viewerBody) return;
+ try { typesetMath(_viewerBody); } catch (_) { /* non-fatal */ }
+}
+
async function _onRevealClick() {
const p = _viewerCurrentPath;
if (!p) return;
try {
const payload = { path: p };
if (_viewerCurrentCid) payload.cid = _viewerCurrentCid;
+ if (_viewerCurrentProjectId) payload.projectId = _viewerCurrentProjectId;
const res = await window.orkas.invoke('workspace.revealPath', payload);
if (!res || !res.ok) {
_viewerLog.warn('reveal failed', { path: p, error: res && res.error });
@@ -181,10 +220,24 @@ async function _onRevealClick() {
}
}
-function _openViewerShell(displayName) {
+function _openViewerShell(displayName, opts) {
const el = _ensureViewer();
+ if (!_confirmDiscardViewerEdits()) return null;
+ _teardownViewerContent();
+ _viewerRenderSeq += 1;
+ const absPath = opts && opts.absPath;
+ const cid = (opts && opts.cid) || null;
+ const projectId = (opts && opts.projectId) || null;
+ _viewerCurrentPath = absPath || null;
+ _viewerCurrentCid = cid;
+ _viewerCurrentProjectId = projectId;
_viewerTitle.textContent = displayName || '';
- _viewerBody.innerHTML = '';
+ // `is-markdown` / `is-text` switch the body to a flex column so an editor
+ // textarea can fill the available height. View-mode `<pre>` and view-mode
+ // markdown both rely on this same layout — see style.css §chat-file-viewer
+ // for the per-class rules.
+ if (opts && opts.kind === 'markdown') el.classList.add('is-markdown');
+ if (opts && opts.kind === 'text') el.classList.add('is-text');
el.classList.add('is-open');
el.setAttribute('aria-hidden', 'false');
if (!_viewerKeyHandler) {
@@ -192,41 +245,48 @@ function _openViewerShell(displayName) {
if (!_isViewerOpen()) return;
// IME guard: Esc commits IME composition cancel — don't double-fire close.
if (e.isComposing || e.keyCode === 229) return;
+ const active = document.activeElement;
+ if (active && active.tagName === 'TEXTAREA' && active.closest('.chat-file-viewer')) return;
if (e.key === 'Escape') closeChatFileViewer();
};
document.addEventListener('keydown', _viewerKeyHandler);
}
- return el;
+ return _viewerRenderSeq;
}
-function closeChatFileViewer() {
+function closeChatFileViewer(opts) {
if (!_viewerEl) return;
+ const force = !!(opts && opts.force);
+ if (!force && !_confirmDiscardViewerEdits()) return false;
+ _viewerRenderSeq += 1;
_viewerEl.classList.remove('is-open');
_viewerEl.setAttribute('aria-hidden', 'true');
// Drop iframe / blob src so big preview docs can be GC'd promptly when
// the user closes the overlay. Without this, hidden iframes keep the
// PDFium / HTML document alive in memory until the next reopen.
- if (_viewerBody) _viewerBody.innerHTML = '';
+ _teardownViewerContent();
_viewerCurrentPath = null;
_viewerCurrentCid = null;
+ _viewerCurrentProjectId = null;
if (_viewerKeyHandler) {
document.removeEventListener('keydown', _viewerKeyHandler);
_viewerKeyHandler = null;
}
+ return true;
}
// ── Per-kind body builders ───────────────────────────────────────────────
-function _renderPdfBody(absPath, displayName) {
- _openViewerShell(displayName);
+function _renderPdfBody(absPath, displayName, cid, projectId) {
+ if (!_openViewerShell(displayName, { kind: 'pdf', absPath, cid, projectId })) return;
const url = _chatMediaLocalUrl(absPath);
// `#toolbar=1&navpanes=0` are Chromium PDFium control hints (keep toolbar,
// hide left sidebar). Same pattern as the KB context PDF viewer.
_viewerBody.innerHTML = `<iframe class="chat-file-viewer-pdf" src="${url}#toolbar=1&navpanes=0" title="${escapeHtml(displayName || '')}"></iframe>`;
}
-function _renderHtmlBody(absPath, displayName) {
- _openViewerShell(displayName);
+function _renderHtmlBody(absPath, displayName, cid, projectId) {
+ if (!_openViewerShell(displayName, { kind: 'html', absPath, cid, projectId })) return;
const url = _chatMediaLocalUrl(absPath);
// sandbox: allow-scripts ONLY. chat-media:// is a distinct origin from
// file://, so SOP blocks parent.* access; we additionally forbid
@@ -238,47 +298,88 @@ function _renderHtmlBody(absPath, displayName) {
_viewerBody.innerHTML = `<iframe class="chat-file-viewer-html" sandbox="${sandbox}" src="${url}" title="${escapeHtml(displayName || '')}"></iframe>`;
}
-function _renderVideoBody(absPath, displayName) {
- _openViewerShell(displayName);
+function _renderVideoBody(absPath, displayName, cid, projectId) {
+ if (!_openViewerShell(displayName, { kind: 'video', absPath, cid, projectId })) return;
const url = _chatMediaLocalUrl(absPath);
_viewerBody.innerHTML = `<div class="chat-file-viewer-video-wrap"><video class="chat-file-viewer-video" controls preload="metadata" src="${url}"></video></div>`;
}
-async function _renderMarkdownBody(absPath, displayName, cid) {
- _openViewerShell(displayName);
+async function _renderMarkdownBody(absPath, displayName, cid, projectId) {
+ const seq = _openViewerShell(displayName, { kind: 'markdown', absPath, cid, projectId });
+ if (!seq) return;
_viewerBody.innerHTML = `<div class="chat-file-viewer-loading">…</div>`;
- const text = await _readTextFile(absPath, cid);
- if (text === null) return; // _readTextFile already routed to the fallback dialog
- // renderMarkdown is the single markdown surface; matches chat bubbles so
- // generated reports look identical here and inline.
- const md = (typeof renderMarkdown === 'function') ? renderMarkdown(text) : escapeHtml(text);
- _viewerBody.innerHTML = `<div class="chat-file-viewer-md">${md}</div>`;
- // Same async LaTeX typeset chat bubbles use. Skipped silently if math.js
- // isn't loaded (e.g. tests).
- if (typeof typesetMath === 'function') {
- try { typesetMath(_viewerBody); } catch (_) { /* non-fatal */ }
+ const text = await _readTextFile(absPath, cid, projectId, seq);
+ if (text === null || seq !== _viewerRenderSeq || !_isViewerOpen()) return; // _readTextFile already routed to the fallback dialog
+ if (typeof mountMdViewEdit !== 'function') {
+ _viewerLog.warn('mountMdViewEdit missing; falling back to read-only markdown preview');
+ const md = (typeof renderMarkdown === 'function') ? renderMarkdown(text) : escapeHtml(text);
+ _viewerBody.innerHTML = `<div class="chat-file-viewer-md">${md}</div>`;
+ _typesetViewerMarkdown();
+ return;
}
+ _viewerEditController = mountMdViewEdit({
+ bodyEl: _viewerBody,
+ actionsEl: _viewerMdActions,
+ source: { kind: 'workspace', absPath, cid: cid || undefined, projectId: projectId || undefined },
+ capabilities: projectId
+ ? { edit: false, save: false, delete: false, reveal: false, taskCheckbox: false }
+ : { reveal: false, delete: false },
+ initialMode: 'view',
+ initialContent: text,
+ actionIconOnly: true,
+ callbacks: {
+ onDirtyChange: (dirty) => { _viewerDirty = !!dirty; },
+ onSaved: () => _typesetViewerMarkdown(),
+ },
+ });
+ _typesetViewerMarkdown();
}
-async function _renderTextBody(absPath, displayName, cid) {
- _openViewerShell(displayName);
+async function _renderTextBody(absPath, displayName, cid, projectId) {
+ const seq = _openViewerShell(displayName, { kind: 'text', absPath, cid, projectId });
+ if (!seq) return;
_viewerBody.innerHTML = `<div class="chat-file-viewer-loading">…</div>`;
- const text = await _readTextFile(absPath, cid);
- if (text === null) return;
- _viewerBody.innerHTML = `<pre class="chat-file-viewer-text">${escapeHtml(text)}</pre>`;
+ // Pre-fetch via _readTextFile so the too_large / read-failure path falls
+ // back to the "open the folder?" dialog (same UX as the read-only path
+ // before). On success, hand the text to mountTextViewEdit; it owns the
+ // view ↔ edit transitions, save IPC, and dirty tracking from there.
+ const text = await _readTextFile(absPath, cid, projectId, seq);
+ if (text === null || seq !== _viewerRenderSeq || !_isViewerOpen()) return;
+ if (typeof mountTextViewEdit !== 'function') {
+ _viewerLog.warn('mountTextViewEdit missing; falling back to read-only text preview');
+ _viewerBody.innerHTML = `<pre class="chat-file-viewer-text">${escapeHtml(text)}</pre>`;
+ return;
+ }
+ _viewerEditController = mountTextViewEdit({
+ bodyEl: _viewerBody,
+ actionsEl: _viewerMdActions,
+ source: { absPath, cid: cid || undefined, projectId: projectId || undefined },
+ // Project files are read-only by design (the LLM owns project workspace
+ // mutations); workspace / per-conv attachments allow edit + save.
+ capabilities: projectId ? { edit: false, save: false } : { edit: true, save: true },
+ initialMode: 'view',
+ initialContent: text,
+ actionIconOnly: true,
+ callbacks: {
+ onDirtyChange: (dirty) => { _viewerDirty = !!dirty; },
+ },
+ });
}
// Fetch a file's text via IPC. Returns the text on success; on rejection
// surfaces the fallback dialog (and returns null so the caller knows to
// stop). Closes the overlay before showing the dialog so it doesn't stack
// on top of a half-built viewer.
-async function _readTextFile(absPath, cid) {
+async function _readTextFile(absPath, cid, projectId, seq) {
_viewerCurrentPath = absPath;
_viewerCurrentCid = cid || null;
+ _viewerCurrentProjectId = projectId || null;
try {
const payload = { path: absPath };
if (cid) payload.cid = cid;
+ if (projectId) payload.projectId = projectId;
const res = await window.orkas.invoke('produced.readText', payload);
+ if (seq && seq !== _viewerRenderSeq) return null;
if (res && res.ok) return String(res.text || '');
// Specifically distinguish too_large so the user sees "file is X MB,
// open in folder?" instead of a generic failure.
@@ -287,13 +388,13 @@ async function _readTextFile(absPath, cid) {
if (err === 'too_large') {
const sizeMb = ((res && res.size) || 0) / 1024 / 1024;
const capMb = ((res && res.cap) || 2 * 1024 * 1024) / 1024 / 1024;
- await _showUnsupportedDialog(absPath, cid, {
+ await _showUnsupportedDialog(absPath, cid, projectId, {
messageKey: 'chat.preview_too_large_message',
vars: { name: absPath.split(/[\\/]/).pop() || absPath, size: sizeMb.toFixed(1), cap: capMb.toFixed(0) },
fallback: `File is ${sizeMb.toFixed(1)} MB (over the ${capMb.toFixed(0)} MB preview cap). Open the containing folder?`,
});
} else {
- await _showUnsupportedDialog(absPath, cid, {
+ await _showUnsupportedDialog(absPath, cid, projectId, {
messageKey: 'chat.preview_read_failed_message',
vars: { name: absPath.split(/[\\/]/).pop() || absPath },
fallback: 'Could not read this file. Open the containing folder?',
@@ -301,9 +402,10 @@ async function _readTextFile(absPath, cid) {
}
return null;
} catch (e) {
+ if (seq && seq !== _viewerRenderSeq) return null;
_viewerLog.warn('readText threw', { path: absPath, error: String(e && e.message || e) });
closeChatFileViewer();
- await _showUnsupportedDialog(absPath, cid, {
+ await _showUnsupportedDialog(absPath, cid, projectId, {
messageKey: 'chat.preview_read_failed_message',
vars: { name: absPath.split(/[\\/]/).pop() || absPath },
fallback: 'Could not read this file. Open the containing folder?',
@@ -315,7 +417,7 @@ async function _readTextFile(absPath, cid) {
// Fallback dialog: explain to the user that we can't preview this file
// inline, and offer to open the folder it lives in. uiConfirm returns true
// on the (relabeled) primary button.
-async function _showUnsupportedDialog(absPath, cid, opts) {
+async function _showUnsupportedDialog(absPath, cid, projectId, opts) {
const name = absPath.split(/[\\/]/).pop() || absPath;
const ext = _extOf(name) || _viewerLabel('chat.preview_unsupported_no_ext', '(no extension)');
const message = opts && opts.messageKey
@@ -343,6 +445,7 @@ async function _showUnsupportedDialog(absPath, cid, opts) {
try {
const payload = { path: absPath };
if (cid) payload.cid = cid;
+ if (projectId) payload.projectId = projectId;
const res = await window.orkas.invoke('workspace.revealPath', payload);
if (!res || !res.ok) _viewerLog.warn('fallback reveal failed', { path: absPath, error: res && res.error });
} catch (err) {
@@ -355,10 +458,9 @@ async function _showUnsupportedDialog(absPath, cid, opts) {
function openChatFileViewer(absPath, displayName, opts) {
if (!absPath) return;
const cid = (opts && opts.cid) || null;
+ const projectId = (opts && opts.projectId) || null;
const name = displayName || (absPath.split(/[\\/]/).pop() || absPath);
const kind = _kindOf(name);
- _viewerCurrentPath = absPath;
- _viewerCurrentCid = cid;
if (kind === 'image') {
// Delegate — the image lightbox already has zoom / pan / keyboard. We
@@ -369,13 +471,13 @@ function openChatFileViewer(absPath, displayName, opts) {
}
return;
}
- if (kind === 'pdf') return _renderPdfBody(absPath, name);
- if (kind === 'video') return _renderVideoBody(absPath, name);
- if (kind === 'html') return _renderHtmlBody(absPath, name);
- if (kind === 'markdown') return _renderMarkdownBody(absPath, name, cid);
- if (kind === 'text') return _renderTextBody(absPath, name, cid);
+ if (kind === 'pdf') return _renderPdfBody(absPath, name, cid, projectId);
+ if (kind === 'video') return _renderVideoBody(absPath, name, cid, projectId);
+ if (kind === 'html') return _renderHtmlBody(absPath, name, cid, projectId);
+ if (kind === 'markdown') return _renderMarkdownBody(absPath, name, cid, projectId);
+ if (kind === 'text') return _renderTextBody(absPath, name, cid, projectId);
// unsupported — go straight to the dialog, never open the shell.
- return _showUnsupportedDialog(absPath, cid, {});
+ return _showUnsupportedDialog(absPath, cid, projectId, {});
}
// CJS bridge for vitest — pure functions only, per PC/CLAUDE.md §9.
diff --git a/src/renderer/modules/chat-md-drawer.js b/src/renderer/modules/chat-md-drawer.js
index 7010bb6f..f8e3fb65 100644
--- a/src/renderer/modules/chat-md-drawer.js
+++ b/src/renderer/modules/chat-md-drawer.js
@@ -1,7 +1,7 @@
// ─── Right-side markdown drawer for the chat view ───
// Hosted next to the main chat history, slides in from the right. Two source
// kinds today (defined by mountMdViewEdit):
-// workspace — clicked an LLM-produced .md / .markdown file chip
+// workspace — a caller explicitly wants a side drawer for a workspace file
// ephemeral — context-menu "临时编辑" on a chat selection
//
// Single-instance: reopening always closes the previous mount. The drawer's
diff --git a/src/renderer/modules/connectors.js b/src/renderer/modules/connectors.js
index a35f5c21..8e77f148 100644
--- a/src/renderer/modules/connectors.js
+++ b/src/renderer/modules/connectors.js
@@ -1,7 +1,7 @@
// Connectors panel — curated OAuth-only catalog of MCP-based integrations.
//
-// Layout: a single scrollable grid view with two stacked sections — "已连接" (Connected) on top,
-// "可用" (Available) below. No tab switching; same shape as the Skills panel's
+// Layout: a single scrollable grid view with two stacked sections — connected on top,
+// available below. No tab switching; same shape as the Skills panel's
// Custom / Built-in split. Both groups hide themselves when empty.
//
// Click model: only the action buttons on a card are clickable. Clicking elsewhere on a card
@@ -9,10 +9,10 @@
// click handler was firing OAuth flows by accident. Buttons stop propagation as a defence in
// depth.
//
-// OAuth flow UX: clicking "连接" just opens the system browser and returns control to the
+// OAuth flow UX: clicking Connect just opens the system browser and returns control to the
// renderer's _runConnect Promise (which the main process resolves once the deep-link callback
-// completes). **The card itself does NOT change state during the flow** — no "授权中" badge, no
-// disabled button. The user can click "连接" again at any time; server-side startOAuth
+// completes). **The card itself does NOT change state during the flow** — no authorizing badge, no
+// disabled button. The user can click Connect again at any time; server-side startOAuth
// supersedes any prior pending flow, and the renderer surfaces only real errors (the
// "superseded" message is swallowed since it's caused by the user's own re-click).
@@ -25,7 +25,7 @@ let _connectorsState = {
/** Per-card connecting flag. Set when `_runConnect` is in flight (browser-open → deep-link
* return → /oauth/exchange → MCP connect → list_tools). Visible only post-return when the
* user comes back to PC and the IPC is still running through exchange + MCP setup — without
- * this they'd see a static "连接" button with no indication anything is happening. Button
+ * this they'd see a static Connect button with no indication anything is happening. Button
* stays clickable; re-click supersedes the prior flow (see comment in `_runConnect`). */
connecting: new Set(),
};
@@ -63,7 +63,7 @@ function _deriveBundleInstance(entry) {
&& members.every((m) => m.status && m.status.kind === 'connected');
const anyErrored = members.find((m) => m.status && m.status.kind === 'error');
const accountLabel = members.map((m) => m.oauth_grant && m.oauth_grant.account_label).find(Boolean) || '';
- // For enabled: bundle is "enabled" iff every member is enabled (any disabled → show "停用-able").
+ // For enabled: bundle is "enabled" iff every member is enabled (any disabled → show Enable).
const allEnabled = members.every((m) => m.enabled !== false);
return {
id: entry.id,
@@ -182,12 +182,12 @@ function _renderCatalogCard(entry, instance) {
const accountLabel = (instance && instance.oauth_grant && instance.oauth_grant.account_label) || '';
const errorMsg = errored && instance && instance.status && instance.status.message;
- // ⋯ menu lives on connected cards only — it hosts the destructive "断开" action so it stays
+ // The ⋯ menu lives on connected cards only — it hosts the destructive disconnect action so it stays
// one click away from accidental triggers. Un-connected / errored cards still surface the
- // disconnect action as a bottom-row button (they need it visible to recover; the "停用" toggle
+ // disconnect action as a bottom-row button (they need it visible to recover; the disable toggle
// doesn't apply when there's nothing to disable).
const menuHtml = connected
- ? `<button class="connector-card-menu-btn" data-act="menu" aria-label="${escapeHtml(t('common.more') || '⋯')}">⋯</button>`
+ ? `<button class="connector-card-menu-btn" data-act="menu" aria-label="${escapeHtml(t('common.more'))}">⋯</button>`
: '';
// Bottom-row action:
@@ -197,10 +197,10 @@ function _renderCatalogCard(entry, instance) {
// before `_runConnect`'s `finally` clears the connecting flag, and the user sees
// the spinner vanish "the moment they return from the browser" even though the
// IPC chain (loadConnectors + grid re-render) is still finishing.
- // - connected: 启用 / 停用 toggle (per-user soft switch — hides instance from LLM)
- // - errored: 断开 (recover from a stuck error state)
- // - oauth_pending: disabled "敬请期待"
- // - default (uninstalled): 连接 (start OAuth)
+ // - connected: enable / disable toggle (per-user soft switch — hides instance from LLM)
+ // - errored: disconnect (recover from a stuck error state)
+ // - oauth_pending: disabled unavailable button
+ // - default (uninstalled): connect (start OAuth)
let action = '';
const isConnecting = _connectorsState.connecting && _connectorsState.connecting.has(e.id);
if (isConnecting) {
@@ -257,7 +257,7 @@ function _renderCatalogCard(entry, instance) {
return card;
}
-// Tiny absolute-positioned popover anchored under the ⋯ button — one item for now ("断开").
+// Tiny absolute-positioned popover anchored under the ⋯ button — one disconnect item for now.
// Closes on outside click / Esc / window scroll. Keeping it inline so we don't pull in a generic
// menu primitive for one use site; if a second connector-card menu item ever lands, refactor.
function _openCardMenu(anchorBtn, entry, instance) {
@@ -357,7 +357,7 @@ async function _runConnect(entry) {
try {
const res = await window.orkas.invoke('connectors.start_oauth', { catalog_id: entry.id });
if (res && res.ok && res.instance) {
- // Refresh state + re-render the grid. The new instance shows up in the "已连接" group.
+ // Refresh state + re-render the grid. The new instance shows up in the connected group.
await loadConnectors();
} else if (res && !res.ok) {
const msg = (res.error || '').toLowerCase();
diff --git a/src/renderer/modules/conversation-info.js b/src/renderer/modules/conversation-info.js
index 54e37e7e..3ac649d9 100644
--- a/src/renderer/modules/conversation-info.js
+++ b/src/renderer/modules/conversation-info.js
@@ -1,8 +1,9 @@
// ─── Conversation info side panel ───────────────────────────────────────
// Right-side companion panel for the active conversation. It summarizes the
-// structured plan, produced files, and attachments without adding new backend
-// state: everything is derived from the same history / plan / attachment
-// endpoints already used by the chat view.
+// structured plan, workspace files, and attachments. The file tab reads the
+// live conversation workspace first, then merges chip-tracked produced files
+// from history so the panel stays aligned with disk even when tools create
+// files through bash / CLI flows.
const ConversationInfo = (() => {
const _infoLog = (typeof createLogger === 'function')
@@ -13,12 +14,20 @@ const ConversationInfo = (() => {
let _open = false;
let _activeTab = 'tasks';
let _seq = 0;
+ let _taskSeq = 0;
+ let _attachmentSeq = 0;
let _loading = false;
let _error = '';
let _snapshot = {
conversation: null,
history: [],
plan: null,
+ members: [],
+ files: [],
+ fileRoot: '',
+ fileRootExists: false,
+ filesTruncated: false,
+ filesCount: 0,
attachments: [],
};
@@ -55,6 +64,19 @@ const ConversationInfo = (() => {
.filter(Boolean);
}
+ function _normalizePath(p) {
+ return String(p || '').replace(/\\/g, '/').replace(/\/+/g, '/');
+ }
+
+ function _relPathUnder(root, target) {
+ const r = _normalizePath(root).replace(/\/+$/, '');
+ const t = _normalizePath(target);
+ if (!r || !t) return '';
+ if (t === r) return '';
+ const prefix = r + '/';
+ return t.startsWith(prefix) ? t.slice(prefix.length) : '';
+ }
+
function _samePrefix(a, b) {
if (!Array.isArray(a) || !Array.isArray(b) || a.length > b.length) return false;
for (let i = 0; i < a.length; i++) {
@@ -76,8 +98,23 @@ const ConversationInfo = (() => {
return common;
}
+ // One-time warn when the shared icons helpers (`window.fileKindIconHtml` /
+ // `window.uiIconHtml`, defined by `modules/icons.js`) are missing — typical
+ // cause is an `index.html` script-list refactor that put `icons.js` after a
+ // consumer. Without this warn the panel just paints rows without icons,
+ // silently degrading; logging once lets DevTools surface the broken load
+ // order on the first render attempt. The flag is per-module-lifetime to
+ // avoid spamming on every refresh.
+ let _warnedIconsMissing = false;
+ function _warnIconsHelperMissingOnce(helperName) {
+ if (_warnedIconsMissing) return;
+ _warnedIconsMissing = true;
+ _infoLog.warn(`icons.js helper missing: ${helperName} — check index.html <script> load order`);
+ }
+
function _iconForName(name, kind) {
if (typeof window !== 'undefined' && typeof window.fileKindIconHtml === 'function') return window.fileKindIconHtml(name, kind);
+ _warnIconsHelperMissingOnce('fileKindIconHtml');
return '';
}
@@ -85,6 +122,7 @@ const ConversationInfo = (() => {
if (typeof window !== 'undefined' && typeof window.uiIconHtml === 'function') {
return window.uiIconHtml(name, className);
}
+ _warnIconsHelperMissingOnce('uiIconHtml');
return '';
}
@@ -114,12 +152,20 @@ const ConversationInfo = (() => {
async function _load(cid) {
const enc = encodeURIComponent(cid);
- const [historyData, planData, attachmentData] = await Promise.all([
+ const [historyData, planData, memberData, filesData, attachmentData] = await Promise.all([
_fetchJson(`/api/conversations/${enc}/history?limit=500`),
_fetchJson(`/api/conversations/${enc}/plan`).catch((err) => {
_infoLog.warn('plan load failed', { cid, error: err && err.message });
return { plan: null };
}),
+ _fetchJson(`/api/conversations/${enc}/members`).catch((err) => {
+ _infoLog.warn('member load failed', { cid, error: err && err.message });
+ return { actors: [] };
+ }),
+ _fetchJson(`/api/conversations/${enc}/files`).catch((err) => {
+ _infoLog.warn('file list load failed', { cid, error: err && err.message });
+ return { items: [], root: '', rootExists: false, truncated: false, count: 0 };
+ }),
_fetchJson(`/api/conversations/${enc}/attachments`).catch((err) => {
_infoLog.warn('attachment load failed', { cid, error: err && err.message });
return { items: [] };
@@ -129,10 +175,51 @@ const ConversationInfo = (() => {
conversation: historyData.conversation || null,
history: Array.isArray(historyData.history) ? historyData.history : [],
plan: planData.plan || null,
+ members: Array.isArray(memberData.actors) ? memberData.actors : [],
+ files: Array.isArray(filesData.items) ? filesData.items : [],
+ fileRoot: typeof filesData.root === 'string' ? filesData.root : '',
+ fileRootExists: filesData.rootExists === true,
+ filesTruncated: filesData.truncated === true,
+ filesCount: Number(filesData.count) || 0,
attachments: Array.isArray(attachmentData.items) ? attachmentData.items : [],
};
}
+ async function _loadTaskSnapshot(cid) {
+ const enc = encodeURIComponent(cid);
+ const [historyData, planData, memberData] = await Promise.all([
+ _fetchJson(`/api/conversations/${enc}/history?limit=500`),
+ _fetchJson(`/api/conversations/${enc}/plan`).catch((err) => {
+ _infoLog.warn('plan load failed', { cid, error: err && err.message });
+ return { plan: null };
+ }),
+ _fetchJson(`/api/conversations/${enc}/members`).catch((err) => {
+ _infoLog.warn('member load failed', { cid, error: err && err.message });
+ return { actors: [] };
+ }),
+ ]);
+ return {
+ conversation: historyData.conversation || null,
+ history: Array.isArray(historyData.history) ? historyData.history : [],
+ plan: planData.plan || null,
+ members: Array.isArray(memberData.actors) ? memberData.actors : [],
+ };
+ }
+
+ function _normalizeAttachmentItems(items) {
+ return (Array.isArray(items) ? items : [])
+ .filter((item) => item && item.status !== 'error')
+ .map((item) => ({
+ name: String(item.name || ''),
+ displayName: item.displayName ? String(item.displayName) : '',
+ kind: item.kind || _kindForName(item.name || item.displayName || ''),
+ bytes: Number(item.bytes) || 0,
+ mtime: item.mtime,
+ status: item.status || '',
+ }))
+ .filter((item) => item.name);
+ }
+
function _currentConversationTitle() {
const c = _snapshot.conversation
|| (Array.isArray(conversations) ? conversations.find((x) => x && x.conversation_id === _cid) : null);
@@ -149,27 +236,147 @@ const ConversationInfo = (() => {
return '';
}
- function _parseAnnouncementSteps(text) {
+ function _normalizeActorKey(value) {
+ return String(value || '').trim().replace(/^@+/, '').replace(/\s+/g, '').toLowerCase();
+ }
+
+ function _buildMemberLookup(members) {
+ const byKey = new Map();
+ const add = (key, actor) => {
+ const norm = _normalizeActorKey(key);
+ if (norm && actor) byKey.set(norm, actor);
+ };
+ for (const actor of Array.isArray(members) ? members : []) {
+ if (!actor || !actor.id) continue;
+ add(actor.id, actor);
+ add(actor.name, actor);
+ }
+ add('commander', { id: 'commander', kind: 'commander', name: _label('chat.recipient_commander', 'Commander') });
+ add('指挥官', { id: 'commander', kind: 'commander', name: _label('chat.recipient_commander', 'Commander') });
+ add('我自己', { id: 'commander', kind: 'commander', name: _label('chat.recipient_commander', 'Commander') });
+ add('自己', { id: 'commander', kind: 'commander', name: _label('chat.recipient_commander', 'Commander') });
+ add('user', { id: 'user', kind: 'user', name: _label('chat.from_user', 'User') });
+ add('用户', { id: 'user', kind: 'user', name: _label('chat.from_user', 'User') });
+ return { byKey };
+ }
+
+ function _normalizeAssignee(raw, lookup) {
+ const ref = String(raw || '').trim().replace(/^@+/, '').trim();
+ if (!ref) return '';
+ const actor = lookup && lookup.byKey ? lookup.byKey.get(_normalizeActorKey(ref)) : null;
+ if (actor && actor.id) return String(actor.id);
+ const key = _normalizeActorKey(ref);
+ if (key === 'commander' || key === '指挥官' || key === '我自己' || key === '自己') return 'commander';
+ if (key === 'user' || key === '用户') return 'user';
+ return ref;
+ }
+
+ function _parseAnnouncementSteps(text, lookup) {
const lines = String(text || '').split(/\r?\n/);
const steps = [];
for (const line of lines) {
const m = line.match(/^\s*(\d+)[\.\)、)]\s+(.+?)\s*$/u);
if (!m) continue;
const idx = Number(m[1]);
- const body = String(m[2] || '').replace(/[((][^))]*[))]\s*$/u, '').trim();
+ let body = String(m[2] || '').trim();
+ let assignee = '';
+ const assigneeMatch = body.match(/[((]([^))]*)[))]\s*$/u);
+ if (assigneeMatch) {
+ assignee = _normalizeAssignee(assigneeMatch[1], lookup);
+ body = body.slice(0, assigneeMatch.index).trim();
+ }
if (!body) continue;
- steps.push({ index: idx, title: body, status: 'pending' });
+ steps.push({ index: idx, title: body, status: 'pending', ...(assignee ? { assignee } : {}) });
}
return steps;
}
+ function _actorMatches(actorValue, ref, lookup) {
+ if (!actorValue || !ref) return false;
+ const actor = lookup && lookup.byKey ? lookup.byKey.get(_normalizeActorKey(actorValue)) : null;
+ const refActor = lookup && lookup.byKey ? lookup.byKey.get(_normalizeActorKey(ref)) : null;
+ const actorId = actor?.id || String(actorValue);
+ const refId = refActor?.id || String(ref);
+ if (actorId === refId) return true;
+ return _normalizeActorKey(actorId) === _normalizeActorKey(refId)
+ || _normalizeActorKey(actor?.name) === _normalizeActorKey(ref)
+ || _normalizeActorKey(actorValue) === _normalizeActorKey(refActor?.name);
+ }
+
+ function _isUserVisibleTurn(m) {
+ return !!m && !m.dispatch && !m.plan_announcement && (m.from || m.role);
+ }
+
+ function _messageTargetsAssignee(m, assignee, lookup) {
+ if (!m || !m.dispatch || !assignee) return false;
+ const to = Array.isArray(m.to) ? m.to : [];
+ if (to.some((target) => _actorMatches(target, assignee, lookup))) return true;
+ const text = String(m.text || m.content || '').trim();
+ const mention = text.match(/^@([^\s,,::]+)/u);
+ return !!mention && _actorMatches(mention[1], assignee, lookup);
+ }
+
+ function _messageCarriesStepIndex(m, step) {
+ const idx = Number(step && step.index);
+ if (!Number.isFinite(idx)) return false;
+ return Number(m && (m.triggered_step ?? m.plan_step_index ?? m.form?.plan_step_index)) === idx;
+ }
+
+ function _inferHistoricalSteps(steps, section, lookup) {
+ if (!Array.isArray(steps) || !steps.length) return [];
+ const usedDispatches = new Set();
+ const usedDoneTurns = new Set();
+ let previousDoneIdx = -1;
+ return steps.map((step) => {
+ const assignee = step.assignee || '';
+ const explicitIdx = section.findIndex((m) => _messageCarriesStepIndex(m, step));
+ const explicitDoneIdx = section.findIndex((m) =>
+ _messageCarriesStepIndex(m, step) && _isUserVisibleTurn(m));
+ const dispatchIdx = explicitIdx >= 0 ? explicitIdx : section.findIndex((m, idx) =>
+ !usedDispatches.has(idx) && _messageTargetsAssignee(m, assignee, lookup));
+ let doneIdx = -1;
+ if (explicitDoneIdx >= 0) {
+ doneIdx = explicitDoneIdx;
+ } else if (assignee === 'commander') {
+ const start = Math.max(0, previousDoneIdx + 1);
+ doneIdx = section.findIndex((m, idx) =>
+ idx >= start
+ && !usedDoneTurns.has(idx)
+ && _isUserVisibleTurn(m)
+ && _actorMatches(m.from || m.role, 'commander', lookup));
+ } else if (assignee) {
+ const start = dispatchIdx >= 0 ? dispatchIdx : Math.max(0, previousDoneIdx + 1);
+ doneIdx = section.findIndex((m, idx) =>
+ idx >= start
+ && !usedDoneTurns.has(idx)
+ && _isUserVisibleTurn(m)
+ && _actorMatches(m.from || m.role, assignee, lookup));
+ }
+ if (doneIdx >= 0) {
+ usedDoneTurns.add(doneIdx);
+ if (dispatchIdx >= 0) usedDispatches.add(dispatchIdx);
+ previousDoneIdx = Math.max(previousDoneIdx, doneIdx);
+ return { ...step, status: 'done' };
+ }
+ if (dispatchIdx >= 0) {
+ usedDispatches.add(dispatchIdx);
+ return { ...step, status: 'in_progress' };
+ }
+ return step;
+ });
+ }
+
function _buildTasks() {
const history = _snapshot.history || [];
const plan = _snapshot.plan;
+ const lookup = _buildMemberLookup(_snapshot.members);
const tasks = [];
for (let i = 0; i < history.length; i++) {
const m = history[i];
if (!m || !m.plan_announcement) continue;
+ const nextPlanIdx = history.findIndex((next, idx) => idx > i && next && next.plan_announcement);
+ const sectionEnd = nextPlanIdx >= 0 ? nextPlanIdx : history.length;
+ const parsedSteps = _parseAnnouncementSteps(m.text || m.content || '', lookup);
const fallbackTitle = _nearestPriorUserText(history, i - 1)
|| _currentConversationTitle()
|| _label('conversation_info.task_plan_fallback', 'Execution plan');
@@ -177,7 +384,8 @@ const ConversationInfo = (() => {
id: `plan-${m.id || i}`,
title: fallbackTitle,
time: m.ts || m.time || '',
- steps: _parseAnnouncementSteps(m.text || m.content || ''),
+ steps: _inferHistoricalSteps(parsedSteps, history.slice(i + 1, sectionEnd), lookup),
+ _historyIndex: i,
current: false,
});
}
@@ -195,13 +403,21 @@ const ConversationInfo = (() => {
current: true,
};
if (tasks.length) {
- tasks[tasks.length - 1] = { ...tasks[tasks.length - 1], ...current };
+ tasks[tasks.length - 1] = { ...tasks[tasks.length - 1], ...current, _historyIndex: tasks[tasks.length - 1]._historyIndex };
} else {
tasks.push(current);
}
}
- return tasks.sort((a, b) => new Date(b.time || 0).getTime() - new Date(a.time || 0).getTime());
+ return tasks
+ .sort((a, b) => {
+ const at = new Date(a.time || 0).getTime();
+ const bt = new Date(b.time || 0).getTime();
+ const timeDiff = (Number.isFinite(bt) ? bt : 0) - (Number.isFinite(at) ? at : 0);
+ if (timeDiff !== 0) return timeDiff;
+ return (b._historyIndex ?? -1) - (a._historyIndex ?? -1);
+ })
+ .map(({ _historyIndex, ...task }) => task);
}
function _statusLabel(status) {
@@ -238,7 +454,9 @@ const ConversationInfo = (() => {
if (!a) return '';
if (a === 'commander') return _label('chat.recipient_commander', 'Commander');
if (a === 'user') return _label('chat.from_user', 'User');
- return '@' + a;
+ const lookup = _buildMemberLookup(_snapshot.members);
+ const actor = lookup.byKey.get(_normalizeActorKey(a));
+ return '@' + (actor?.name || a);
}
function _renderStep(step) {
@@ -282,7 +500,7 @@ const ConversationInfo = (() => {
}).join('')}</div>`;
}
- function _collectProducedFiles() {
+ function _collectHistoryProducedFiles() {
const byPath = new Map();
for (const m of _snapshot.history || []) {
const ts = m && (m.ts || m.time || '');
@@ -305,11 +523,60 @@ const ConversationInfo = (() => {
return Array.from(byPath.values()).sort((a, b) => String(a.path).localeCompare(String(b.path)));
}
+ function _collectVisibleFiles() {
+ const byPath = new Map();
+ const fileRoot = _snapshot.fileRoot || '';
+ const workspaceFiles = Array.isArray(_snapshot.files) ? _snapshot.files : [];
+ for (const item of workspaceFiles) {
+ const p = item && item.path ? String(item.path) : '';
+ if (!p) continue;
+ const key = _normalizePath(p);
+ const relPath = item.relPath ? String(item.relPath) : _relPathUnder(fileRoot, p);
+ byPath.set(key, {
+ path: p,
+ relPath,
+ name: item.name || _baseName(p),
+ time: item.mtime ? new Date(Number(item.mtime)).toISOString() : '',
+ bytes: Number(item.bytes) || 0,
+ source: 'workspace',
+ });
+ }
+
+ const hasAuthoritativeWorkspaceSnapshot = !!fileRoot && _snapshot.fileRootExists === true;
+ for (const produced of _collectHistoryProducedFiles()) {
+ const p = produced && produced.path ? String(produced.path) : '';
+ if (!p) continue;
+ const key = _normalizePath(p);
+ if (byPath.has(key)) continue;
+ const relPath = _relPathUnder(fileRoot, p);
+ if (relPath && hasAuthoritativeWorkspaceSnapshot && !_snapshot.filesTruncated) {
+ // The workspace snapshot is authoritative for files under its root.
+ // If a produced file was deleted or renamed, don't keep showing the
+ // stale history record.
+ continue;
+ }
+ byPath.set(key, {
+ ...produced,
+ relPath,
+ name: _baseName(p),
+ source: 'produced',
+ });
+ }
+
+ return Array.from(byPath.values()).sort((a, b) => {
+ const ar = a.relPath || a.path || '';
+ const br = b.relPath || b.path || '';
+ return String(ar).localeCompare(String(br));
+ });
+ }
+
function _buildFileTree(files) {
const root = { dirs: new Map(), files: [] };
- const common = _commonDirSegments(files.map((f) => f.path));
+ const hasRelPaths = files.some((f) => f && f.relPath);
+ const common = hasRelPaths ? [] : _commonDirSegments(files.map((f) => f.path));
for (const file of files) {
- const all = _splitPath(file.path);
+ const treePath = hasRelPaths ? (file.relPath || _baseName(file.path)) : file.path;
+ const all = _splitPath(treePath);
const rel = _samePrefix(common, all) ? all.slice(common.length) : all;
const parts = rel.length ? rel : [_baseName(file.path)];
let node = root;
@@ -326,7 +593,7 @@ const ConversationInfo = (() => {
const dirs = Array.from(node.dirs.entries()).sort((a, b) => a[0].localeCompare(b[0]));
const files = node.files.slice().sort((a, b) => a.name.localeCompare(b.name));
const dirHtml = dirs.map(([name, child]) => `
- <details class="conversation-info-dir" style="--depth:${depth}" open>
+ <details class="conversation-info-dir" style="--depth:${depth}">
<summary class="conversation-info-dir-summary">
<span class="conversation-info-dir-folder-icon conversation-info-dir-folder-closed">${_uiIcon('folder', 'ui-icon conversation-info-dir-svg-icon')}</span>
<span class="conversation-info-dir-folder-icon conversation-info-dir-folder-open">${_uiIcon('folder-open', 'ui-icon conversation-info-dir-svg-icon')}</span>
@@ -337,7 +604,7 @@ const ConversationInfo = (() => {
`).join('');
const fileHtml = files.map((file) => `
<button type="button" class="conversation-info-file" style="--depth:${depth}"
- data-file-path="${escapeHtml(file.path)}" title="${escapeHtml(file.path)}">
+ data-file-path="${escapeHtml(file.path)}" draggable="true" title="${escapeHtml(file.path)}">
<span class="conversation-info-file-icon">${_iconForName(file.name)}</span>
<span class="conversation-info-file-name">${escapeHtml(file.name)}</span>
</button>
@@ -346,12 +613,15 @@ const ConversationInfo = (() => {
}
function _renderFiles() {
- const files = _collectProducedFiles();
+ const files = _collectVisibleFiles();
if (!files.length) {
- return `<div class="conversation-info-empty">${escapeHtml(_label('conversation_info.empty_files', 'No generated files yet'))}</div>`;
+ return `<div class="conversation-info-empty">${escapeHtml(_label('conversation_info.empty_files', 'No files yet'))}</div>`;
}
const tree = _buildFileTree(files);
- return `<div class="conversation-info-tree">${_renderTreeNode(tree, 0)}</div>`;
+ const trunc = _snapshot.filesTruncated
+ ? `<div class="conversation-info-empty is-small">${escapeHtml(_label('conversation_info.files_truncated', 'Showing first {count} files', { count: _snapshot.filesCount || files.length }))}</div>`
+ : '';
+ return `${trunc}<div class="conversation-info-tree">${_renderTreeNode(tree, 0)}</div>`;
}
function _collectConversationAttachments() {
@@ -371,6 +641,7 @@ const ConversationInfo = (() => {
if (!name) continue;
byName.set(name, {
name,
+ displayName: item.displayName ? String(item.displayName) : '',
kind: item.kind || _kindForName(name),
bytes: Number(item.bytes) || 0,
time: item.mtime ? new Date(Number(item.mtime) * 1000).toISOString() : '',
@@ -392,13 +663,14 @@ const ConversationInfo = (() => {
}
return `<div class="conversation-info-attachment-list">${items.map((item) => {
const name = String(item.name || '');
+ const label = String(item.displayName || item.name || '');
const size = _formatBytes(item.bytes);
const meta = [item.kind || '', size].filter(Boolean).join(' · ');
return `
- <button type="button" class="conversation-info-attachment" data-attachment-name="${escapeHtml(name)}" title="${escapeHtml(name)}">
- <span class="conversation-info-file-icon">${_iconForName(name, item.kind)}</span>
+ <button type="button" class="conversation-info-attachment" data-attachment-name="${escapeHtml(name)}" title="${escapeHtml(label)}">
+ <span class="conversation-info-file-icon">${_iconForName(label, item.kind)}</span>
<span class="conversation-info-attachment-main">
- <span class="conversation-info-file-name">${escapeHtml(name)}</span>
+ <span class="conversation-info-file-name">${escapeHtml(label)}</span>
${meta ? `<span class="conversation-info-attachment-meta">${escapeHtml(meta)}</span>` : ''}
</span>
</button>
@@ -471,13 +743,65 @@ const ConversationInfo = (() => {
}
}
+ async function refreshAttachments(cid, opts = {}) {
+ const target = cid || _cid;
+ if (!target || target !== _cid || !_open) return;
+ const items = Array.isArray(opts.items) ? _normalizeAttachmentItems(opts.items) : null;
+ if (items) {
+ _snapshot = { ..._snapshot, attachments: items };
+ if (_activeTab === 'attachments') _renderBody();
+ return;
+ }
+
+ const seq = ++_attachmentSeq;
+ try {
+ const data = await _fetchJson(`/api/conversations/${encodeURIComponent(target)}/attachments`);
+ if (seq !== _attachmentSeq || target !== _cid) return;
+ _snapshot = {
+ ..._snapshot,
+ attachments: Array.isArray(data.items) ? data.items : [],
+ };
+ if (_activeTab === 'attachments') _renderBody();
+ } catch (err) {
+ _infoLog.warn('attachment refresh failed', { cid: target, error: err && err.message });
+ }
+ }
+
+ async function refreshTasks(cid, opts = {}) {
+ const target = cid || _cid;
+ if (!target || target !== _cid || !_open) return;
+ const seq = ++_taskSeq;
+ const silent = !!opts.silent;
+ if (!silent && _activeTab === 'tasks') {
+ _loading = true;
+ _error = '';
+ _renderBody();
+ }
+ try {
+ const partial = await _loadTaskSnapshot(target);
+ if (seq !== _taskSeq || target !== _cid) return;
+ _snapshot = { ..._snapshot, ...partial };
+ _error = '';
+ } catch (err) {
+ if (seq !== _taskSeq || target !== _cid) return;
+ _error = (err && err.message) || String(err);
+ } finally {
+ if (seq === _taskSeq && target === _cid) {
+ _loading = false;
+ if (_activeTab === 'tasks') _renderBody();
+ }
+ }
+ }
+
function bind(cid) {
_cid = cid || null;
_open = false;
- _snapshot = { conversation: null, history: [], plan: null, attachments: [] };
+ _snapshot = { conversation: null, history: [], plan: null, members: [], files: [], fileRoot: '', fileRootExists: false, filesTruncated: false, filesCount: 0, attachments: [] };
_error = '';
_loading = false;
_seq++;
+ _taskSeq++;
+ _attachmentSeq++;
_syncChrome();
_renderBody();
if (_open && _cid) refresh(_cid);
@@ -542,6 +866,18 @@ const ConversationInfo = (() => {
_openAttachment(attachment.dataset.attachmentName || '');
}
});
+ body.addEventListener('dragstart', (ev) => {
+ const file = ev.target.closest('.conversation-info-file[data-file-path]');
+ if (!file || !ev.dataTransfer) return;
+ const path = file.dataset.filePath || '';
+ if (!path) return;
+ const name = (file.querySelector('.conversation-info-file-name')?.textContent || _baseName(path)).trim();
+ try {
+ ev.dataTransfer.effectAllowed = 'copy';
+ ev.dataTransfer.setData('application/x-orkas-file', JSON.stringify({ path, name }));
+ ev.dataTransfer.setData('text/plain', path);
+ } catch (_) { /* best-effort */ }
+ });
}
_syncChrome();
_renderBody();
@@ -561,7 +897,8 @@ const ConversationInfo = (() => {
bind,
unbind,
refresh,
- refreshAttachments: refresh,
+ refreshTasks,
+ refreshAttachments,
};
})();
diff --git a/src/renderer/modules/conversation.js b/src/renderer/modules/conversation.js
index aac78309..65f79b49 100644
--- a/src/renderer/modules/conversation.js
+++ b/src/renderer/modules/conversation.js
@@ -215,6 +215,8 @@ function _initAllMentionMirrors() {
if (chatInput) _initMentionMirror(chatInput);
const newChatInput = document.getElementById('new-chat-input');
if (newChatInput) _initMentionMirror(newChatInput);
+ const projectChatInput = document.getElementById('project-chat-input');
+ if (projectChatInput) _initMentionMirror(projectChatInput);
}
const CHAT_INPUT_RESERVE_FALLBACK = 140;
@@ -275,6 +277,7 @@ const _RECIPIENT_LS_KEY = 'chat.recipientByCid';
let _recipientByCid = {}; // { [cid]: {kind,id,name} } — agents only; commander = absence
let _newChatRecipient = { ..._COMMANDER }; // ephemeral, reset on view-enter
let _pendingNewChatRecipient = null; // captured at send time, transferred to new cid
+let _projectChatRecipient = { ..._COMMANDER }; // ephemeral recipient for project detail composer
function _loadRecipientMap() {
try {
@@ -305,6 +308,7 @@ function _normRecipient(next) {
function _activeRecipient(target) {
if (target === 'new-chat') return _newChatRecipient;
+ if (target === 'project') return _projectChatRecipient;
if (currentCid && _recipientByCid[currentCid]) return _recipientByCid[currentCid];
return _COMMANDER;
}
@@ -340,6 +344,8 @@ function setChatRecipient(target, next, _opts = {}) {
if (!r) return;
if (target === 'new-chat') {
_newChatRecipient = r;
+ } else if (target === 'project') {
+ _projectChatRecipient = r;
} else if (currentCid) {
if (r.kind === 'commander') delete _recipientByCid[currentCid];
else _recipientByCid[currentCid] = r;
@@ -364,9 +370,11 @@ function _transferNewChatRecipientTo(cid) {
}
function _renderRecipientChip(target) {
- const targets = target ? [target] : ['conversation', 'new-chat'];
+ const targets = target ? [target] : ['conversation', 'new-chat', 'project'];
for (const tg of targets) {
- const id = tg === 'new-chat' ? 'new-chat-recipient-name' : 'chat-recipient-name';
+ const id = tg === 'new-chat'
+ ? 'new-chat-recipient-name'
+ : (tg === 'project' ? 'project-chat-recipient-name' : 'chat-recipient-name');
const nameEl = document.getElementById(id);
if (!nameEl) continue;
const r = _activeRecipient(tg);
@@ -496,10 +504,77 @@ function _forgetCidRecipient(cid) {
const _autoEvalInflight = new Set(); // cid set
const _latestInFlight = new Map(); // cid → string[] (mirrors state_changed.state.in_flight)
const _runtimeRecoveryTimers = new Map(); // cid → timeout id
+const _lastGroupWorkEventAt = new Map(); // cid → ms timestamp of process/artifact/assistant message
+const _groupObserverCtrls = new Map(); // cid → recovery observer controller
+const _groupEventDedupe = new Map(); // cid → recently handled visible group events
// cid → most recent interactive agent that produced an end-of-turn message.
// Used as Phase 1.5 sticky when there's no plan + nobody in flight (single
// agent dispatch via @-mention finished a turn but user hasn't replied yet).
const _lastInteractiveTurnAgent = new Map();
+const GROUP_EVENT_DEDUPE_LIMIT = 600;
+
+function _groupEventDedupeKey(evData) {
+ if (!evData || typeof evData !== 'object') return '';
+ if (evData.type === 'message' && evData.msg) {
+ const gm = evData.msg;
+ if (gm.id) return `message:${gm.id}:${evData.turn_end ? 1 : 0}`;
+ return `message:${gm.from || ''}:${gm.to || ''}:${gm.ts || ''}:${(gm.text || '').length}:${evData.turn_end ? 1 : 0}`;
+ }
+ if (evData.type === 'artifact_created') {
+ const artifact = evData.artifact || {};
+ if (artifact.id) return `artifact:${evData.actor || artifact.agent_id || ''}:${artifact.id}`;
+ }
+ if (evData.type === 'process') {
+ const actor = String(evData.actor || '');
+ const data = evData.data || {};
+ if (data.type === 'delta') return '';
+ if (data.type === 'progress' && data.text) {
+ return `process:progress:${actor}:${String(data.text).slice(0, 500)}`;
+ }
+ if (data.type === 'event') {
+ const evt = data.event || {};
+ if (evt.stream === 'assistant') return '';
+ const payload = evt.data || {};
+ const eventId = payload.id || payload.call_id || payload.tool_call_id || payload.request_id || '';
+ const phase = payload.phase || payload.status || payload.type || '';
+ const name = payload.name || payload.tool || payload.command || '';
+ if (eventId || phase || name) {
+ return `process:event:${actor}:${evt.stream || ''}:${eventId}:${phase}:${name}`;
+ }
+ }
+ }
+ return '';
+}
+
+// Mirror of the carve-outs in `_groupEventDedupeKey`: text-streaming events
+// (per-token deltas and assistant-stream chunks) carry no stable ID, so they
+// can't be deduped between the primary sendStream and the secondary observer.
+// The observer uses this to drop them while the primary is still alive.
+function _isUndedupableLiveEvent(evData) {
+ if (!evData || evData.type !== 'process') return false;
+ const data = evData.data || {};
+ if (data.type === 'delta') return true;
+ if (data.type === 'event' && data.event && data.event.stream === 'assistant') return true;
+ return false;
+}
+
+function _rememberGroupEventIfDuplicate(cid, evData) {
+ const key = _groupEventDedupeKey(evData);
+ if (!cid || !key) return false;
+ let seen = _groupEventDedupe.get(cid);
+ if (!seen) {
+ seen = new Map();
+ _groupEventDedupe.set(cid, seen);
+ }
+ if (seen.has(key)) return true;
+ seen.set(key, Date.now());
+ while (seen.size > GROUP_EVENT_DEDUPE_LIMIT) {
+ const oldest = seen.keys().next().value;
+ if (oldest == null) break;
+ seen.delete(oldest);
+ }
+ return false;
+}
function _stopRuntimeActorRecovery(cid) {
const timer = _runtimeRecoveryTimers.get(cid);
@@ -507,6 +582,13 @@ function _stopRuntimeActorRecovery(cid) {
_runtimeRecoveryTimers.delete(cid);
}
+function _stopGroupEventObserver(cid) {
+ const ctrl = _groupObserverCtrls.get(cid);
+ if (!ctrl) return;
+ _groupObserverCtrls.delete(cid);
+ try { ctrl.abort(); } catch (_) {}
+}
+
async function _syncPendingActorsFromRuntime(cid, opts = {}) {
if (!cid || !isConvPending(cid)) return false;
const state = pendingConvs.get(cid);
@@ -539,9 +621,10 @@ async function _syncPendingActorsFromRuntime(cid, opts = {}) {
} catch (_) {}
}
if (window.ConversationInfo) {
- try { window.ConversationInfo.refresh(cid, { silent: true }); } catch (_) {}
+ try { window.ConversationInfo.refreshTasks(cid, { silent: true }); } catch (_) {}
}
if (!processing) {
+ _stopGroupEventObserver(cid);
if (hasLiveController) {
_latestInFlight.set(cid, []);
setGroupConversationBusy(cid, false);
@@ -567,6 +650,17 @@ async function _syncPendingActorsFromRuntime(cid, opts = {}) {
_updateConvSidebarBadge(cid, true);
startPolling(cid);
if (cid === currentCid) _updateConvSendUI(cid);
+ if (hasLiveController) {
+ // Keep a second, read-only group event stream attached while a plan-driven
+ // run is active. The request-scoped send stream can be interrupted by
+ // renderer lifecycle / abort races; this observer is cheap and idempotent,
+ // and process-event de-dupe below is now render-aware so duplicates are
+ // less dangerous than a missing live rail.
+ _observeConversationRunFromPlanAction(cid, {
+ attachExisting: true,
+ allowWithController: true,
+ });
+ }
if (!hasLiveController) {
// Runtime polling can tell us who is working, but it does not carry
// process/delta events. When the local send-stream was lost (refresh,
@@ -598,7 +692,7 @@ function _startRuntimeActorRecovery(cid) {
_stopRuntimeActorRecovery(cid);
return;
}
- const recovered = await _syncPendingActorsFromRuntime(cid);
+ const recovered = await _syncPendingActorsFromRuntime(cid, { allowController: true });
if (!isConvPending(cid) || pendingConvs.get(cid)?.aborted) {
_stopRuntimeActorRecovery(cid);
return;
@@ -828,9 +922,18 @@ async function _openActorAgentDetail(actorId) {
try {
const res = await apiFetch(`/api/agents/${encodeURIComponent(aid)}`);
const data = await res.json();
- if (!data?.ok || !data?.agent) return;
+ if (!data?.ok || !data?.agent) {
+ // Agent was deleted / renamed / never installed under this uid — the
+ // user clicked the actor header and landed on the Agents tab with no
+ // detail panel. Tell them why instead of silently dropping the click.
+ _convLog.warn('open message actor: agent not found', { agent_id: aid });
+ try { await uiAlert(t('agents.agent_not_found')); } catch (_) {}
+ return;
+ }
} catch (err) {
- _convLog.warn('open message actor failed', { agent_id: aid, error: String(err && err.message || err) });
+ const msg = String(err && err.message || err);
+ _convLog.warn('open message actor failed', { agent_id: aid, error: msg });
+ try { await uiAlert(t('chat.unknown_error') + ': ' + msg); } catch (_) {}
return;
}
if (typeof _showAgentsDetailView === 'function') await _showAgentsDetailView(aid);
@@ -1023,9 +1126,11 @@ function _groupMsgToLegacy(gm) {
// ─── Chat attachments (pending-send pool per cid) ─────────────────────────
// User picks files via "+" → we upload them to `<cid>/` and remember them in
// this Map. On send we hand the filenames to the server; on success the list
-// for that cid is cleared (per-message granularity). Each entry is {name, kind, bytes, dataUrl?}.
+// for that cid is cleared (per-message granularity). Each entry is
+// {name, displayName?, kind, bytes, dataUrl?, sha256?, reused?}. `name` is the
+// real attachment-pool filename; `displayName` is the stable composer label.
-const _chatAttachments = new Map(); // cid → Array<{name, kind, bytes, dataUrl?}>
+const _chatAttachments = new Map(); // cid → Array<{name, displayName?, kind, bytes, dataUrl?, sha256?, reused?}>
// Draft cid used by the commander (new-chat) tab — files land under
// `data/<uid>/chat_attachments/main_chat/` until the user hits send, at which
@@ -1042,6 +1147,7 @@ const CHAT_ATTACH_ACCEPT = [
const CHAT_IMAGE_EXTS = ['.png', '.jpg', '.jpeg', '.webp', '.gif'];
const CHAT_VIDEO_EXTS = ['.mp4', '.webm', '.mov', '.m4v', '.ogv'];
+const ORKAS_FILE_DRAG_MIME = 'application/x-orkas-file';
function _chatFileIconHtml(name, kind) {
if (typeof window !== 'undefined' && typeof window.fileKindIconHtml === 'function') return window.fileKindIconHtml(name, kind);
@@ -1061,6 +1167,61 @@ function _chatAttachKindFromExt(ext) {
return 'text';
}
+function _chatAttachBaseName(p) {
+ const parts = String(p || '').split(/[\\/]/).filter(Boolean);
+ return parts.length ? parts[parts.length - 1] : String(p || '');
+}
+
+function _chatAttachHex(buf) {
+ return Array.from(new Uint8Array(buf))
+ .map((b) => b.toString(16).padStart(2, '0'))
+ .join('');
+}
+
+async function _chatAttachSha256(buf) {
+ const subtle = globalThis.crypto && globalThis.crypto.subtle;
+ if (!subtle || typeof subtle.digest !== 'function') return '';
+ const digest = await subtle.digest('SHA-256', buf);
+ return _chatAttachHex(digest);
+}
+
+async function _chatAttachPrepareUploadFiles(cid, fileList) {
+ const files = Array.from(fileList || []);
+ const rejected = [];
+ const prepared = [];
+ const seenHashes = new Set(
+ _chatAttachList(cid)
+ .map((item) => item && item.sha256 ? `${Number(item.bytes) || 0}:${item.sha256}` : '')
+ .filter(Boolean),
+ );
+
+ for (const file of files) {
+ const ext = _chatAttachExtOf(file.name);
+ if (!CHAT_ATTACH_ACCEPT.includes(ext)) {
+ rejected.push(t('chat.attach_unsupported', { name: file.name }));
+ continue;
+ }
+
+ let buf;
+ let sha256 = '';
+ try {
+ buf = await file.arrayBuffer();
+ } catch (err) {
+ rejected.push(t('chat.attach_upload_fail', { name: file.name, reason: err.message || t('chat.attach_upload_generic_fail') }));
+ continue;
+ }
+ try { sha256 = await _chatAttachSha256(buf); }
+ catch (_) { sha256 = ''; }
+
+ const key = sha256 ? `${buf.byteLength}:${sha256}` : '';
+ if (key && seenHashes.has(key)) continue;
+ if (key) seenHashes.add(key);
+ prepared.push({ file, ext, buf, sha256 });
+ }
+
+ return { prepared, rejected };
+}
+
// Build the `chat-media://` URL for serving an attachment's raw bytes to an
// <img> or <video>. The main-process handler enforces uid + path safety; we
// just URL-encode both path segments here.
@@ -1076,13 +1237,16 @@ function _chatAttachSet(cid, items) {
_chatAttachments.set(cid, items);
_chatAttachRenderChips(cid);
if (cid && cid === currentCid && window.ConversationInfo) {
- window.ConversationInfo.refreshAttachments(cid);
+ window.ConversationInfo.refreshAttachments(cid, { items });
}
}
function _chatAttachClear(cid) {
_chatAttachments.delete(cid);
_chatAttachRenderChips(cid);
+ if (cid && cid === currentCid && window.ConversationInfo) {
+ window.ConversationInfo.refreshAttachments(cid, { items: [] });
+ }
}
// cid → DOM host id for the chip row. Draft cid (commander tab) renders into the
@@ -1109,10 +1273,11 @@ function _chatAttachRenderChips(cid) {
}
host.style.display = '';
host.innerHTML = items.map((it, i) => {
+ const displayName = it.displayName || it.name;
const icon = it.kind === 'image'
- ? (it.dataUrl ? `<img class="chat-attach-thumb" src="${it.dataUrl}" alt="">` : _chatFileIconHtml(it.name, it.kind))
- : _chatFileIconHtml(it.name, it.kind);
- const label = escapeHtml(it.name);
+ ? (it.dataUrl ? `<img class="chat-attach-thumb" src="${it.dataUrl}" alt="">` : _chatFileIconHtml(displayName, it.kind))
+ : _chatFileIconHtml(displayName, it.kind);
+ const label = escapeHtml(displayName);
const busy = it.status === 'uploading';
const errored = it.status === 'error';
const klass = `chat-attach-chip${busy ? ' is-uploading' : ''}${errored ? ' is-error' : ''}`;
@@ -1146,9 +1311,9 @@ async function _chatAttachRemove(cid, idx) {
if (item.dataUrl && item.dataUrl.startsWith('blob:')) {
try { URL.revokeObjectURL(item.dataUrl); } catch (_) { /* ignore */ }
}
- // Only hit the server if the upload actually completed — placeholders
- // that errored out or are still uploading have no disk counterpart yet.
- if (item.status !== 'uploading') {
+ // Only delete files this chip created. Reused files can belong to earlier
+ // messages, so removing the pending chip must leave the original on disk.
+ if (item.status !== 'uploading' && !item.reused) {
try {
await apiFetch(`/api/conversations/${encodeURIComponent(cid)}/attachments?name=${encodeURIComponent(item.name)}`, {
method: 'DELETE',
@@ -1169,23 +1334,36 @@ function _chatAttachReplaceByTempId(cid, tempId, patch) {
// Drop the entry entirely (error path).
items.splice(idx, 1);
} else {
- items[idx] = { ...items[idx], ...patch };
+ const next = { ...items[idx], ...patch };
+ const dupIdx = next.name
+ ? items.findIndex((it, i) => i !== idx && it.name === next.name && it.status !== 'uploading')
+ : -1;
+ if (dupIdx >= 0) {
+ if (items[idx].dataUrl && items[idx].dataUrl.startsWith('blob:')) {
+ try { URL.revokeObjectURL(items[idx].dataUrl); } catch (_) { /* ignore */ }
+ }
+ items.splice(idx, 1);
+ } else {
+ items[idx] = next;
+ }
}
_chatAttachSet(cid, items);
}
async function _chatAttachUpload(cid, fileList) {
- const files = Array.from(fileList || []);
- if (!files.length) return;
+ const { prepared, rejected } = await _chatAttachPrepareUploadFiles(cid, fileList);
+ if (!prepared.length) {
+ if (rejected.length) uiAlert(t('chat.attach_rejected_prefix', { list: rejected.join('\n') }));
+ return;
+ }
- // ── Step 1: show placeholders immediately ────────────────────────────
- // Append a busy chip for each file *before* any network I/O so the user
- // sees the chip appear the instant they pick files. Images get a cheap
- // local preview via URL.createObjectURL — no extra roundtrip needed.
+ // ── Step 1: show placeholders after client-side hash dedupe ───────────
+ // Hashing before painting chips prevents one add/drop action with several
+ // identical files from flashing duplicate thumbnails above the composer.
const placeholders = [];
const current = _chatAttachList(cid).slice();
- for (const file of files) {
- const ext = _chatAttachExtOf(file.name);
+ for (const item of prepared) {
+ const { file, ext, buf, sha256 } = item;
const kind = _chatAttachKindFromExt(ext);
const tempId = `att-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
let localPreview = null;
@@ -1193,31 +1371,24 @@ async function _chatAttachUpload(cid, fileList) {
try { localPreview = URL.createObjectURL(file); } catch (_) { /* ignore */ }
}
const entry = {
- tempId, name: file.name, kind, bytes: file.size || 0,
- dataUrl: localPreview, status: 'uploading',
+ tempId, name: file.name, displayName: file.name, kind, bytes: file.size || 0,
+ dataUrl: localPreview, sha256, status: 'uploading',
};
current.push(entry);
- placeholders.push({ tempId, file, ext });
+ placeholders.push({ tempId, file, ext, buf, sha256 });
}
_chatAttachSet(cid, current);
// ── Step 2: upload all in parallel; rendering is already done ─────────
- const rejected = [];
await Promise.all(placeholders.map(async (ph) => {
- if (!CHAT_ATTACH_ACCEPT.includes(ph.ext)) {
- _chatAttachReplaceByTempId(cid, ph.tempId, null);
- rejected.push(t('chat.attach_unsupported', { name: ph.file.name }));
- return;
- }
try {
- const buf = await ph.file.arrayBuffer();
const res = await apiFetch(`/api/conversations/${encodeURIComponent(cid)}/attachments/upload`, {
method: 'POST',
headers: {
'Content-Type': 'application/octet-stream',
'X-Filename': encodeURIComponent(ph.file.name),
},
- body: buf,
+ body: ph.buf,
});
const data = await res.json();
if (!data.ok) {
@@ -1230,8 +1401,11 @@ async function _chatAttachUpload(cid, fileList) {
// roundtrip to fetch a dataUrl the server would just base64 back.
_chatAttachReplaceByTempId(cid, ph.tempId, {
name: info.name,
+ displayName: ph.file.name,
kind: info.kind,
bytes: info.bytes,
+ reused: !!data.reused,
+ sha256: ph.sha256,
status: 'ready',
});
} catch (err) {
@@ -1245,6 +1419,93 @@ async function _chatAttachUpload(cid, fileList) {
}
}
+function _chatAttachInternalDragItems(dataTransfer) {
+ if (!dataTransfer || !dataTransfer.types) return [];
+ let hasInternal = false;
+ for (let i = 0; i < dataTransfer.types.length; i++) {
+ if (dataTransfer.types[i] === ORKAS_FILE_DRAG_MIME) {
+ hasInternal = true;
+ break;
+ }
+ }
+ if (!hasInternal) return [];
+ let raw = '';
+ try { raw = dataTransfer.getData(ORKAS_FILE_DRAG_MIME); }
+ catch (_) { return []; }
+ if (!raw) return [];
+ try {
+ const parsed = JSON.parse(raw);
+ const list = Array.isArray(parsed) ? parsed : [parsed];
+ return list
+ .map((item) => ({
+ path: typeof item?.path === 'string' ? item.path : '',
+ name: typeof item?.name === 'string' ? item.name : '',
+ }))
+ .filter((item) => item.path);
+ } catch (_) {
+ return [];
+ }
+}
+
+async function _chatAttachImportPaths(cid, entries) {
+ const files = Array.isArray(entries) ? entries.filter((it) => it && it.path) : [];
+ if (!files.length) return;
+
+ const placeholders = [];
+ const current = _chatAttachList(cid).slice();
+ for (const item of files) {
+ const displayName = item.name || _chatAttachBaseName(item.path);
+ const ext = _chatAttachExtOf(displayName);
+ const kind = _chatAttachKindFromExt(ext);
+ const tempId = `att-${Date.now()}-${Math.random().toString(36).slice(2, 8)}`;
+ current.push({
+ tempId,
+ name: displayName,
+ displayName,
+ kind,
+ bytes: 0,
+ dataUrl: null,
+ status: 'uploading',
+ });
+ placeholders.push({ tempId, path: item.path, name: displayName });
+ }
+ _chatAttachSet(cid, current);
+
+ const rejected = [];
+ await Promise.all(placeholders.map(async (ph) => {
+ try {
+ const res = await apiFetch(`/api/conversations/${encodeURIComponent(cid)}/attachments/import`, {
+ method: 'POST',
+ headers: { 'Content-Type': 'application/json' },
+ body: JSON.stringify({ path: ph.path, name: ph.name }),
+ });
+ const data = await res.json();
+ if (!data.ok) {
+ _chatAttachReplaceByTempId(cid, ph.tempId, null);
+ rejected.push(t('chat.attach_upload_fail', { name: ph.name, reason: data.error || t('chat.attach_upload_generic_fail') }));
+ return;
+ }
+ const info = data.info;
+ _chatAttachReplaceByTempId(cid, ph.tempId, {
+ name: info.name,
+ displayName: ph.name,
+ kind: info.kind,
+ bytes: info.bytes,
+ dataUrl: (info.kind === 'image' || info.kind === 'video') ? _chatMediaUrl(cid, info.name) : null,
+ reused: !!data.reused,
+ status: 'ready',
+ });
+ } catch (err) {
+ _chatAttachReplaceByTempId(cid, ph.tempId, null);
+ rejected.push(t('chat.attach_upload_fail', { name: ph.name, reason: err.message || t('chat.attach_upload_generic_fail') }));
+ }
+ }));
+
+ if (rejected.length) {
+ uiAlert(t('chat.attach_rejected_prefix', { list: rejected.join('\n') }));
+ }
+}
+
async function _chatAttachRefreshFromServer(cid) {
try {
const res = await apiFetch(`/api/conversations/${encodeURIComponent(cid)}/attachments`);
@@ -1409,32 +1670,14 @@ function _hydrateMessageProducedChips(msgDiv) {
e.stopPropagation();
const p = chip.dataset.producedPath;
if (!p) return;
- // Markdown lands in the right-side drawer with view/edit semantics
- // (reuses the KB editor — see modules/md-view-edit.js). Other kinds
- // continue to use the existing in-app preview overlay / fallback.
- if (_isMarkdownPath(p) && typeof openChatMdDrawer === 'function') {
- const base = p.split(/[\\/]/).pop() || p;
- openChatMdDrawer({
- source: { kind: 'workspace', absPath: p, cid: currentCid || undefined },
- initialMode: 'view',
- title: base,
- });
- return;
- }
if (typeof openChatFileViewer === 'function') {
const base = p.split(/[\\/]/).pop() || p;
- openChatFileViewer(p, base);
+ openChatFileViewer(p, base, currentCid ? { cid: currentCid } : undefined);
}
});
});
}
-function _isMarkdownPath(p) {
- if (!p) return false;
- const lower = p.toLowerCase();
- return lower.endsWith('.md') || lower.endsWith('.markdown');
-}
-
function _hydrateMessageAttachmentThumbs(msgDiv, cid) {
// Image chips have a thumb we want to enlarge via the lightbox; the rest
// (pdf / docx / text / video) get the same kind-aware viewer as produced
@@ -1512,26 +1755,31 @@ function _bindChatPasteAttach(inputSelector, getCid) {
function _bindChatDropAttach(wrapSelector, getCid) {
const el = document.querySelector(wrapSelector);
if (!el || el.dataset.dropBound === '1') return;
- const isFileDrag = (e) => {
+ const isAttachDrag = (e) => {
const types = e.dataTransfer && e.dataTransfer.types;
if (!types) return false;
for (let i = 0; i < types.length; i++) {
- if (types[i] === 'Files') return true;
+ if (types[i] === 'Files' || types[i] === ORKAS_FILE_DRAG_MIME) return true;
}
return false;
};
const allow = (e) => {
- if (!isFileDrag(e)) return;
+ if (!isAttachDrag(e)) return;
e.preventDefault();
e.dataTransfer.dropEffect = 'copy';
};
el.addEventListener('dragover', allow);
el.addEventListener('dragenter', allow);
el.addEventListener('drop', (e) => {
- if (!isFileDrag(e)) return;
+ if (!isAttachDrag(e)) return;
e.preventDefault();
const cid = getCid();
if (!cid) return;
+ const internalFiles = _chatAttachInternalDragItems(e.dataTransfer);
+ if (internalFiles.length) {
+ _chatAttachImportPaths(cid, internalFiles);
+ return;
+ }
_chatAttachUpload(cid, e.dataTransfer.files);
});
el.dataset.dropBound = '1';
@@ -1703,6 +1951,10 @@ async function _toggleConversationPinned(cid, pinned) {
conversations = snapshot;
renderConversationList();
_convLog.warn('toggle conversation pin failed', err);
+ // Surface failure to the user — silent rollback leaves a chip-flicker
+ // with no explanation (typical race: another tab/device deleted the
+ // conv while this pin was in flight, or the server transient-errored).
+ try { await uiAlert(t('chat.pin_failed')); } catch (_) {}
}
}
@@ -2008,10 +2260,10 @@ async function loadConversationHistory(cid) {
// adds it post-scroll and it ends up below the visible area.
_ensureConvCreateAgentInline();
_scrollToBottomNoAnim(container);
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid);
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid);
} catch (e) {
container.innerHTML = `<div class="empty">${escapeHtml(t('chat.load_failed', { msg: e.message || '' }))}</div>`;
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid);
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid);
}
}
@@ -2083,7 +2335,7 @@ async function _recoverPolledVisibleMessages(cid, rawMessages) {
}
if (changed) {
try { if (window.PlanRail) window.PlanRail.refresh(cid, { force: true }); } catch (_) {}
- try { if (window.ConversationInfo) window.ConversationInfo.refresh(cid); } catch (_) {}
+ try { if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid); } catch (_) {}
}
return changed;
}
@@ -3063,9 +3315,12 @@ function _makeConvChatController(cid) {
controller: { abort: () => _convChatCtrls.get(id)?.abort() },
aborted: false,
});
+ _lastGroupWorkEventAt.set(id, Date.now());
+ _groupEventDedupe.delete(id);
_updateConvSendUI(id);
_updateConvSidebarBadge(id, true);
startPolling(id);
+ _startRuntimeActorRecovery(id);
},
onAbort(_msgEl, id) {
const state = pendingConvs.get(id);
@@ -3181,7 +3436,9 @@ function _makeStreamPaintYield() {
function _observeConversationRunFromPlanAction(cid, opts = {}) {
if (!cid) return null;
const attachExisting = !!opts.attachExisting;
- if (_convChatCtrls.has(cid)) return null;
+ const allowWithController = !!opts.allowWithController;
+ if (_convChatCtrls.has(cid) && !allowWithController) return null;
+ if (allowWithController && _groupObserverCtrls.has(cid)) return null;
if (!attachExisting && (pendingConvs.has(cid) || isGroupConversationBusy(cid))) return null;
const controller = new AbortController();
@@ -3195,6 +3452,7 @@ function _observeConversationRunFromPlanAction(cid, opts = {}) {
try { controller.abort(); } catch (_) {}
},
};
+ if (allowWithController) _groupObserverCtrls.set(cid, ctrl);
const activate = () => {
if (activated || controller.signal.aborted) return;
@@ -3205,12 +3463,14 @@ function _observeConversationRunFromPlanAction(cid, opts = {}) {
msgEl = existing.loadingEl || (cid === currentCid
? _createStreamingAssistantMessage(container, { hiddenUntilActor: true })
: null);
- _convChatCtrls.set(cid, ctrl);
+ if (!allowWithController) {
+ _convChatCtrls.set(cid, ctrl);
+ }
pendingConvs.set(cid, {
...existing,
loadingEl: msgEl,
needsIndicator: false,
- controller: ctrl,
+ controller: allowWithController ? (existing.controller || ctrl) : ctrl,
aborted: false,
});
setGroupConversationBusy(cid, true);
@@ -3266,6 +3526,17 @@ function _observeConversationRunFromPlanAction(cid, opts = {}) {
if (st.status === 'running' || inFlight.length > 0) sawActivity = true;
}
if (sawActivity) activate();
+ // While the primary IPC sendStream still holds the live source of
+ // truth for rendering, skip events the dedupe layer can't
+ // disambiguate. `_groupEventDedupeKey` returns empty for delta
+ // tokens and assistant-stream events (no stable ID), so feeding
+ // them from both subscribers doubles every appended chunk into
+ // the streaming bubble. The observer remains active for ID-bearing
+ // events so the resilience story (primary dies mid-turn → bubble
+ // still receives message/artifact/state_changed) is preserved.
+ if (allowWithController && _convChatCtrls.has(cid) && _isUndedupableLiveEvent(evData)) {
+ continue;
+ }
_handleGroupBusEvent(cid, msgEl, evData, { archive: true });
const paintWait = maybeYieldToPaint();
if (paintWait) await paintWait;
@@ -3281,14 +3552,23 @@ function _observeConversationRunFromPlanAction(cid, opts = {}) {
} finally {
settled = true;
clearTimeout(noActivityTimer);
- if (_convChatCtrls.get(cid) === ctrl) _convChatCtrls.delete(cid);
+ if (allowWithController) {
+ if (_groupObserverCtrls.get(cid) === ctrl) _groupObserverCtrls.delete(cid);
+ } else if (_convChatCtrls.get(cid) === ctrl) {
+ _convChatCtrls.delete(cid);
+ }
if (activated) {
+ if (allowWithController) {
+ if (window.PlanRail) window.PlanRail.refresh(cid, { force: true });
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid, { silent: true });
+ return;
+ }
setGroupConversationBusy(cid, false);
_removeEmptyStreamingPlaceholder(msgEl);
_finishStreamingMsg(cid);
_scheduleHistoryReconcileAfterStream(cid);
if (window.PlanRail) window.PlanRail.refresh(cid, { force: true });
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid, { silent: true });
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid, { silent: true });
}
}
})();
@@ -3511,6 +3791,10 @@ function _setProcessLineContent(line, text, kind) {
line.innerHTML = `${icon}<span class="stream-process-text">${escapeHtml(body)}</span>`;
}
+function _processLineCount(msg) {
+ return msg?.querySelectorAll?.('[data-role="process"] .stream-process-line')?.length || 0;
+}
+
function _eventProcessKind(evt, text) {
if (!evt || typeof evt !== 'object') return _processKindOf(text);
const stream = evt.stream;
@@ -3730,6 +4014,9 @@ function _streamingMarkAborted(msg) {
function _finishStreamingMsg(cid) {
_stopRuntimeActorRecovery(cid);
+ _stopGroupEventObserver(cid);
+ _lastGroupWorkEventAt.delete(cid);
+ _groupEventDedupe.delete(cid);
pendingConvs.delete(cid);
if (isGroupConversationBusy(cid)) startPolling(cid);
else stopPolling(cid);
@@ -4179,7 +4466,12 @@ function createChatController(config) {
}
const paintWait = maybeYieldToPaint();
if (paintWait) await paintWait;
- } catch (_) { /* skip malformed */ }
+ } catch (err) {
+ _convLog.warn('stream event handler failed', {
+ cid: id,
+ error: String(err && err.message || err),
+ });
+ }
}
}
} catch (err) {
@@ -4583,6 +4875,13 @@ function _finalizeActorPlaceholder(ph, gm, cid, archive) {
// { type: 'aborted', cid }
function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {}) {
if (!evData || typeof evData !== 'object') return;
+ if (
+ evData.type === 'process'
+ || evData.type === 'artifact_created'
+ || (evData.type === 'message' && evData.msg && evData.msg.from !== 'user')
+ ) {
+ _lastGroupWorkEventAt.set(cid, Date.now());
+ }
// Bump the conv to the top of the sidebar list whenever a user-visible
// message lands — applies to both currently-viewed and background convs,
// so the sidebar stays ordered by last activity in real time. Skip
@@ -4590,7 +4889,7 @@ function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {
// user's view; visible end-of-turn replies will bump shortly after).
if (evData.type === 'message' && evData.msg && !evData.msg.dispatch) {
_bumpConvToTop(cid);
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid);
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid);
}
// Cross-cid leakage guard: per-cid controllers stay alive when the user
// navigates away mid-stream (a legit pattern — let the conv finish in
@@ -4605,6 +4904,7 @@ function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {
// and on switch-back `loadConversationHistory` rebuilds the cid's view
// from disk — so dropping live UI updates here is safe.
if (cid !== currentCid) return;
+ if (evData.type !== 'process' && _rememberGroupEventIfDuplicate(cid, evData)) return;
if (evData.type === 'message') {
const gm = evData.msg;
if (!gm) return;
@@ -4705,7 +5005,15 @@ function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {
if (cid === currentCid) _updateConvSendUI(cid);
}
const target = _ensureActorPlaceholder(cid, actor, streamingMsg);
- if (!target) return;
+ if (!target) {
+ _convLog.warn('group process target missing', {
+ cid,
+ actor,
+ kind: data.type || '',
+ });
+ return;
+ }
+ if (_rememberGroupEventIfDuplicate(cid, evData)) return;
// Diagnostic — count how many deltas reach the renderer per actor.
// If this number is much smaller than the bus emit count from the
// turn-end log, the bottleneck is upstream (IPC batching). If it's
@@ -4716,14 +5024,37 @@ function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {
window._convDeltaCount[actor] = (window._convDeltaCount[actor] || 0) + 1;
}
}
- if (data.type === 'delta' && typeof data.text === 'string') {
- // Token-by-token streaming → write into the placeholder's final
- // body so the user sees the reply form character-by-character.
- _streamingAppendFinalDelta(target, data.text);
- } else if (data.type === 'progress' && data.text) {
- _streamingAppendProgress(target, String(data.text));
- } else if (data.type === 'event') {
- _renderAgentEvent(target, data.event);
+ try {
+ if (data.type === 'delta' && typeof data.text === 'string') {
+ // Token-by-token streaming → write into the placeholder's final
+ // body so the user sees the reply form character-by-character.
+ _streamingAppendFinalDelta(target, data.text);
+ } else if (data.type === 'progress' && data.text) {
+ _streamingAppendProgress(target, String(data.text));
+ } else if (data.type === 'event') {
+ const before = _processLineCount(target);
+ _renderAgentEvent(target, data.event);
+ const evt = data.event || {};
+ const line = evt.stream === 'tool' ? _formatEventLine(evt) : null;
+ if (line && _processLineCount(target) <= before) {
+ _streamingAppendProgress(target, line, _eventProcessKind(evt, line));
+ }
+ }
+ } catch (err) {
+ const evt = data.event || {};
+ const fallback = data.type === 'progress'
+ ? String(data.text || '')
+ : (evt && evt.stream ? _formatEventLine(evt) : '');
+ _convLog.warn('group process render failed', {
+ cid,
+ actor,
+ kind: data.type || '',
+ stream: evt.stream || '',
+ error: String(err && err.message || err),
+ });
+ if (fallback) {
+ try { _streamingAppendProgress(target, fallback, _eventProcessKind(evt, fallback)); } catch (_) {}
+ }
}
} else if (evData.type === 'artifact_created') {
const actor = String(evData.actor || evData.artifact?.agent_id || '');
@@ -4747,7 +5078,7 @@ function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {
// reply should default to that agent without them having to @-mention.
_evaluateAutoRecipient(cid);
if (window.PlanRail) window.PlanRail.refresh(cid, { force: true });
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid, { silent: true });
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid, { silent: true });
} else if (evData.type === 'state_changed') {
// Each in_flight actor gets a placeholder so its delta tokens / tool
// calls render in its own bubble even before its `message` arrives.
@@ -4774,7 +5105,7 @@ function _handleGroupBusEvent(cid, streamingMsg, evData, { archive = false } = {
if (window.PlanRail) {
window.PlanRail.setInFlight(cid, inFlight);
}
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid, { silent: true });
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid, { silent: true });
_updateConvSidebarBadge(cid, false);
if (cid === currentCid) _updateConvSendUI(cid);
} else if (evData.type === 'aborted') {
@@ -4934,19 +5265,18 @@ function _stripAgentCreateBlocksForStream(buf) {
return _replaceOuterAgentBlocks(buf, _streamPlaceholderHtml('chat.create_agent_streaming_placeholder'));
}
+function _skillStreamPlaceholderHtml() {
+ return _streamPlaceholderHtml('chat.create_skill_streaming_placeholder');
+}
+
// `<<<skill-file path=X ... >>>` blocks (skill edit chat). Different fence
// shape from `<agent>` (see strip-structural-blocks.js header) but same user-facing
-// contract: streaming placeholder hides the raw block + reveals the file
-// being written. The path attribute (when present) flows into the
-// localised "Writing X…" label so the placeholder is informative;
-// unattributed blocks fall back to a generic "Writing file…" label.
+// contract: streaming placeholder hides raw file content while the skill is
+// being assembled. Do not surface per-file paths here: one skill can emit
+// several file blocks, and repeating "Writing SKILL.md" made the stream noisy.
function _stripSkillFileBlocksForStream(buf) {
- return _replaceOuterSkillFileBlocks(buf, (path) => {
- const label = path
- ? t('chat.skill_file_streaming_placeholder', { path })
- : t('chat.skill_file_streaming_placeholder_unknown');
- return `\n<em class="stream-placeholder">${escapeHtml(label)}</em>\n`;
- });
+ const placeholder = _skillStreamPlaceholderHtml();
+ return _replaceOuterSkillFileBlocks(buf, () => placeholder);
}
// `<skill>` container (commander create / edit). Pure logic lives in
@@ -4957,10 +5287,11 @@ function _stripSkillFileBlocksForStream(buf) {
// composes `_replaceOuterAgentBlocks`). Set A / set B fixtures pinned in
// `test/renderer/strip-structural-blocks.test.ts`.
function _stripSkillCreateBlocksForStream(buf) {
- return _stripSkillCreateContainer(
+ const placeholder = _skillStreamPlaceholderHtml();
+ return _collapseRepeatedStructuralPlaceholders(_stripSkillCreateContainer(
buf,
- _streamPlaceholderHtml('chat.create_skill_streaming_placeholder'),
- );
+ placeholder,
+ ), placeholder);
}
function _stripAgentFormBlockForStream(buf) {
@@ -5608,13 +5939,14 @@ function _refreshTaskSurfacesAfterAbort(cid) {
}
} catch (_) {}
try {
- if (window.ConversationInfo) window.ConversationInfo.refresh(cid, { silent: true });
+ if (window.ConversationInfo) window.ConversationInfo.refreshTasks(cid, { silent: true });
} catch (_) {}
}
function abortConvStream(cid) {
const state = pendingConvs.get(cid);
_stopRuntimeActorRecovery(cid);
+ _stopGroupEventObserver(cid);
setGroupConversationBusy(cid, false);
// Group chat: also tell the bus to abort all in-flight worker turns + clear
// queues. Cancelling just the IPC stream would leave agents running in the
@@ -5767,4 +6099,3 @@ if (typeof window !== 'undefined') {
_initChatSelectionMenu();
}
}
-
diff --git a/src/renderer/modules/ipc-shim.js b/src/renderer/modules/ipc-shim.js
index c674a18f..1de35989 100644
--- a/src/renderer/modules/ipc-shim.js
+++ b/src/renderer/modules/ipc-shim.js
@@ -61,6 +61,7 @@ const _IPC_ROUTES = [
['DELETE', /^\/api\/conversations\/([^/]+)$/, 'conversations.delete', ['cid']],
['POST', /^\/api\/conversations\/([^/]+)\/pin$/, 'conversations.pin', ['cid']],
['GET', /^\/api\/conversations\/([^/]+)\/history$/, 'conversations.history', ['cid']],
+ ['GET', /^\/api\/conversations\/([^/]+)\/files$/, 'conversations.files.list', ['cid']],
['POST', /^\/api\/conversations\/([^/]+)\/send$/, 'groupChat.send', ['cid']],
['POST', /^\/api\/conversations\/([^/]+)\/send\/stream$/, 'conversations.sendStream', ['cid'], { stream: true }],
['POST', /^\/api\/conversations\/([^/]+)\/abort$/, 'groupChat.abort', ['cid']],
@@ -74,6 +75,7 @@ const _IPC_ROUTES = [
['POST', /^\/api\/conversations\/([^/]+)\/marketplace-install$/, 'groupChat.resolveMarketplaceInstallRequest', ['cid']],
['GET', /^\/api\/conversations\/([^/]+)\/attachments$/, 'conversations.attachments.list', ['cid']],
['POST', /^\/api\/conversations\/([^/]+)\/attachments\/upload$/, { upload: 'conversations.attachments.upload' }, ['cid']],
+ ['POST', /^\/api\/conversations\/([^/]+)\/attachments\/import$/, 'conversations.attachments.import', ['cid']],
['DELETE', /^\/api\/conversations\/([^/]+)\/attachments$/, 'conversations.attachments.delete', ['cid']],
['GET', /^\/api\/agents\/([^/]+)$/, 'agents.get', ['agent_id']],
['DELETE', /^\/api\/agents\/([^/]+)$/, 'agents.delete', ['agent_id']],
diff --git a/src/renderer/modules/md-view-edit.js b/src/renderer/modules/md-view-edit.js
index eb5a6fc6..3539a337 100644
--- a/src/renderer/modules/md-view-edit.js
+++ b/src/renderer/modules/md-view-edit.js
@@ -15,7 +15,7 @@
//
// API:
// mountMdViewEdit({ bodyEl, actionsEl, source, capabilities?, initialMode?,
-// initialContent?, initialDraft?, callbacks? })
+// initialContent?, initialDraft?, actionIconOnly?, callbacks? })
// → { destroy, refreshContent, getMode, setMode, isDirty,
// getDraft, setDraftAsContent, getSource }
//
@@ -103,6 +103,7 @@ function mountMdViewEdit(opts) {
callbacks,
bodyEl,
actionsEl,
+ actionIconOnly: opts.actionIconOnly === true,
// Canonical content (from disk / source). Ephemeral seeds this with
// initialText; the "is dirty" check compares draft vs. this.
content: '',
@@ -288,6 +289,18 @@ function _mveRenderError(state, msg) {
state.actionsEl.innerHTML = '';
}
+function _mveActionButton(state, action, labelKey, iconName, extraClass) {
+ const label = t(labelKey);
+ const cls = `btn btn-sm${state.actionIconOnly ? ' btn-icon ctx-viewer-action-icon-btn' : ''}${extraClass ? ` ${extraClass}` : ''}`;
+ if (!state.actionIconOnly) {
+ return `<button class="${cls}" data-mve-action="${action}">${escapeHtml(label)}</button>`;
+ }
+ const icon = (typeof window !== 'undefined' && typeof window.uiIconHtml === 'function')
+ ? window.uiIconHtml(iconName, 'ui-icon ctx-viewer-action-icon')
+ : escapeHtml(label);
+ return `<button type="button" class="${cls}" data-mve-action="${action}" aria-label="${escapeHtml(label)}" title="${escapeHtml(label)}">${icon}</button>`;
+}
+
function _mveRenderView(state) {
state.mode = 'view';
const bodyEl = state.bodyEl;
@@ -297,9 +310,9 @@ function _mveRenderView(state) {
if (state.caps.taskCheckbox) _mveBindTaskCheckboxes(state, bodyEl);
const actions = [];
- if (state.caps.edit) actions.push(`<button class="btn btn-sm" data-mve-action="edit">${escapeHtml(t('contexts.viewer.edit'))}</button>`);
- if (state.caps.reveal) actions.push(`<button class="btn btn-sm" data-mve-action="reveal">${escapeHtml(t('contexts.viewer.open_system'))}</button>`);
- if (state.caps.delete) actions.push(`<button class="btn btn-sm btn-danger" data-mve-action="delete">${escapeHtml(t('contexts.viewer.delete'))}</button>`);
+ if (state.caps.edit) actions.push(_mveActionButton(state, 'edit', 'contexts.viewer.edit', 'pencil'));
+ if (state.caps.reveal) actions.push(_mveActionButton(state, 'reveal', 'contexts.viewer.open_system', 'folder-open'));
+ if (state.caps.delete) actions.push(_mveActionButton(state, 'delete', 'contexts.viewer.delete', 'trash', 'btn-danger'));
actionsEl.innerHTML = actions.join('');
actionsEl.querySelector('[data-mve-action="edit"]')?.addEventListener('click', () => _mveEnterEdit(state));
@@ -350,11 +363,11 @@ function _mveRenderEditor(state) {
// Actions: save/cancel for persisted sources; copy/sendToChat for ephemeral.
const actions = [];
if (state.caps.save) {
- actions.push(`<button class="btn btn-sm btn-primary" data-mve-action="save">${escapeHtml(t('contexts.viewer.save'))}</button>`);
- actions.push(`<button class="btn btn-sm" data-mve-action="cancel">${escapeHtml(t('contexts.viewer.cancel'))}</button>`);
+ actions.push(_mveActionButton(state, 'save', 'contexts.viewer.save', 'check', 'btn-primary'));
+ actions.push(_mveActionButton(state, 'cancel', 'contexts.viewer.cancel', 'x'));
} else if (state.source.kind === 'ephemeral') {
- actions.push(`<button class="btn btn-sm" data-mve-action="copy-draft">${escapeHtml(t('chat.md_drawer.copy_draft'))}</button>`);
- actions.push(`<button class="btn btn-sm btn-primary" data-mve-action="send-to-chat">${escapeHtml(t('chat.md_drawer.send_to_chat'))}</button>`);
+ actions.push(_mveActionButton(state, 'copy-draft', 'chat.md_drawer.copy_draft', 'clipboard-list'));
+ actions.push(_mveActionButton(state, 'send-to-chat', 'chat.md_drawer.send_to_chat', 'send', 'btn-primary'));
}
actionsEl.innerHTML = actions.join('');
diff --git a/src/renderer/modules/plan-rail.js b/src/renderer/modules/plan-rail.js
index ef28cfa3..c921d7ba 100644
--- a/src/renderer/modules/plan-rail.js
+++ b/src/renderer/modules/plan-rail.js
@@ -135,8 +135,8 @@ async function _finishStepAction(cid, stepIndex, opts = {}) {
}
function _refreshConversationInfo(cid) {
- if (window.ConversationInfo && typeof window.ConversationInfo.refresh === 'function') {
- window.ConversationInfo.refresh(cid, { silent: true });
+ if (window.ConversationInfo && typeof window.ConversationInfo.refreshTasks === 'function') {
+ window.ConversationInfo.refreshTasks(cid, { silent: true });
}
}
@@ -222,6 +222,20 @@ function _buildProgressText(plan) {
return `${done}/${plan.steps.length}`;
}
+function _isPlanFullyComplete(plan) {
+ const steps = plan && Array.isArray(plan.steps) ? plan.steps : [];
+ if (!steps.length) return false;
+ if (Array.isArray(_currentInFlight) && _currentInFlight.length > 0) return false;
+ return steps.every((s) => s && (s.status === 'done' || s.status === 'skipped'));
+}
+
+function _hideRail(root, body, progress, wrap) {
+ root.style.display = 'none';
+ body.innerHTML = '';
+ progress.textContent = '';
+ if (wrap) wrap.classList.remove('has-plan-rail');
+}
+
function _render(plan) {
const root = document.getElementById('plan-rail');
const body = document.getElementById('plan-rail-body');
@@ -235,10 +249,13 @@ function _render(plan) {
const wrap = root.closest('.chat-input-wrapper');
if (!plan || !Array.isArray(plan.steps) || plan.steps.length === 0) {
_lastPlan = null;
- root.style.display = 'none';
- body.innerHTML = '';
- progress.textContent = '';
- if (wrap) wrap.classList.remove('has-plan-rail');
+ _hideRail(root, body, progress, wrap);
+ return;
+ }
+ if (_isPlanFullyComplete(plan)) {
+ _lastPlan = plan;
+ _clearStepActionRequests(_currentCid);
+ _hideRail(root, body, progress, wrap);
return;
}
_lastPlan = plan;
diff --git a/src/renderer/modules/project-detail.js b/src/renderer/modules/project-detail.js
index b6f4af97..e9099203 100644
--- a/src/renderer/modules/project-detail.js
+++ b/src/renderer/modules/project-detail.js
@@ -1,8 +1,6 @@
// ─── Project detail panel ───────────────────────────────────────────────
-// Drives `#panel-project`: shows the project's bindings (agents + skills)
-// and exposes Add / Remove for each. Opened by clicking a project row in
-// the sidebar (`projects.js::_renderProjectRow` — single click both
-// expands the row and switches view here).
+// Drives `#panel-project`: shows the project composer, files, and bindings
+// (agents + skills), and exposes Add / Remove for each.
//
// Header carries 3 actions: + New chat / Rename / Delete. The "Back" button
// was intentionally removed — the sidebar IS the back navigation.
@@ -13,7 +11,7 @@
const _projectDetailLog = createLogger('project-detail');
let _projectDetailPid = ''; // pid currently rendered in the panel
-let _projectDetailMeta = null; // { project, agentDetails, skillDetails }
+let _projectDetailMeta = null; // { project, agentDetails, skillDetails, files }
// ── Public: navigate-into ──────────────────────────────────────────────
@@ -25,17 +23,19 @@ async function loadProjectDetail(pid) {
return;
}
try {
- const [getRes, listRes] = await Promise.all([
+ const [getRes, listRes, filesRes] = await Promise.all([
window.orkas.invoke('projects.get', { projectId: pid }),
window.orkas.invoke('projects.bindings.list', { projectId: pid }),
+ window.orkas.invoke('projects.files.list', { projectId: pid }),
]);
- if (!getRes?.ok || !listRes?.ok) {
- throw new Error((getRes && getRes.error) || (listRes && listRes.error) || 'load_failed');
+ if (!getRes?.ok || !listRes?.ok || !filesRes?.ok) {
+ throw new Error((getRes && getRes.error) || (listRes && listRes.error) || (filesRes && filesRes.error) || 'load_failed');
}
_projectDetailMeta = {
project: getRes.project,
agentDetails: Array.isArray(listRes.agentDetails) ? listRes.agentDetails : [],
skillDetails: Array.isArray(listRes.skillDetails) ? listRes.skillDetails : [],
+ files: Array.isArray(filesRes.files) ? filesRes.files : [],
};
_renderProjectDetail();
} catch (err) {
@@ -48,13 +48,17 @@ async function loadProjectDetail(pid) {
function _renderProjectDetailEmpty() {
const titleEl = document.getElementById('project-detail-title');
if (titleEl) titleEl.textContent = '';
- document.getElementById('project-agents-list').innerHTML = '';
- document.getElementById('project-skills-list').innerHTML = '';
+ const agents = document.getElementById('project-agents-list');
+ const skills = document.getElementById('project-skills-list');
+ const files = document.getElementById('project-files-list');
+ if (agents) agents.innerHTML = '';
+ if (skills) skills.innerHTML = '';
+ if (files) files.innerHTML = '';
}
function _renderProjectDetail() {
if (!_projectDetailMeta) { _renderProjectDetailEmpty(); return; }
- const { project, agentDetails, skillDetails } = _projectDetailMeta;
+ const { project, agentDetails, skillDetails, files } = _projectDetailMeta;
const titleEl = document.getElementById('project-detail-title');
if (titleEl) titleEl.textContent = project?.name || '';
@@ -65,9 +69,13 @@ function _renderProjectDetail() {
document.getElementById('project-skills-list').innerHTML = _renderBindingsRows(
'skill', skillDetails,
);
+ _renderProjectFiles(files || []);
applyDomI18n();
_bindRemoveButtons();
+ _bindProjectFileRows();
+ if (typeof refreshWorkspaceChip === 'function') refreshWorkspaceChip();
+ if (typeof hydrateUiIcons === 'function') hydrateUiIcons(document.getElementById('project-detail-content'));
}
function _renderBindingsRows(kind, items) {
@@ -78,9 +86,9 @@ function _renderBindingsRows(kind, items) {
for (const it of sorted) {
const id = (kind === 'agent') ? it.agent_id : it.id;
const name = escapeHtml(it.name || id);
- const desc = _pickItemDescription(it, lang);
+ const desc = kind === 'agent' ? '' : _pickItemDescription(it, lang);
const descHtml = desc ? `<div class="project-binding-desc muted">${escapeHtml(desc)}</div>` : '';
- const sourceTag = it.source ? `<span class="project-binding-source">${escapeHtml(it.source)}</span>` : '';
+ const sourceTag = _projectBindingSourceHtml(it.source, kind);
rows.push(`
<div class="project-binding-row" data-kind="${kind}" data-id="${escapeHtml(id)}">
<div class="project-binding-main">
@@ -101,6 +109,204 @@ function _renderBindingsRows(kind, items) {
return rows.join('');
}
+function _formatProjectFileBytes(bytes) {
+ const n = Number(bytes);
+ if (!Number.isFinite(n) || n <= 0) return '';
+ if (n < 1024) return `${n} B`;
+ if (n < 1024 * 1024) return `${(n / 1024).toFixed(n < 10 * 1024 ? 1 : 0)} KB`;
+ return `${(n / 1024 / 1024).toFixed(n < 10 * 1024 * 1024 ? 1 : 0)} MB`;
+}
+
+function _renderProjectFiles(files) {
+ const list = document.getElementById('project-files-list');
+ const status = document.getElementById('project-files-status');
+ if (!list) return;
+ if (status) status.textContent = '';
+ const sorted = (files || []).slice().sort(_byDisplayName);
+ if (!sorted.length) {
+ list.innerHTML = `<div class="empty" data-i18n="project.files.empty">${escapeHtml(t('project.files.empty'))}</div>`;
+ return;
+ }
+ const removeLabel = escapeHtml(t('project.files.remove'));
+ list.innerHTML = sorted.map((f) => {
+ const name = f.name || '';
+ const label = escapeHtml(name);
+ const icon = (typeof window !== 'undefined' && typeof window.fileKindIconHtml === 'function')
+ ? window.fileKindIconHtml(name, f.kind)
+ : '';
+ const meta = _formatProjectFileBytes(f.bytes);
+ return `
+ <div class="project-file-row" data-project-file="${label}">
+ <button type="button" class="project-file-main" title="${label}" data-action="open-file">
+ <span class="project-file-icon">${icon}</span>
+ <span class="project-file-name">${label}</span>
+ ${meta ? `<span class="project-file-meta">${escapeHtml(meta)}</span>` : ''}
+ </button>
+ <button type="button" class="project-file-remove" data-action="remove-file"
+ title="${removeLabel}" aria-label="${removeLabel}">×</button>
+ </div>
+ `;
+ }).join('');
+}
+
+function _bindProjectFileRows() {
+ const root = document.getElementById('project-files-list');
+ if (!root) return;
+ root.querySelectorAll('.project-file-row[data-project-file]').forEach((row) => {
+ const name = row.dataset.projectFile || '';
+ row.querySelector('[data-action="open-file"]')?.addEventListener('click', async (e) => {
+ e.stopPropagation();
+ await _openProjectFile(name);
+ });
+ row.querySelector('[data-action="remove-file"]')?.addEventListener('click', async (e) => {
+ e.stopPropagation();
+ await _deleteProjectFile(name);
+ });
+ });
+}
+
+async function _openProjectFile(name) {
+ if (!_projectDetailPid || !name || typeof openChatFileViewer !== 'function') return;
+ try {
+ const res = await window.orkas.invoke('projects.files.absPath', {
+ projectId: _projectDetailPid,
+ name,
+ });
+ if (!res?.ok || !res.path) throw new Error(res?.error || 'not_found');
+ openChatFileViewer(res.path, name, { projectId: _projectDetailPid });
+ } catch (err) {
+ _projectDetailLog.warn('open project file failed', err);
+ }
+}
+
+async function _deleteProjectFile(name) {
+ if (!_projectDetailPid || !name) return;
+ try {
+ const res = await window.orkas.invoke('projects.files.delete', {
+ projectId: _projectDetailPid,
+ name,
+ });
+ if (!res?.ok) throw new Error(res?.error || 'delete_failed');
+ await loadProjectDetail(_projectDetailPid);
+ } catch (err) {
+ _projectDetailLog.warn('delete project file failed', err);
+ if (typeof uiAlert === 'function') uiAlert(t('project.files.delete_failed'));
+ }
+}
+
+function _arrayBufferToBase64(buf) {
+ const bytes = new Uint8Array(buf || new ArrayBuffer(0));
+ let binary = '';
+ const step = 0x8000;
+ for (let i = 0; i < bytes.length; i += step) {
+ binary += String.fromCharCode.apply(null, bytes.subarray(i, Math.min(i + step, bytes.length)));
+ }
+ return btoa(binary);
+}
+
+function _setProjectFilesStatus(text) {
+ const status = document.getElementById('project-files-status');
+ if (status) status.textContent = text || '';
+}
+
+async function _uploadProjectFiles(fileList) {
+ if (!_projectDetailPid || !fileList || !fileList.length) return;
+ const files = Array.from(fileList).filter(Boolean);
+ if (!files.length) return;
+ _setProjectFilesStatus(t('project.files.uploading'));
+ const failed = [];
+ await Promise.all(files.map(async (file) => {
+ try {
+ const buf = await file.arrayBuffer();
+ const res = await window.orkas.invoke('projects.files.upload', {
+ projectId: _projectDetailPid,
+ name: file.name || 'file',
+ data: _arrayBufferToBase64(buf),
+ });
+ if (!res?.ok) {
+ failed.push(t('project.files.upload_failed', {
+ name: file.name || 'file',
+ reason: res?.error || t('chat.attach_upload_generic_fail'),
+ }));
+ }
+ } catch (err) {
+ failed.push(t('project.files.upload_failed', {
+ name: file.name || 'file',
+ reason: err?.message || t('chat.attach_upload_generic_fail'),
+ }));
+ }
+ }));
+ await loadProjectDetail(_projectDetailPid);
+ if (failed.length && typeof uiAlert === 'function') {
+ await uiAlert(t('project.files.upload_failed_list', { list: failed.join('\n') }));
+ }
+}
+
+async function _submitProjectChat() {
+ if (!_projectDetailPid) return;
+ const input = document.getElementById('project-chat-input');
+ const btn = document.getElementById('project-chat-send-btn');
+ const raw = (input?.value || '').trim();
+ if (!raw) return;
+ if (typeof ensureModelConfigured === 'function' && !ensureModelConfigured()) return;
+ const useSelection = (typeof consumeChatUseSelection === 'function')
+ ? consumeChatUseSelection('project')
+ : null;
+ const recipient = (typeof getChatRecipient === 'function')
+ ? getChatRecipient('project')
+ : null;
+ const withUse = (typeof transformWithChatUse === 'function')
+ ? transformWithChatUse(raw, useSelection)
+ : raw;
+ const content = (typeof applyRecipientPrefix === 'function')
+ ? applyRecipientPrefix(withUse, 'project')
+ : withUse;
+ if (btn) btn.disabled = true;
+ let convId = '';
+ try {
+ const res = await apiFetch('/api/conversations/create', {
+ method: 'POST',
+ headers: { 'Content-Type': 'application/json' },
+ body: JSON.stringify({ kind: 'normal', projectId: _projectDetailPid }),
+ });
+ const data = await res.json();
+ if (!data.ok) throw new Error(data.error || t('chat.create_conv_failed'));
+ const conv = data.conversation;
+ convId = conv.conversation_id;
+ conv.title = (typeof _autoTitle === 'function') ? _autoTitle(raw) : raw.slice(0, 32);
+ conversations.unshift(conv);
+ renderConversationList();
+ if (typeof loadProjects === 'function') loadProjects(true);
+ } catch (err) {
+ if (typeof uiAlert === 'function') {
+ await uiAlert(t('chat.create_conv_failed_with_reason', { reason: err?.message || err }));
+ }
+ if (btn) btn.disabled = false;
+ return;
+ }
+
+ if (input) {
+ input.value = '';
+ if (typeof autoGrow === 'function') autoGrow(input, 180);
+ }
+ const chatInput = document.getElementById('chat-input');
+ if (chatInput) {
+ chatInput.value = '';
+ if (typeof autoGrow === 'function') autoGrow(chatInput, 200);
+ }
+ if (typeof setView === 'function') setView('conversation', convId, { skipLoad: true });
+ if (recipient && typeof setChatRecipient === 'function') {
+ setChatRecipient('conversation', recipient);
+ }
+ if (useSelection && typeof setChatUseSelection === 'function') {
+ setChatUseSelection('conversation', useSelection);
+ }
+ if (btn) btn.disabled = false;
+ if (typeof sendInCurrentConversation === 'function') {
+ await sendInCurrentConversation(content);
+ }
+}
+
function _activeLang() {
if (typeof getLang === 'function') return getLang();
return 'en';
@@ -124,6 +330,18 @@ function _pickItemDescription(item, lang) {
return (typeof pickDesc === 'function') ? pickDesc(item, lang) : '';
}
+function _projectBindingSourceHtml(source, kind) {
+ if (!source) return '';
+ const label = (typeof catalogSourceLabel === 'function')
+ ? catalogSourceLabel(source, kind === 'skill' ? 'skills' : 'agents')
+ : String(source);
+ if (!label) return '';
+ const normalized = (typeof normalizeCatalogSource === 'function')
+ ? normalizeCatalogSource(source)
+ : String(source);
+ return `<span class="project-binding-source project-binding-source--${escapeHtml(normalized)}">${escapeHtml(label)}</span>`;
+}
+
function _bindRemoveButtons() {
document.querySelectorAll('#project-detail-content [data-action="remove"]').forEach((btn) => {
btn.addEventListener('click', async (e) => {
@@ -216,7 +434,7 @@ async function _openAddPicker(kind) {
const name = escapeHtml(c.name || id);
const desc = _pickItemDescription(c, lang);
const descHtml = desc ? `<div class="project-binding-desc muted">${escapeHtml(desc)}</div>` : '';
- const source = c.source ? `<span class="project-binding-source">${escapeHtml(c.source)}</span>` : '';
+ const source = _projectBindingSourceHtml(c.source, kind);
return `
<div class="project-binding-picker-item" data-kind="${kind}" data-id="${escapeHtml(id)}">
<div class="project-binding-main">
@@ -286,16 +504,7 @@ async function _openAddPicker(kind) {
setTimeout(() => searchEl.focus(), 0);
}
-// ── Header actions: New chat / Rename / Delete ────────────────────────
-
-function _onNewConvAction() {
- if (!_projectDetailPid) return;
- // Pin pid into the commander chip's localStorage key so the new-chat
- // panel mounts with this project pre-selected. Mirrors
- // `projects.js::_newConvWithProject` semantics — keep the key in sync.
- try { localStorage.setItem('commander.lastProject', _projectDetailPid); } catch (_) {}
- if (typeof setView === 'function') setView('new-chat');
-}
+// ── Header actions: Rename / Delete ────────────────────────────────────
function _onRenameAction() {
if (!_projectDetailMeta) return;
@@ -373,6 +582,11 @@ async function _onDeleteAction() {
// ── Empty-bindings banner inside in-project conversations ─────────────
+function _setConvProjectBannerVisible(visible) {
+ const pane = document.querySelector('#panel-conversation .chat-main-pane');
+ if (pane) pane.classList.toggle('has-project-empty-banner', !!visible);
+}
+
async function refreshConvProjectEmptyBanner(cid) {
const banner = document.getElementById('conv-project-empty-banner');
if (!banner) return;
@@ -384,6 +598,7 @@ async function refreshConvProjectEmptyBanner(cid) {
if (!pid) {
banner.style.display = 'none';
delete banner.dataset.pid;
+ _setConvProjectBannerVisible(false);
return;
}
try {
@@ -393,9 +608,13 @@ async function refreshConvProjectEmptyBanner(cid) {
// gates whether the user can do anything inside this project.
const isEmpty = !(b.agents || []).length;
banner.style.display = isEmpty ? '' : 'none';
- banner.dataset.pid = pid;
+ if (isEmpty) banner.dataset.pid = pid;
+ else delete banner.dataset.pid;
+ _setConvProjectBannerVisible(isEmpty);
} catch (_) {
banner.style.display = 'none';
+ delete banner.dataset.pid;
+ _setConvProjectBannerVisible(false);
}
}
@@ -410,10 +629,6 @@ document.addEventListener('DOMContentLoaded', () => {
e.stopPropagation();
_openAddPicker('skill');
});
- document.getElementById('project-action-new-conv')?.addEventListener('click', (e) => {
- e.stopPropagation();
- _onNewConvAction();
- });
document.getElementById('project-action-rename')?.addEventListener('click', (e) => {
e.stopPropagation();
_onRenameAction();
@@ -422,6 +637,35 @@ document.addEventListener('DOMContentLoaded', () => {
e.stopPropagation();
_onDeleteAction();
});
+ document.getElementById('project-add-file-btn')?.addEventListener('click', (e) => {
+ e.stopPropagation();
+ document.getElementById('project-file-input')?.click();
+ });
+ document.getElementById('project-chat-attach-btn')?.addEventListener('click', (e) => {
+ e.stopPropagation();
+ document.getElementById('project-file-input')?.click();
+ });
+ document.getElementById('project-file-input')?.addEventListener('change', async (e) => {
+ await _uploadProjectFiles(e.target.files);
+ e.target.value = '';
+ });
+ const projectInput = document.getElementById('project-chat-input');
+ if (projectInput) {
+ projectInput.addEventListener('input', () => {
+ if (typeof autoGrow === 'function') autoGrow(projectInput, 180);
+ });
+ projectInput.addEventListener('keydown', (e) => {
+ if (e.isComposing || e.keyCode === 229) return;
+ if (e.key === 'Enter' && !e.shiftKey) {
+ e.preventDefault();
+ _submitProjectChat();
+ }
+ });
+ }
+ document.getElementById('project-chat-send-btn')?.addEventListener('click', (e) => {
+ e.stopPropagation();
+ _submitProjectChat();
+ });
// Inline rename input handlers.
const inputEl = document.getElementById('project-detail-title-input');
if (inputEl) {
diff --git a/src/renderer/modules/projects.js b/src/renderer/modules/projects.js
index 62e65253..15348a0c 100644
--- a/src/renderer/modules/projects.js
+++ b/src/renderer/modules/projects.js
@@ -155,6 +155,7 @@ function _renderInlineCreateRow() {
function _renderProjectRow(p, convs) {
const expanded = !!_projectsExpanded[p.project_id];
+ const selected = _isProjectSelected(p.project_id);
const editing = _projectsInlineRenamePid === p.project_id;
const folderIcon = expanded
? _projectUiIconHtml('folder-open', 'project-folder-icon')
@@ -166,7 +167,7 @@ function _renderProjectRow(p, convs) {
value="${escapeHtml(p.name || '')}" autocomplete="off" spellcheck="false" />`
: `<span class="project-name" title="${safeName}">${safeName}</span>`;
let html = `
- <div class="project-row" data-pid="${escapeHtml(p.project_id)}">
+ <div class="project-row${selected ? ' active' : ''}" data-pid="${escapeHtml(p.project_id)}">
<span class="project-icon">${folderIcon}</span>
${nameNode}
<button type="button" class="ctx-row-menu-btn project-row-menu-btn" data-project-menu
@@ -190,6 +191,12 @@ function _renderProjectRow(p, convs) {
return html;
}
+function _isProjectSelected(pid) {
+ return currentView === 'project'
+ && typeof _projectDetailPid !== 'undefined'
+ && _projectDetailPid === pid;
+}
+
// ── Event wiring ────────────────────────────────────────────────────────
function _bindProjectsHandlers(container) {
@@ -202,9 +209,9 @@ function _bindProjectsHandlers(container) {
_bindInlineRenameInput(input);
});
- // Project rows: any click → BOTH open the detail page AND toggle the
- // expand state of nested conversations (per design — single click does
- // both). Exceptions: ⋯ opens its menu; rename input swallows click.
+ // Project rows: first click selects/opens the project detail. Once that
+ // project is already selected, clicking the row toggles its nested
+ // conversations.
container.querySelectorAll('.project-row[data-pid]').forEach((row) => {
const pid = row.dataset.pid;
row.addEventListener('click', (e) => {
@@ -217,8 +224,12 @@ function _bindProjectsHandlers(container) {
_openProjectRowMenu(e.target.closest('[data-project-menu]'), pid);
return;
}
+ if (!_isProjectSelected(pid)) {
+ if (typeof setView === 'function') setView('project', pid);
+ renderProjectsSection();
+ return;
+ }
_toggleProjectExpand(pid);
- if (typeof setView === 'function') setView('project', pid);
});
});
@@ -405,7 +416,6 @@ function _openProjectRowMenu(anchorBtn, pid) {
_closeProjectRowMenu();
const items = [
- { action: 'new_conv', label: t('project.menu.new_conv') },
{ action: 'rename', label: t('project.menu.rename') },
{ action: 'delete', label: t('project.menu.delete'), danger: true },
];
@@ -456,20 +466,10 @@ function _closeProjectRowMenu() {
}
async function _runProjectMenuAction(action, pid) {
- if (action === 'new_conv') return _newConvWithProject(pid);
if (action === 'rename') return _startProjectInlineRename(pid);
if (action === 'delete') return _confirmDeleteProject(pid);
}
-// Pin pid into the commander chip's localStorage key BEFORE switching view —
-// onEnterCommanderProjectChip (fired from boot.js view-change) reads the same
-// key and re-validates against _projectsCache, so by the time the new-chat
-// panel mounts the chip is already showing this project.
-function _newConvWithProject(pid) {
- try { localStorage.setItem(_COMMANDER_LAST_PROJECT_KEY, pid); } catch (_) {}
- setView('new-chat');
-}
-
// ── Delete flow ─────────────────────────────────────────────────────────
async function _confirmDeleteProject(pid) {
diff --git a/src/renderer/modules/search.js b/src/renderer/modules/search.js
index 057766c1..91351980 100644
--- a/src/renderer/modules/search.js
+++ b/src/renderer/modules/search.js
@@ -212,7 +212,8 @@ const _SEARCH_KIND_META = {
const _SEARCH_SOURCE_LABEL = {
custom: 'search.source.custom',
- builtin: 'search.source.builtin',
+ builtin: 'search.source.marketplace',
+ marketplace: 'search.source.marketplace',
};
// Render one result into HTML. Uses the shared `_renderSearchRow` helper for
@@ -235,7 +236,8 @@ function _renderSearchRow(r, dataIdx, query) {
}
} else if (r.kind === 'agent' || r.kind === 'skill') {
title = r.name || r.id;
- const srcKey = _SEARCH_SOURCE_LABEL[r.source] || '';
+ const source = (typeof normalizeCatalogSource === 'function') ? normalizeCatalogSource(r.source) : r.source;
+ const srcKey = _SEARCH_SOURCE_LABEL[source] || _SEARCH_SOURCE_LABEL[r.source] || '';
sub = srcKey ? t(srcKey) : '';
}
const active = dataIdx === _searchActiveIdx ? ' active' : '';
@@ -418,4 +420,3 @@ function _saveSearchHistoryEntry(query) {
cur.unshift(q);
_saveSearchHistory(cur.slice(0, _SEARCH_HISTORY_MAX));
}
-
diff --git a/src/renderer/modules/skills.js b/src/renderer/modules/skills.js
index 66708dc4..a3dcb59f 100644
--- a/src/renderer/modules/skills.js
+++ b/src/renderer/modules/skills.js
@@ -4,6 +4,19 @@ const _skillsLog = createLogger('skills');
let _skillsCache = null;
let _selectedSkill = null; // { source, id }
const _SKILL_SPEC_CATEGORY_CODE_RE = /^[a-z][a-z0-9_-]{0,79}$/;
+
+function _skillSource(source) {
+ return (typeof normalizeCatalogSource === 'function')
+ ? normalizeCatalogSource(source)
+ : String(source || '');
+}
+
+function _isSkillPlatformSource(source) {
+ return (typeof isMarketplaceCatalogSource === 'function')
+ ? isMarketplaceCatalogSource(source)
+ : _skillSource(source) === 'marketplace';
+}
+
function _normalizeSkillCategoryForHiddenSave(category) {
const code = String(category || '').trim().toLowerCase();
return _SKILL_SPEC_CATEGORY_CODE_RE.test(code) ? code : 'general';
@@ -55,7 +68,7 @@ async function refreshSkillsAfterMarketplaceReconcile() {
_skillTreeCache.clear();
await loadSkills(true);
if (_skillEditMode) return;
- if (_selectedSkill?.id && _selectedSkill.source === 'builtin') {
+ if (_selectedSkill?.id && _isSkillPlatformSource(_selectedSkill.source)) {
const source = _selectedSkill.source;
const id = _selectedSkill.id;
const filepath = _selectedSkill.filepath || 'SKILL.md';
@@ -75,7 +88,10 @@ async function loadSkills(forceRefresh) {
const res = await apiFetch('/api/skills/list');
const data = await res.json();
if (data.ok) {
- _skillsCache = data.skills || [];
+ _skillsCache = (data.skills || []).map((s) => ({
+ ...s,
+ source: _skillSource(s.source),
+ }));
renderSkillsList(_skillsCache);
}
} catch (e) {
@@ -89,7 +105,7 @@ async function loadSkills(forceRefresh) {
// mutually exclusive because both are textual routing hints prepended to the
// next message; agents remain separate in the recipient chip.
-const _chatUse = { 'new-chat': null, 'conversation': null };
+const _chatUse = { 'new-chat': null, 'conversation': null, project: null };
function bindSkillPicker() {
// Chip remove (×)
@@ -163,7 +179,9 @@ function _renderChatUseChipLabel(labelEl, selection) {
function _renderChatUseChip(target) {
const next = _normalizeChatUseSelection(_chatUse[target]);
- const chipId = target === 'new-chat' ? 'new-chat-skill-chip' : 'chat-skill-chip';
+ const chipId = target === 'new-chat'
+ ? 'new-chat-skill-chip'
+ : (target === 'project' ? 'project-chat-skill-chip' : 'chat-skill-chip');
const chip = document.getElementById(chipId);
if (!chip) return;
if (!next) {
@@ -187,6 +205,7 @@ function _renderChatUseChip(target) {
function refreshChatUseChips() {
_renderChatUseChip('new-chat');
_renderChatUseChip('conversation');
+ _renderChatUseChip('project');
}
function setChatUseSelection(target, selection) {
@@ -200,7 +219,7 @@ function setChatUseSelection(target, selection) {
// Return focus to the textarea after selection.
const input = target === 'new-chat'
? document.getElementById('new-chat-input')
- : document.getElementById('chat-input');
+ : (target === 'project' ? document.getElementById('project-chat-input') : document.getElementById('chat-input'));
input?.focus();
}
@@ -242,7 +261,7 @@ function renderSkillsList(skills) { renderSkillsGrid(skills); }
function renderSkillsGrid(skills) {
const custom = skills.filter(s => s.source === 'custom');
- const builtin = skills.filter(s => s.source === 'builtin');
+ const marketplace = skills.filter(s => _isSkillPlatformSource(s.source));
const emptyEl = document.getElementById('skills-empty');
const customGroup = document.getElementById('skills-grid-custom-group');
const builtinGroup = document.getElementById('skills-grid-builtin-group');
@@ -264,14 +283,14 @@ function renderSkillsGrid(skills) {
const desc = pickDesc(s, getLang()).trim();
const descClass = desc ? 'skill-card-desc' : 'skill-card-desc is-empty';
const descText = desc || t('skills.no_desc');
- // Source label (custom / builtin) is shown on the detail page only;
+ // Source label (custom / marketplace) is shown on the detail page only;
// the grid cards stay clean and use the ⋯ menu for actions.
const moreBtn = `<button type="button" class="skill-card-more" data-skill-more title="${moreTitle}" aria-label="${moreTitle}">⋯</button>`;
const enabled = s.enabled !== false;
- // Marketplace-installed (`builtin`) skills carry version + category from _install.json /
+ // Marketplace-installed skills carry version + category from _install.json /
// SKILL.md frontmatter — show them as inline chips so users see provenance without
// opening the detail page. Custom skills skip (no published version).
- const platformChips = s.source === 'builtin' ? _skillPlatformChipsHtml(s) : '';
+ const platformChips = _isSkillPlatformSource(s.source) ? _skillPlatformChipsHtml(s) : '';
return `
<div class="skill-card${enabled ? '' : ' is-disabled'}" data-id="${escapeHtml(s.id)}" data-source="${s.source}">
<div class="skill-card-header">
@@ -295,7 +314,7 @@ function renderSkillsGrid(skills) {
gridEl.innerHTML = list.map(cardHtml).join('');
};
renderGroup(custom, customGroup, customGrid);
- renderGroup(builtin, builtinGroup, builtinGrid);
+ renderGroup(marketplace, builtinGroup, builtinGrid);
// Wire card / ▶ / ⋯ click handlers. (Enable/disable lives in the ⋯ menu
// now — no toggle switch on the card.)
@@ -373,6 +392,7 @@ function _showSkillsGridView() {
}
async function _showSkillsDetailView(source, id) {
+ source = _skillSource(source);
const grid = document.getElementById('skills-grid-view');
const detail = document.getElementById('skills-detail-view');
if (grid) grid.style.display = 'none';
@@ -447,6 +467,10 @@ function _openSkillRowMenu(anchorBtn, id, source) {
// Dev-only entry on builtin: tag the label so the user knows this isn't a
// normal user capability (mirrors marketplace.upload's "(dev)" treatment).
const editLabelSuffix = (source === 'builtin' && false) ? t('common.dev_suffix') : '';
+ const canEdit = source === 'custom' || (_isSkillPlatformSource(source) && false);
+ // Dev-only entry on marketplace items: tag the label so the user knows this isn't a
+ // normal user capability (mirrors marketplace.upload's "(dev)" treatment).
+ const editLabelSuffix = (_isSkillPlatformSource(source) && false) ? t('common.dev_suffix') : '';
const items = [];
if (canEdit) {
items.push(`<div class="skill-row-menu-item" data-action="edit">${escapeHtml(t('skills.edit') + editLabelSuffix)}</div>`);
@@ -548,6 +572,7 @@ async function _toggleSkillsSource() {
}
async function expandSkillTree(source, id, childrenEl) {
+ source = _skillSource(source);
const key = `${source}:${id}`;
let tree = _skillTreeCache.get(key);
if (!tree) {
@@ -675,6 +700,7 @@ function findDirInTree(tree, targetPath) {
}
async function selectSkillFile(source, id, filepath, nodeEl) {
+ source = _skillSource(source);
const skill = _skillsCache?.find(s => s.id === id && s.source === source);
// Detect "same-skill file switch" vs initial / cross-skill load — only the
// former needs scroll preservation (user is mid-page browsing source files).
@@ -721,6 +747,7 @@ async function selectSkillFile(source, id, filepath, nodeEl) {
// In edit mode only the "done" button (the relabeled "edit") is
// shown; everything else hides.
const canEditThisSkill = source === 'custom';
+ const canEditThisSkill = source === 'custom' || (_isSkillPlatformSource(source) && false);
const editingThis = _skillEditMode && _skillEditSkillId === id && canEditThisSkill;
const actions = document.getElementById('skills-detail-actions');
if (actions) {
@@ -1160,9 +1187,10 @@ function _updateEditButtonLabel() {
btn.textContent = t('skills.edit_btn_done');
return;
}
- // Tag the "Edit" label on builtin skills (dev-only entry); the "Done"
+ // Tag the "Edit" label on marketplace skills (dev-only entry); the "Done"
// branch above stays bare — in edit mode the marker is redundant.
const suffix = (_selectedSkill?.source === 'builtin' && false) ? t('common.dev_suffix') : '';
+ const suffix = (_isSkillPlatformSource(_selectedSkill?.source) && false) ? t('common.dev_suffix') : '';
btn.textContent = t('skills.edit_btn_edit') + suffix;
}
@@ -1176,6 +1204,9 @@ function _updateEditButtonLabel() {
async function toggleSkillEditMode(opts = {}) {
if (!_selectedSkill) return;
if (_selectedSkill.source !== 'custom') return;
+ // Marketplace editing is dev-only; lift the source guard accordingly.
+ if (_selectedSkill.source !== 'custom'
+ && !(_isSkillPlatformSource(_selectedSkill.source) && false)) return;
if (_skillEditMode && _skillEditSkillId === _selectedSkill.id) {
// Abort any in-flight reply so "done" means "stop + exit", not
// "exit but keep streaming". The chat controller is a singleton;
@@ -1570,11 +1601,16 @@ function editSelectedSkill() {
async function deleteSelectedSkill() {
if (!_selectedSkill) return;
if (_selectedSkill.source !== 'custom') return;
+ const src = _selectedSkill.source;
+ if (src !== 'custom' && !(_isSkillPlatformSource(src) && false)) return;
const sid = _selectedSkill.id;
const cached = _skillsCache?.find(s => s.id === sid && s.source === 'custom');
if (!(await uiConfirm(t('skills.delete_confirm', { name: cached?.name || sid })))) return;
try {
const result = await (await apiFetch(`/api/skills/${sid}`, { method: 'DELETE' })).json();
+ const result = _isSkillPlatformSource(src)
+ ? await window.orkas.invoke('skills.builtin.delete', { id: sid })
+ : await (await apiFetch(`/api/skills/${sid}`, { method: 'DELETE' })).json();
if (!result.ok) {
await uiAlert(t('skills.delete_failed_with', { reason: result.error || '' }));
return;
diff --git a/src/renderer/modules/state.js b/src/renderer/modules/state.js
index ddc4e372..89277c1e 100644
--- a/src/renderer/modules/state.js
+++ b/src/renderer/modules/state.js
@@ -287,9 +287,6 @@ function bindStaticHandlers() {
});
document.getElementById('agent-edit-btn')?.addEventListener('click', toggleAgentEditMode);
document.getElementById('agent-delete-btn')?.addEventListener('click', deleteSelectedAgent);
- document.getElementById('agent-promote-btn')?.addEventListener('click', () => {
- if (_selectedAgent?.source === 'custom') promoteCustomAgent(_selectedAgent.id);
- });
document.getElementById('agent-upload-marketplace-btn')?.addEventListener('click', () => {
// Upload is allowed for both custom AND builtin (dev mode) — same
// marketplace publishing flow per CLAUDE.md §11. The row-menu handler
@@ -329,9 +326,6 @@ function bindStaticHandlers() {
});
document.getElementById('skill-edit-btn')?.addEventListener('click', toggleSkillEditMode);
document.getElementById('skill-delete-btn')?.addEventListener('click', deleteSelectedSkill);
- document.getElementById('skill-promote-btn')?.addEventListener('click', () => {
- if (_selectedSkill?.source === 'custom') promoteCustomSkill(_selectedSkill.id);
- });
document.getElementById('skill-upload-marketplace-btn')?.addEventListener('click', () => {
// Upload is allowed for both custom AND builtin (dev mode) — same as
// the agent upload handler above; see comment there.
diff --git a/src/renderer/modules/strip-structural-blocks.js b/src/renderer/modules/strip-structural-blocks.js
index de5df558..b7b3f196 100644
--- a/src/renderer/modules/strip-structural-blocks.js
+++ b/src/renderer/modules/strip-structural-blocks.js
@@ -437,6 +437,15 @@ function _replaceOuterSkillFileBlocks(buf, makePlaceholder) {
return out;
}
+function _collapseRepeatedStructuralPlaceholders(buf, placeholder) {
+ if (!buf || !placeholder) return buf;
+ const marker = String(placeholder).trim();
+ if (!marker) return buf;
+ const escaped = marker.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
+ const collapsed = String(buf).replace(new RegExp(`${escaped}(?:\\s*${escaped})+`, 'g'), marker);
+ return collapsed.replace(/\n{3,}/g, '\n\n');
+}
+
// Streaming-time strip for the commander's `<skill>` container. Two modes:
// - **Closed** `<skill>...</skill>` → strip outer tags + any `<skill_id>`
// sub-tag, keep the inner content. The inner `<<<skill-file>>>` blocks
@@ -610,5 +619,6 @@ if (typeof module !== 'undefined' && typeof module.exports === 'object') {
_replaceKnownSkillIdsForDisplay,
_simplifyKnownSkillFollowPhrasesForDisplay,
_normalizeKnownSkillRefsForDisplay,
+ _collapseRepeatedStructuralPlaceholders,
};
}
diff --git a/src/renderer/modules/text-view-edit.js b/src/renderer/modules/text-view-edit.js
new file mode 100644
index 00000000..720bff1b
--- /dev/null
+++ b/src/renderer/modules/text-view-edit.js
@@ -0,0 +1,257 @@
+// ─── Plain-text view/edit controller ────────────────────────────────────
+// Sibling to `mountMdViewEdit` for chat-file-viewer's non-markdown text
+// files (.py, .json, .csv, .log, source code, etc.). Same controller shape
+// (`destroy / isDirty / getMode / setMode`) so chat-file-viewer can hold
+// either MD or text controller in `_viewerEditController` without branching.
+//
+// Why a sibling instead of a `plainText` flag on mountMdViewEdit:
+// - Plain text has no markdown toolbar, no preview toggle, no
+// task-checkbox interactivity. Threading `plainText: true` through
+// md-view-edit would mean three `if (!plainText) …` branches in the
+// editor / view paths — duplicating logic at the call site is less
+// code than gating each rendering primitive.
+// - PC/CLAUDE.md §8 "only md files show the md toolbar" — a physical
+// module boundary makes that invariant unforgeable; adding a flag
+// leaves the toolbar visible if anyone later forgets the gate.
+//
+// Backend reuse: same `produced.readText` / `produced.writeText` IPC as
+// the workspace branch of md-view-edit. Path-sandbox + 2 MB read cap
+// already enforced on the main side, so nothing new to validate here.
+//
+// Source shape: { absPath: string, cid?: string, projectId?: string }
+// `cid` widens the read/write scope to include this conversation's
+// attachment dir; `projectId` widens it to the project file pool.
+
+const _tveLog = createLogger('text-view-edit');
+
+/**
+ * Mount a plain-text view/edit surface inside `bodyEl` with action buttons
+ * placed in `actionsEl` (Edit in view mode; Save+Cancel in edit mode).
+ *
+ * Required: opts.bodyEl, opts.actionsEl, opts.source.absPath.
+ * Optional: opts.initialContent (skips the initial read), opts.initialMode
+ * ('view' default), opts.actionIconOnly, opts.capabilities
+ * ({ edit, save } — defaults true; pass { edit: false } to lock
+ * read-only for project files etc.), opts.callbacks
+ * ({ onDirtyChange, onSaved, onContentLoaded }).
+ *
+ * Returns a noop-shell controller on misuse rather than throwing — the
+ * call site has no good fallback if mount fails after the overlay opened.
+ */
+function mountTextViewEdit(opts) {
+ const bodyEl = opts && opts.bodyEl;
+ const actionsEl = opts && opts.actionsEl;
+ if (!bodyEl || !actionsEl) {
+ _tveLog.warn('mountTextViewEdit called without bodyEl/actionsEl', { hasBody: !!bodyEl, hasActions: !!actionsEl });
+ return _tveNoopController();
+ }
+ const source = opts.source;
+ if (!source || typeof source.absPath !== 'string' || !source.absPath) {
+ _tveLog.warn('mountTextViewEdit called without source.absPath');
+ return _tveNoopController();
+ }
+
+ const caps = Object.assign({ edit: true, save: true }, opts.capabilities || {});
+ const callbacks = opts.callbacks || {};
+
+ const state = {
+ source,
+ caps,
+ callbacks,
+ bodyEl,
+ actionsEl,
+ actionIconOnly: opts.actionIconOnly === true,
+ content: typeof opts.initialContent === 'string' ? opts.initialContent : '',
+ draft: '',
+ mode: opts.initialMode === 'edit' ? 'edit' : 'view',
+ destroyed: false,
+ };
+ state.draft = state.content;
+
+ if (typeof opts.initialContent !== 'string') {
+ _tveLoadAndRender(state);
+ } else {
+ _tveRender(state);
+ }
+
+ function destroy() {
+ state.destroyed = true;
+ if (bodyEl) bodyEl.innerHTML = '';
+ if (actionsEl) actionsEl.innerHTML = '';
+ }
+ function isDirty() { return state.mode === 'edit' && state.draft !== state.content; }
+ function getMode() { return state.mode; }
+ function setMode(m) {
+ if (state.destroyed) return;
+ if (m !== 'view' && m !== 'edit') return;
+ if (m === 'edit') _tveEnterEdit(state);
+ else _tveRenderView(state);
+ }
+ async function refreshContent() {
+ if (state.destroyed) return;
+ await _tveLoadAndRender(state);
+ }
+
+ return { destroy, isDirty, getMode, setMode, refreshContent };
+}
+
+function _tveNoopController() {
+ return {
+ destroy() {},
+ isDirty() { return false; },
+ getMode() { return 'view'; },
+ setMode() {},
+ refreshContent() {},
+ };
+}
+
+async function _tveLoadAndRender(state) {
+ state.bodyEl.innerHTML = `<div class="chat-file-viewer-loading">…</div>`;
+ state.actionsEl.innerHTML = '';
+ try {
+ const payload = { path: state.source.absPath };
+ if (state.source.cid) payload.cid = state.source.cid;
+ if (state.source.projectId) payload.projectId = state.source.projectId;
+ const res = await window.orkas.invoke('produced.readText', payload);
+ if (state.destroyed) return;
+ if (!res || !res.ok) {
+ _tveRenderError(state, (res && res.error) || 'read_failed');
+ return;
+ }
+ state.content = String(res.text || '');
+ state.draft = state.content;
+ if (state.callbacks.onContentLoaded) {
+ try { state.callbacks.onContentLoaded(state.content); }
+ catch (err) { _tveLog.warn('onContentLoaded threw', err); }
+ }
+ _tveRender(state);
+ } catch (err) {
+ if (state.destroyed) return;
+ _tveRenderError(state, err && err.message || String(err));
+ }
+}
+
+function _tveRenderError(state, msg) {
+ state.bodyEl.innerHTML = `<div class="chat-file-viewer-loading">${escapeHtml(_tveLabel('contexts.viewer.read_failed', `Failed to read file: ${msg}`, { reason: msg }))}</div>`;
+ state.actionsEl.innerHTML = '';
+}
+
+function _tveRender(state) {
+ if (state.destroyed) return;
+ if (state.mode === 'edit') _tveRenderEditor(state);
+ else _tveRenderView(state);
+}
+
+function _tveRenderView(state) {
+ state.mode = 'view';
+ state.bodyEl.innerHTML = `<pre class="chat-file-viewer-text">${escapeHtml(state.content)}</pre>`;
+ const actions = [];
+ if (state.caps.edit) actions.push(_tveActionButton(state, 'edit', 'contexts.viewer.edit', 'pencil'));
+ state.actionsEl.innerHTML = actions.join('');
+ state.actionsEl.querySelector('[data-tve-action="edit"]')?.addEventListener('click', () => _tveEnterEdit(state));
+ _tveEmitDirty(state);
+}
+
+function _tveEnterEdit(state) {
+ if (!state.caps.edit) return;
+ state.mode = 'edit';
+ state.draft = state.content;
+ _tveRenderEditor(state);
+ _tveEmitDirty(state);
+}
+
+function _tveRenderEditor(state) {
+ // Tab key in the textarea inserts \t instead of moving focus — for code /
+ // config files this matters (the alternative — sending tab to focus the
+ // next action button — has no point inside a fullscreen editor).
+ state.bodyEl.innerHTML = `<textarea class="ctx-viewer-editor" data-tve-textarea spellcheck="false">${escapeHtml(state.draft)}</textarea>`;
+ const actions = [];
+ if (state.caps.save) {
+ actions.push(_tveActionButton(state, 'save', 'contexts.viewer.save', 'check', 'btn-primary'));
+ actions.push(_tveActionButton(state, 'cancel', 'contexts.viewer.cancel', 'x'));
+ }
+ state.actionsEl.innerHTML = actions.join('');
+ state.actionsEl.querySelector('[data-tve-action="save"]')?.addEventListener('click', () => _tveSave(state));
+ state.actionsEl.querySelector('[data-tve-action="cancel"]')?.addEventListener('click', () => _tveCancelEdit(state));
+
+ const ta = state.bodyEl.querySelector('[data-tve-textarea]');
+ if (ta) {
+ ta.addEventListener('input', () => {
+ state.draft = ta.value;
+ _tveEmitDirty(state);
+ });
+ ta.addEventListener('keydown', (e) => {
+ if (e.key === 'Tab') {
+ e.preventDefault();
+ const start = ta.selectionStart;
+ const end = ta.selectionEnd;
+ const before = ta.value.slice(0, start);
+ const after = ta.value.slice(end);
+ ta.value = `${before}\t${after}`;
+ ta.selectionStart = ta.selectionEnd = start + 1;
+ state.draft = ta.value;
+ _tveEmitDirty(state);
+ }
+ });
+ ta.focus();
+ }
+ _tveEmitDirty(state);
+}
+
+function _tveCancelEdit(state) {
+ state.draft = state.content;
+ _tveRenderView(state);
+}
+
+async function _tveSave(state) {
+ const ta = state.bodyEl.querySelector('[data-tve-textarea]');
+ if (ta) state.draft = ta.value;
+ const next = state.draft;
+ try {
+ const payload = { path: state.source.absPath, content: next };
+ if (state.source.cid) payload.cid = state.source.cid;
+ if (state.source.projectId) payload.projectId = state.source.projectId;
+ const res = await window.orkas.invoke('produced.writeText', payload);
+ if (!res || !res.ok) {
+ await uiAlert(_tveLabel('contexts.save_failed_with', `Save failed: ${(res && res.error) || 'unknown'}`, { reason: (res && res.error) || 'unknown' }));
+ return;
+ }
+ } catch (err) {
+ await uiAlert(_tveLabel('contexts.save_failed_with', `Save failed: ${err && err.message || err}`, { reason: err && err.message || String(err) }));
+ return;
+ }
+ state.content = next;
+ _tveRenderView(state);
+ if (state.callbacks.onSaved) {
+ try { state.callbacks.onSaved(next); }
+ catch (err) { _tveLog.warn('onSaved threw', err); }
+ }
+}
+
+function _tveActionButton(state, action, labelKey, iconName, extraClass) {
+ const label = _tveLabel(labelKey, labelKey);
+ const cls = `btn btn-sm${state.actionIconOnly ? ' btn-icon ctx-viewer-action-icon-btn' : ''}${extraClass ? ` ${extraClass}` : ''}`;
+ if (!state.actionIconOnly) {
+ return `<button type="button" class="${cls}" data-tve-action="${action}">${escapeHtml(label)}</button>`;
+ }
+ const icon = (typeof window !== 'undefined' && typeof window.uiIconHtml === 'function')
+ ? window.uiIconHtml(iconName, 'ui-icon ctx-viewer-action-icon')
+ : escapeHtml(label);
+ return `<button type="button" class="${cls}" data-tve-action="${action}" aria-label="${escapeHtml(label)}" title="${escapeHtml(label)}">${icon}</button>`;
+}
+
+function _tveEmitDirty(state) {
+ if (state.callbacks.onDirtyChange) {
+ try { state.callbacks.onDirtyChange(state.mode === 'edit' && state.draft !== state.content); }
+ catch (err) { _tveLog.warn('onDirtyChange threw', err); }
+ }
+}
+
+// Localized label helper — falls back to the provided English when the
+// key is missing. Mirrors chat-file-viewer's `_viewerLabel`.
+function _tveLabel(key, fallback, vars) {
+ try {
+ const v = vars ? t(key, vars) : t(key);
+ return v === key ? fallback : v;
+ } catch (_) { return fallback; }
+}
diff --git a/src/renderer/modules/user-workspace.js b/src/renderer/modules/user-workspace.js
index 0d9dce3b..db5e1490 100644
--- a/src/renderer/modules/user-workspace.js
+++ b/src/renderer/modules/user-workspace.js
@@ -24,6 +24,7 @@ const _wsLog = createLogger('user-workspace');
const _wsInfoByTarget = {
'new-chat': { currentPath: '', defaultPath: '', isDefault: true, recentPaths: [], scope: 'default' },
'conversation': { currentPath: '', defaultPath: '', isDefault: true, recentPaths: [], scope: 'default' },
+ project: { currentPath: '', defaultPath: '', isDefault: true, recentPaths: [], scope: 'default' },
};
// ── Workspace display helpers ───────────────────────────────────────
@@ -42,6 +43,10 @@ function _wsScopeHintFor(target) {
if (target === 'conversation') {
return currentCid ? { cid: currentCid } : {};
}
+ if (target === 'project') {
+ const pid = (typeof _projectDetailPid !== 'undefined') ? (_projectDetailPid || '') : '';
+ return pid ? { projectId: pid } : {};
+ }
// new-chat / commander panel
if (typeof getCommanderProjectId === 'function') {
const pid = getCommanderProjectId();
@@ -55,7 +60,7 @@ function _updateChipForTarget(target) {
const info = _wsInfoByTarget[target] || _wsInfoByTarget['conversation'];
const sel = target === 'new-chat'
? '#panel-new-chat .workspace-chip'
- : '#panel-conversation .workspace-chip';
+ : (target === 'project' ? '#panel-project .workspace-chip' : '#panel-conversation .workspace-chip');
const chip = document.querySelector(sel);
if (!chip) return;
const label = chip.querySelector('.workspace-chip-label');
@@ -75,6 +80,7 @@ function _updateChipForTarget(target) {
function _updateAllChips() {
_updateChipForTarget('new-chat');
_updateChipForTarget('conversation');
+ _updateChipForTarget('project');
}
// ── Core actions ────────────────────────────────────────────────────
@@ -101,7 +107,7 @@ async function _fetchWorkspaceInfo(target) {
/** Refetch info for both targets — called on boot, on commander chip change,
* and on conversation switch. */
async function _refreshAllWorkspaceInfo() {
- await Promise.all([_fetchWorkspaceInfo('new-chat'), _fetchWorkspaceInfo('conversation')]);
+ await Promise.all([_fetchWorkspaceInfo('new-chat'), _fetchWorkspaceInfo('conversation'), _fetchWorkspaceInfo('project')]);
_updateAllChips();
}
@@ -312,6 +318,13 @@ async function initUserWorkspace() {
if (sendBtn) convBar.insertBefore(chip, sendBtn);
else convBar.appendChild(chip);
}
+ const projectBar = document.querySelector('#panel-project .chat-bottom-bar');
+ if (projectBar) {
+ const sendBtn = projectBar.querySelector('.chat-send-btn');
+ const chip = _createWorkspaceChip('project');
+ if (sendBtn) projectBar.insertBefore(chip, sendBtn);
+ else projectBar.appendChild(chip);
+ }
await _refreshAllWorkspaceInfo();
}
diff --git a/src/renderer/modules/utils.js b/src/renderer/modules/utils.js
index ccb38592..5e1504fd 100644
--- a/src/renderer/modules/utils.js
+++ b/src/renderer/modules/utils.js
@@ -52,6 +52,29 @@ function escapeHtml(str) {
.replace(/"/g, '&quot;').replace(/'/g, '&#039;');
}
+function normalizeCatalogSource(source) {
+ const s = String(source || '').trim().toLowerCase();
+ if (s === 'builtin' || s === 'platform') return 'marketplace';
+ if (s === 'custom') return 'custom';
+ return s;
+}
+
+function isCustomCatalogSource(source) {
+ return normalizeCatalogSource(source) === 'custom';
+}
+
+function isMarketplaceCatalogSource(source) {
+ return normalizeCatalogSource(source) === 'marketplace';
+}
+
+function catalogSourceLabel(source, kind = 'agents') {
+ const normalized = normalizeCatalogSource(source);
+ const base = kind === 'skills' ? 'skills' : 'agents';
+ if (normalized === 'custom') return t(`${base}.source_custom`);
+ if (normalized === 'marketplace') return t(`${base}.source_marketplace`);
+ return source ? String(source) : '';
+}
+
// Full markdown renderer (used for skill detail view and chat)
function renderMarkdownFull(md) {
if (!md) return '';
@@ -730,27 +753,11 @@ function nowIsoLocal() {
* refreshes with backend's value and the sidebar entry visibly flips.
* Empty input returns '' so the caller can fall back to its own
* placeholder (`t('chat.new_conv_title')` for the conv list). */
-const _AUTO_TITLE_ZH_FILLER = /^(帮我看一下|可不可以|帮我看下|帮我看看|麻烦你|帮我看|我想要|想问问|能不能|可以不|可以吗|请帮我|看一下|麻烦|帮我|我想|想问|请问|看下|看看)\s*/;
-const _AUTO_TITLE_EN_FILLER = /^(could you|would you|can you|help me|i'?d like to|i want to|please)\s+/i;
-const _AUTO_TITLE_CLAUSE = /[,。?!;;,.?!]/;
-const _AUTO_TITLE_MAX = 30;
-function _autoTitle(text) {
- const raw = String(text == null ? '' : text).trim().replace(/\s+/g, ' ');
- if (!raw) return '';
- let s = raw;
- for (let i = 0; i < 5; i++) {
- const before = s;
- s = s.replace(_AUTO_TITLE_ZH_FILLER, '').replace(_AUTO_TITLE_EN_FILLER, '');
- if (s === before) break;
- }
- s = s.trim();
- const clauseIdx = s.search(_AUTO_TITLE_CLAUSE);
- if (clauseIdx >= 4) s = s.slice(0, clauseIdx);
- s = s.trim();
- if (!s) s = raw;
- if (s.length > _AUTO_TITLE_MAX) s = s.slice(0, _AUTO_TITLE_MAX) + '…';
- return s;
-}
+// `_autoTitle` + its regex constants live in `modules/auto-title.js` so the
+// regex set has a single renderer-side home and a clean target for the
+// `test/renderer/auto-title-parity.test.ts` parity check against
+// `src/main/util/auto-title.ts`. The new file is loaded before consumers
+// via `<script src="./modules/auto-title.js">` in index.html.
function formatTime(iso) {
if (!iso) return '-';
diff --git a/src/renderer/style.css b/src/renderer/style.css
index 95bb9bbe..475ff259 100644
--- a/src/renderer/style.css
+++ b/src/renderer/style.css
@@ -474,11 +474,22 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
text-align: left;
}
+.sidebar-btn [data-ui-icon] {
+ display: inline-flex;
+ align-items: center;
+ justify-content: center;
+ width: 15px;
+ height: 15px;
+ flex: 0 0 15px;
+ line-height: 0;
+}
+
.sidebar-btn-icon {
width: 15px;
height: 15px;
flex: 0 0 15px;
color: #94a3b8;
+ display: block;
}
.sidebar-btn.active .sidebar-btn-icon {
@@ -625,6 +636,8 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
position: relative;
}
.project-row:hover { background: #f1f5f9; color: var(--text); }
+.project-row.active { background: #eff6ff; color: var(--primary); }
+.project-row.active .project-icon { color: var(--primary); }
.project-row.is-menu-open { background: #f1f5f9; }
.project-icon {
@@ -707,20 +720,25 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
.project-detail-content {
flex: 1;
overflow-y: auto;
- padding: 22px 32px 32px 32px;
- max-width: 880px;
+ padding: 28px 32px 36px 32px;
+ max-width: 1180px;
width: 100%;
margin: 0 auto;
box-sizing: border-box;
+ display: flex;
+ flex-direction: column;
}
.project-detail-header {
display: flex;
align-items: center;
- gap: 12px;
- margin-bottom: 22px;
+ gap: 16px;
+ margin-bottom: 20px;
}
-.project-detail-title {
+.project-detail-title-wrap {
flex: 1;
+ min-width: 0;
+}
+.project-detail-title {
font-size: 22px;
font-weight: 600;
color: var(--text);
@@ -731,7 +749,8 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
white-space: nowrap;
}
.project-detail-title-input {
- flex: 1;
+ width: 100%;
+ box-sizing: border-box;
font-size: 22px;
font-weight: 600;
color: var(--text);
@@ -748,13 +767,76 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
gap: 8px;
flex-shrink: 0;
}
+.project-detail-grid {
+ flex: 1;
+ display: grid;
+ grid-template-columns: minmax(0, 1fr) 320px;
+ gap: 28px;
+ align-items: start;
+ min-height: 380px;
+}
+.project-detail-main,
+.project-detail-side {
+ min-width: 0;
+}
+.project-detail-main {
+ display: flex;
+ flex-direction: column;
+ justify-content: flex-start;
+ padding-top: 8px;
+}
+.project-composer-card {
+ width: 100%;
+}
+.project-composer-input-area {
+ pointer-events: auto;
+ display: flex;
+ flex-direction: column;
+ gap: 6px;
+ background: #fff;
+ border: 1px solid var(--primary);
+ border-radius: 16px;
+ padding: 10px 16px;
+ box-shadow: 0 8px 32px rgba(37, 99, 235, 0.08);
+ position: relative;
+ transition: all 0.2s;
+}
+.project-composer-input-area:focus-within {
+ border-color: var(--primary);
+ box-shadow: 0 8px 32px rgba(37, 99, 235, 0.16);
+ transform: translateY(-2px);
+}
+.project-composer-input-area textarea {
+ flex: 0 1 auto;
+ width: 100%;
+ min-height: 62px;
+ max-height: 220px;
+ border: none;
+ outline: none;
+ resize: none;
+ padding: 4px 4px;
+ font-size: 15px;
+ line-height: 1.55;
+ color: var(--text);
+ background: transparent;
+ font-family: inherit;
+}
+.project-composer-input-area .chat-input-mirror {
+ font-size: 15px;
+}
+.project-composer-bottom {
+ min-height: 40px;
+}
+.project-chat-send-btn {
+ margin-left: auto;
+}
/* Card-style section. Mirrors the look in shipping-product references —
white card on neutral page bg, rounded corners, subtle border. */
.project-detail-card {
background: #fff;
border: 1px solid #e5e7eb;
- border-radius: 12px;
+ border-radius: 8px;
padding: 16px 18px;
margin-bottom: 14px;
box-shadow: 0 1px 2px rgba(15, 23, 42, 0.03);
@@ -811,6 +893,8 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
padding: 1px 6px;
border-radius: 4px;
}
+.project-binding-source--custom { color: #1d4ed8; background: #dbeafe; }
+.project-binding-source--marketplace { color: #475569; background: #e2e8f0; }
.project-binding-source--stale { color: #b45309; background: #fef3c7; }
.project-binding-desc {
font-size: 12px;
@@ -900,6 +984,109 @@ body.is-sidebar-resizing iframe { pointer-events: none; }
-webkit-box-orient: vertical;
overflow: hidden;
}
+.project-card-icon-btn {
+ width: 28px;
+ height: 28px;
+ display: inline-flex;
+ align-items: center;
+ justify-content: center;
+ border: none;
+ border-radius: 6px;
+ background: transparent;
+ color: #64748b;
+ font-size: 20px;
+ line-height: 1;
+ cursor: pointer;
+}
+.project-card-icon-btn:hover {
+ background: #f1f5f9;
+ color: var(--text);
+}
+.project-files-status {
+ min-height: 18px;
+ margin-top: 8px;
+ font-size: 12px;
+}
+.project-files-list {
+ display: flex;
+ flex-direction: column;
+ gap: 4px;
+ margin-top: 4px;
+}
+.project-file-row {
+ display: flex;
+ align-items: center;
+ gap: 8px;
+ min-height: 32px;
+ padding: 6px 4px;
+ border-radius: 6px;
+ color: #334155;
+}
+.project-file-row:hover { background: #f8fafc; }
+.project-file-main {
+ flex: 1;
+ min-width: 0;
+ display: flex;
+ align-items: center;
+ gap: 7px;
+ border: none;
+ background: transparent;
+ color: inherit;
+ font-family: inherit;
+ text-align: left;
+ cursor: pointer;
+ padding: 0;
+}
+.project-file-icon {
+ width: 18px;
+ height: 18px;
+ flex: 0 0 auto;
+ display: inline-flex;
+ align-items: center;
+ justify-content: center;
+}
+.project-file-name {
+ min-width: 0;
+ overflow: hidden;
+ text-overflow: ellipsis;
+ white-space: nowrap;
+ font-size: 12.5px;
+}
+.project-file-meta {
+ flex: 0 0 auto;
+ color: #94a3b8;
+ font-size: 11px;
+}
+.project-file-remove {
+ width: 24px;
+ height: 24px;
+ display: inline-flex;
+ align-items: center;
+ justify-content: center;
+ border: none;
+ border-radius: 6px;
+ background: transparent;
+ color: #94a3b8;
+ cursor: pointer;
+ font-size: 16px;
+ line-height: 1;
+}
+.project-file-row:hover .project-file-remove { color: #64748b; }
+.project-file-remove:hover {
+ background: #fee2e2;
+ color: #dc2626;
+}
+
+@media (max-width: 980px) {
+ .project-detail-grid {
+ grid-template-columns: 1fr;
+ min-height: 0;
+ }
+ .project-detail-actions {
+ flex-wrap: wrap;
+ justify-content: flex-end;
+ }
+}
/* Empty-bindings banner inside an in-project conversation. */
.conv-project-empty-banner {
@@ -2304,7 +2491,10 @@ body.model-not-configured .main-content { flex-direction: column; }
color: #64748b;
box-shadow: 0 6px 18px rgba(15, 23, 42, 0.08);
cursor: pointer;
- transition: background 120ms ease, color 120ms ease, border-color 120ms ease, box-shadow 120ms ease;
+ transition: top 120ms ease, background 120ms ease, color 120ms ease, border-color 120ms ease, box-shadow 120ms ease;
+}
+.chat-main-pane.has-project-empty-banner .conversation-info-toggle {
+ top: 62px;
}
.conversation-info-toggle:hover,
.conversation-info-toggle.is-active {
@@ -3342,6 +3532,49 @@ body.model-not-configured .main-content { flex-direction: column; }
gap: 6px;
flex-shrink: 0;
}
+.chat-file-viewer-md-actions {
+ display: flex;
+ align-items: center;
+ gap: 6px;
+}
+.chat-file-viewer-md-actions:empty { display: none; }
+.chat-file-viewer-md-actions .btn {
+ min-height: 28px;
+ white-space: nowrap;
+}
+.chat-file-viewer-md-actions .ctx-viewer-action-icon-btn {
+ width: 28px;
+ height: 28px;
+ min-height: 28px;
+ padding: 0;
+ border-radius: 6px;
+ border: none;
+ background: transparent;
+ color: #475569;
+ display: inline-flex;
+ align-items: center;
+ justify-content: center;
+}
+.chat-file-viewer-md-actions .ctx-viewer-action-icon-btn:hover {
+ background: #e2e8f0;
+ color: #0f172a;
+}
+.chat-file-viewer-md-actions .ctx-viewer-action-icon-btn.btn-primary {
+ background: transparent;
+ color: var(--primary);
+}
+.chat-file-viewer-md-actions .ctx-viewer-action-icon-btn.btn-primary:hover {
+ background: rgba(59, 130, 246, 0.10);
+ color: var(--primary-hover);
+}
+.chat-file-viewer-md-actions .ctx-viewer-action-icon-btn.btn-danger {
+ color: var(--danger);
+}
+.chat-file-viewer-md-actions .ctx-viewer-action-icon {
+ width: 15px;
+ height: 15px;
+ display: block;
+}
.chat-file-viewer-reveal,
.chat-file-viewer-close {
width: 28px;
@@ -3368,6 +3601,11 @@ body.model-not-configured .main-content { flex-direction: column; }
overflow: auto;
background: #fff;
}
+.chat-file-viewer.is-markdown .chat-file-viewer-body {
+ display: flex;
+ flex-direction: column;
+ overflow: hidden;
+}
/* iframe-based body (pdf / html) fills the body slot. The body itself
handles scroll for non-iframe bodies (md / text). */
.chat-file-viewer-pdf,
@@ -3398,6 +3636,53 @@ body.model-not-configured .main-content { flex-direction: column; }
max-width: 880px;
margin: 0 auto;
}
+.chat-file-viewer.is-markdown .chat-file-viewer-body > .ctx-viewer-md {
+ flex: 1 1 auto;
+ width: 100%;
+ max-width: 880px;
+ min-height: 0;
+ margin: 0 auto;
+ padding: 24px 32px;
+ box-sizing: border-box;
+ overflow: auto;
+ color: #0f172a;
+ font-size: 14px;
+ line-height: 1.7;
+}
+.chat-file-viewer.is-markdown .ctx-editor-toolbar {
+ flex: 0 0 auto;
+ margin: 0;
+ padding: 8px 12px;
+ border: 0;
+ border-bottom: 1px solid var(--border);
+ border-radius: 0;
+}
+.chat-file-viewer.is-markdown .ctx-editor-body {
+ flex: 1 1 auto;
+ min-height: 0;
+ display: flex;
+}
+.chat-file-viewer.is-markdown .ctx-viewer-editor {
+ flex: 1 1 auto;
+ min-height: 0;
+ padding: 16px 20px;
+ border: 0;
+ border-radius: 0;
+ outline: 0;
+ resize: none;
+}
+.chat-file-viewer.is-markdown .ctx-editor-preview {
+ flex: 1 1 auto;
+ width: 100%;
+ max-width: 880px;
+ min-height: 0;
+ margin: 0 auto;
+ padding: 24px 32px;
+ border: 0;
+ border-radius: 0;
+ box-sizing: border-box;
+ overflow: auto;
+}
.chat-file-viewer-text {
padding: 20px 24px;
margin: 0;
@@ -3411,6 +3696,28 @@ body.model-not-configured .main-content { flex-direction: column; }
min-height: 100%;
box-sizing: border-box;
}
+/* Mirrors .is-markdown's flex layout: body becomes a flex column so the
+ <pre> view or <textarea> editor (both direct children of the body) can
+ fill the available height and own their own scroll. */
+.chat-file-viewer.is-text .chat-file-viewer-body {
+ display: flex;
+ flex-direction: column;
+ overflow: hidden;
+}
+.chat-file-viewer.is-text .chat-file-viewer-body > .chat-file-viewer-text {
+ flex: 1 1 auto;
+ min-height: 0;
+ overflow: auto;
+}
+.chat-file-viewer.is-text .chat-file-viewer-body > .ctx-viewer-editor {
+ flex: 1 1 auto;
+ min-height: 0;
+ padding: 16px 20px;
+ border: 0;
+ border-radius: 0;
+ outline: 0;
+ resize: none;
+}
.chat-file-viewer-loading {
padding: 40px;
text-align: center;
@@ -7056,10 +7363,9 @@ body.model-not-configured .main-content { flex-direction: column; }
right: 0;
/* Popovers portal to <body> at open time (see _aiSelectMount) so
they can escape parent overflow:auto. They must therefore beat
- every overlay they could appear inside. `.modal-overlay` is 100,
- `.ui-dialog-overlay` is 200, the avatar picker is 300, common
- toasts top out at 1000 — pick a value above all of them. */
- z-index: 1200;
+ every overlay they could appear inside, including the login gate
+ at z-index 10000. */
+ z-index: 12000;
background: #fff;
border: 1px solid var(--border, #e5e7eb);
border-radius: 8px;
diff --git a/test/main/conversation-files.test.ts b/test/main/conversation-files.test.ts
new file mode 100644
index 00000000..d9aa862b
--- /dev/null
+++ b/test/main/conversation-files.test.ts
@@ -0,0 +1,47 @@
+import { describe, it, expect } from 'vitest';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+
+import { listWorkspaceFiles } from '../../src/main/features/conversation_files';
+
+describe('conversation workspace file listing', () => {
+ it('returns a relative tree snapshot from the current workspace on disk', () => {
+ const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-conv-files-'));
+ try {
+ fs.mkdirSync(path.join(dir, 'agent_skill_review'), { recursive: true });
+ fs.writeFileSync(path.join(dir, '处理索引.md'), 'index');
+ fs.writeFileSync(path.join(dir, 'agent_skill_review', 'skill_batch.md'), 'skill');
+ fs.writeFileSync(path.join(dir, '.hidden.md'), 'hidden');
+
+ const result = listWorkspaceFiles(dir);
+
+ expect(result.root).toBe(path.resolve(dir));
+ expect(result.rootExists).toBe(true);
+ expect(result.truncated).toBe(false);
+ expect(result.items.map((f) => f.relPath)).toEqual([
+ 'agent_skill_review/skill_batch.md',
+ '处理索引.md',
+ ]);
+ expect(result.items[0].path).toBe(path.join(dir, 'agent_skill_review', 'skill_batch.md'));
+ } finally {
+ fs.rmSync(dir, { recursive: true, force: true });
+ }
+ });
+
+ it('reports truncation instead of walking forever on huge workspaces', () => {
+ const dir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-conv-files-'));
+ try {
+ fs.writeFileSync(path.join(dir, 'a.md'), 'a');
+ fs.writeFileSync(path.join(dir, 'b.md'), 'b');
+
+ const result = listWorkspaceFiles(dir, { maxFiles: 1 });
+
+ expect(result.items).toHaveLength(1);
+ expect(result.truncated).toBe(true);
+ expect(result.count).toBe(1);
+ } finally {
+ fs.rmSync(dir, { recursive: true, force: true });
+ }
+ });
+});
diff --git a/test/main/features/agents.test.ts b/test/main/features/agents.test.ts
index 2d7a516e..a025abf6 100644
--- a/test/main/features/agents.test.ts
+++ b/test/main/features/agents.test.ts
@@ -104,12 +104,12 @@ describe('agents › normalizeAgent', () => {
const norm = a.normalizeAgent({
agent_id: 'x', name: 'N', description: 'D', workflow: 'W',
created_at: 't1', updated_at: 't2',
- }, 'builtin');
+ }, 'marketplace');
expect(norm).toEqual({
agent_id: 'x', name: 'N',
description_zh: '', description_en: 'D',
workflow: 'W',
- source: 'builtin', created_at: 't1', updated_at: 't2',
+ source: 'marketplace', created_at: 't1', updated_at: 't2',
category: '',
// computed-at-load default; overridden by listAgents/getAgent at the boundary
enabled: true,
@@ -766,7 +766,7 @@ describe('agents › listAgents', () => {
const a = await loadAgents();
const list = await a.listAgents();
const sources = Object.fromEntries(list.map((x) => [x.agent_id, x.source]));
- expect(sources).toEqual({ c1: 'custom', b1: 'builtin' });
+ expect(sources).toEqual({ c1: 'custom', b1: 'marketplace' });
});
it('custom wins when builtin has same id', async () => {
diff --git a/test/main/features/chat_attachments.test.ts b/test/main/features/chat_attachments.test.ts
index 5b146e7f..4e3540a9 100644
--- a/test/main/features/chat_attachments.test.ts
+++ b/test/main/features/chat_attachments.test.ts
@@ -34,9 +34,9 @@ async function loadMod() {
return import('../../../src/main/features/chat_attachments');
}
-async function makePng(): Promise<Buffer> {
+async function makePng(color = 0xAACCEEFF): Promise<Buffer> {
const { Jimp } = await import('jimp' as any);
- const img: any = new Jimp({ width: 200, height: 200, color: 0xAACCEEFF });
+ const img: any = new Jimp({ width: 200, height: 200, color });
return await img.getBuffer('image/png');
}
@@ -142,6 +142,92 @@ describe('chat_attachments › uploadAttachment', () => {
expect(r2.info.name).not.toBe('dup.txt');
expect(r2.info.name.endsWith('.txt')).toBe(true);
});
+
+ it('reuses an existing attachment when upload content hash matches', async () => {
+ const m = await loadMod();
+ const r1 = await m.uploadAttachment(UID, CID, 'first.txt', Buffer.from('same bytes'));
+ const r2 = await m.uploadAttachment(UID, CID, 'second.txt', Buffer.from('same bytes'));
+
+ expect(r1.ok && r2.ok).toBe(true);
+ if (!(r1.ok && r2.ok)) return;
+ expect(r2.info.name).toBe(r1.info.name);
+ expect(r2.reused).toBe(true);
+ expect(fs.readdirSync(attDir()).filter((n) => !n.startsWith('.'))).toEqual(['first.txt']);
+ });
+
+ it('reuses a single attachment when matching uploads arrive concurrently', async () => {
+ const m = await loadMod();
+ const results = await Promise.all([
+ m.uploadAttachment(UID, CID, 'parallel-a.txt', Buffer.from('same parallel bytes')),
+ m.uploadAttachment(UID, CID, 'parallel-b.txt', Buffer.from('same parallel bytes')),
+ m.uploadAttachment(UID, CID, 'parallel-c.txt', Buffer.from('same parallel bytes')),
+ ]);
+
+ expect(results.every((r) => r.ok)).toBe(true);
+ if (!results.every((r) => r.ok)) return;
+ expect(new Set(results.map((r) => r.info.name)).size).toBe(1);
+ expect(results.filter((r) => r.reused).length).toBe(2);
+ expect(fs.readdirSync(attDir()).filter((n) => !n.startsWith('.'))).toEqual(['parallel-a.txt']);
+ });
+
+ it('imports a workspace file by path into the attachment pool', async () => {
+ const m = await loadMod();
+ const source = path.join(tmpDir, 'workspace-note.md');
+ fs.writeFileSync(source, '# hello\n', 'utf8');
+
+ const r = await m.importAttachmentFromPath(UID, CID, source);
+
+ expect(r.ok).toBe(true);
+ if (!r.ok) return;
+ expect(r.info.name).toBe('workspace-note.md');
+ expect(r.info.kind).toBe('text');
+ expect(fs.readFileSync(path.join(attDir(), 'workspace-note.md'), 'utf8')).toBe('# hello\n');
+ });
+
+ it('reuses an existing attachment when imported file hash matches', async () => {
+ const m = await loadMod();
+ const r1 = await m.uploadAttachment(UID, CID, 'kept.md', Buffer.from('same body'));
+ const source = path.join(tmpDir, 'copy.md');
+ fs.writeFileSync(source, 'same body', 'utf8');
+
+ const r2 = await m.importAttachmentFromPath(UID, CID, source);
+
+ expect(r1.ok && r2.ok).toBe(true);
+ if (!(r1.ok && r2.ok)) return;
+ expect(r2.info.name).toBe('kept.md');
+ expect(r2.reused).toBe(true);
+ expect(fs.readdirSync(attDir()).filter((n) => !n.startsWith('.'))).toEqual(['kept.md']);
+ });
+
+ it('reuses a single attachment when matching imports arrive concurrently', async () => {
+ const m = await loadMod();
+ const sourceA = path.join(tmpDir, 'source-a.md');
+ const sourceB = path.join(tmpDir, 'source-b.md');
+ fs.writeFileSync(sourceA, 'same imported body', 'utf8');
+ fs.writeFileSync(sourceB, 'same imported body', 'utf8');
+
+ const results = await Promise.all([
+ m.importAttachmentFromPath(UID, CID, sourceA),
+ m.importAttachmentFromPath(UID, CID, sourceB),
+ ]);
+
+ expect(results.every((r) => r.ok)).toBe(true);
+ if (!results.every((r) => r.ok)) return;
+ expect(new Set(results.map((r) => r.info.name)).size).toBe(1);
+ expect(results.filter((r) => r.reused).length).toBe(1);
+ expect(fs.readdirSync(attDir()).filter((n) => !n.startsWith('.'))).toEqual(['source-a.md']);
+ });
+
+ it('validates text encoding when importing by path', async () => {
+ const m = await loadMod();
+ const source = path.join(tmpDir, 'bad.txt');
+ fs.writeFileSync(source, Buffer.from([0xFF, 0xFE]));
+
+ const r = await m.importAttachmentFromPath(UID, CID, source);
+
+ expect(r.ok).toBe(false);
+ expect(fs.existsSync(path.join(attDir(), 'bad.txt'))).toBe(false);
+ });
});
describe('chat_attachments › listAttachments', () => {
@@ -640,7 +726,7 @@ describe('chat_attachments › buildAttachmentManifest', () => {
it('caps images per message: inlined ones appear in manifest, over-cap ones go to skipped[] only', async () => {
const m = await loadMod();
for (let i = 0; i < 7; i++) {
- await m.uploadAttachment(UID, CID, `img${i}.png`, await makePng());
+ await m.uploadAttachment(UID, CID, `img${i}.png`, await makePng(0x11223300 + i));
}
const r = await m.buildAttachmentManifest(
UID, CID,
diff --git a/test/main/features/connectors/dedupe-google-workspace.test.ts b/test/main/features/connectors/dedupe-google-workspace.test.ts
new file mode 100644
index 00000000..f6f142a3
--- /dev/null
+++ b/test/main/features/connectors/dedupe-google-workspace.test.ts
@@ -0,0 +1,149 @@
+/**
+ * `_dedupeGoogleWorkspaceTools` is the gate that prevents the model from
+ * seeing two routes to the same Gmail / Calendar / Docs / Sheets / Tasks
+ * tool when a user has both the all-in-one `google-workspace` connector
+ * AND a standalone single-service connector installed.
+ *
+ * Why this fixture exists: the dedup criterion was historically a
+ * `description.startsWith('[Gmail]')` match — fragile because the
+ * `[Gmail]` / `[Calendar]` / … prefix is added at runtime in
+ * `bin/google-workspace-mcp-server.cjs` and could be dropped in a future
+ * UI polish. Migrating to a tool-name-set match (the strict invariant —
+ * both backends preserve the underlying adapter's `name`) requires
+ * locking down BOTH directions:
+ *
+ * - set A: a workspace tool whose `name` matches a visible
+ * standalone-service tool MUST be filtered out (regardless
+ * of whether the description still carries a `[Gmail]`
+ * prefix).
+ * - set B: a workspace tool whose `name` does NOT match any
+ * visible standalone tool MUST be preserved — even if its
+ * description happens to start with `[Gmail]` for cosmetic
+ * reasons (the old dedup would have eaten this; the new one
+ * must not).
+ */
+
+import { describe, it, expect } from 'vitest';
+
+import { _dedupeGoogleWorkspaceToolsForTest as dedupe } from '../../../../src/main/features/connectors/tools-adapter';
+
+type FakeInstance = { id: string };
+type FakeTool = { name: string; description: string; inputSchema: object };
+
+function visible(
+ entries: Array<{ id: string; tools: Array<{ name: string; description?: string }> }>,
+): Array<{ instance: any; tools: any[] }> {
+ return entries.map((e) => ({
+ instance: { id: e.id } as FakeInstance,
+ tools: e.tools.map((t) => ({
+ name: t.name,
+ description: t.description || '',
+ inputSchema: { type: 'object', properties: {} },
+ })) as FakeTool[],
+ }));
+}
+
+describe('dedupeGoogleWorkspaceTools › set A (shadowed by name — MUST be filtered)', () => {
+ it('shadows workspace tools whose name matches a visible standalone gmail tool', () => {
+ const input = visible([
+ {
+ id: 'google-workspace',
+ tools: [
+ { name: 'gmail_search', description: '[Gmail] Search messages' },
+ { name: 'gmail_send', description: '[Gmail] Send message' },
+ { name: 'gcal_list', description: '[Calendar] List events' },
+ ],
+ },
+ {
+ id: 'gmail',
+ tools: [
+ { name: 'gmail_search', description: 'Search messages' },
+ { name: 'gmail_send', description: 'Send message' },
+ ],
+ },
+ ]);
+ const out = dedupe(input);
+ const workspace = out.find((v: any) => v.instance.id === 'google-workspace');
+ expect(workspace?.tools.map((t: any) => t.name)).toEqual(['gcal_list']);
+ // standalone gmail untouched
+ const gmail = out.find((v: any) => v.instance.id === 'gmail');
+ expect(gmail?.tools.map((t: any) => t.name)).toEqual(['gmail_search', 'gmail_send']);
+ });
+
+ it('hides the workspace entry entirely when every workspace tool is shadowed', () => {
+ const input = visible([
+ {
+ id: 'google-workspace',
+ tools: [
+ { name: 'gmail_search', description: '[Gmail] Search' },
+ { name: 'gcal_list', description: '[Calendar] List' },
+ ],
+ },
+ { id: 'gmail', tools: [{ name: 'gmail_search', description: 'Search' }] },
+ { id: 'gcal', tools: [{ name: 'gcal_list', description: 'List events' }] },
+ ]);
+ const out = dedupe(input);
+ expect(out.find((v: any) => v.instance.id === 'google-workspace')).toBeUndefined();
+ expect(out.map((v: any) => v.instance.id).sort()).toEqual(['gcal', 'gmail']);
+ });
+
+ it('matches name regardless of description prefix presence (robust to future UI polish that drops [Gmail])', () => {
+ // The workspace tool no longer carries the `[Gmail]` description prefix
+ // (simulating a future UI change). With name-based dedup, the standalone
+ // gmail tool still shadows it.
+ const input = visible([
+ {
+ id: 'google-workspace',
+ tools: [
+ // NB: no `[Gmail]` prefix
+ { name: 'gmail_search', description: 'Search messages' },
+ ],
+ },
+ { id: 'gmail', tools: [{ name: 'gmail_search', description: 'Search' }] },
+ ]);
+ const out = dedupe(input);
+ expect(out.find((v: any) => v.instance.id === 'google-workspace')).toBeUndefined();
+ });
+});
+
+describe('dedupeGoogleWorkspaceTools › set B (NOT shadowed — MUST be preserved)', () => {
+ it('preserves a workspace tool whose description starts with [Gmail] when no standalone gmail connector is visible', () => {
+ // The old description-prefix dedup would have eaten this entry just
+ // because of the cosmetic prefix. The name-based dedup correctly leaves
+ // it alone: no standalone gmail connector → no name shadow → keep.
+ const input = visible([
+ {
+ id: 'google-workspace',
+ tools: [
+ { name: 'gmail_search', description: '[Gmail] Search messages' },
+ { name: 'gcal_list', description: '[Calendar] List events' },
+ ],
+ },
+ // No standalone gmail / gcal visible — workspace is the only route.
+ ]);
+ const out = dedupe(input);
+ const workspace = out.find((v: any) => v.instance.id === 'google-workspace');
+ expect(workspace?.tools.map((t: any) => t.name)).toEqual(['gmail_search', 'gcal_list']);
+ });
+
+ it('does not shadow tools from unrelated visible connectors (e.g. Notion)', () => {
+ const input = visible([
+ {
+ id: 'google-workspace',
+ tools: [{ name: 'gmail_search', description: '[Gmail] Search' }],
+ },
+ // Unrelated connector with a coincidentally-named tool — must NOT
+ // shadow the workspace's tool.
+ { id: 'notion', tools: [{ name: 'gmail_search', description: 'Notion gmail-named tool' }] },
+ ]);
+ const out = dedupe(input);
+ const workspace = out.find((v: any) => v.instance.id === 'google-workspace');
+ expect(workspace?.tools.map((t: any) => t.name)).toEqual(['gmail_search']);
+ });
+
+ it('no-ops when google-workspace is not visible', () => {
+ const input = visible([{ id: 'gmail', tools: [{ name: 'gmail_search' }] }]);
+ const out = dedupe(input);
+ expect(out).toEqual(input);
+ });
+});
diff --git a/test/main/features/group_chat/agents-index.test.ts b/test/main/features/group_chat/agents-index.test.ts
index a97764a7..8275fd17 100644
--- a/test/main/features/group_chat/agents-index.test.ts
+++ b/test/main/features/group_chat/agents-index.test.ts
@@ -75,7 +75,7 @@ describe('agents_index block — header + per-entry shape', () => {
const text = await buildBlock(TEST_UID);
expect(text).toContain('`read_file(<ROOT>/<id>/agent.json)`');
expect(text).toContain(`- custom: ${path.resolve(customAgentsDir())}`);
- expect(text).toContain(`- builtin: ${path.resolve(builtinAgentsDir())}`);
+ expect(text).toContain(`- marketplace: ${path.resolve(builtinAgentsDir())}`);
expect(text).toContain('Use these ROOT values verbatim');
expect(text).toContain('@<name>');
});
diff --git a/test/main/features/group_chat/bus-integration.test.ts b/test/main/features/group_chat/bus-integration.test.ts
index 6879a320..82fbd276 100644
--- a/test/main/features/group_chat/bus-integration.test.ts
+++ b/test/main/features/group_chat/bus-integration.test.ts
@@ -25,16 +25,30 @@ function _setScript(sessionId: string, events: any[]) {
}
function _resetScripts() { _scripts.clear(); }
+const modelAbortMock = vi.hoisted(() => vi.fn(() => 0));
+
vi.mock('../../../../src/main/model/client', () => ({
async *streamChatWithModel(opts: any) {
const sid = opts.sessionId || '';
const queue = _scripts.get(sid) || [];
const events = queue.shift() || [{ type: 'final', text: '' }];
_scripts.set(sid, queue);
- for (const ev of events) yield ev;
+ for (const ev of events) {
+ if (ev?.type === '__wait_for_abort__') {
+ if (!opts.abortSignal?.aborted) {
+ await new Promise<void>((resolve) => {
+ opts.abortSignal?.addEventListener('abort', () => resolve(), { once: true });
+ });
+ }
+ yield { type: 'error', text: 'aborted', aborted: true };
+ continue;
+ }
+ yield ev;
+ }
yield { type: 'done' };
},
async chatWithModel() { return { ok: true, text: '', error: '', aborted: false }; },
+ abortActiveSessionsForConversation: modelAbortMock,
}));
let tmpDir: string;
@@ -52,6 +66,8 @@ beforeEach(async () => {
prevWs = process.env.ORKAS_WORKSPACE_ROOT;
process.env.ORKAS_WORKSPACE_ROOT = tmpDir;
_resetScripts();
+ modelAbortMock.mockClear();
+ modelAbortMock.mockReturnValue(0);
vi.resetModules();
const users = await import('../../../../src/main/features/users');
users.activateUser(TEST_UID);
@@ -211,6 +227,15 @@ describe.skip('group_chat bus integration › multi-recipient parallel dispatch'
});
describe('group_chat bus integration › abort sticky across worker post-cleanup', () => {
+ it('abort also targets active core-agent sessions by conversation id', async () => {
+ const cid = newCid();
+ const bus = await import('../../../../src/main/features/group_chat/bus');
+
+ await bus.abort(TEST_UID, cid);
+
+ expect(modelAbortMock).toHaveBeenCalledWith(cid);
+ });
+
it('abort during a turn keeps state.aborted; subsequent worker reply does NOT un-stick', async () => {
const cid = newCid();
const state = await import('../../../../src/main/features/group_chat/state');
@@ -236,6 +261,44 @@ describe('group_chat bus integration › abort sticky across worker post-cleanup
expect(st.status).toBe('aborted');
});
+ it('abort during a live agent turn propagates to the worker AbortSignal', async () => {
+ const cid = newCid();
+ const state = await import('../../../../src/main/features/group_chat/state');
+ const bus = await import('../../../../src/main/features/group_chat/bus');
+
+ bus.subscribe(TEST_UID, cid, () => {});
+ _setScript(state.buildGmemberSessionId(TEST_UID, cid, AGENT_ID), [
+ { type: '__wait_for_abort__' },
+ { type: 'final', text: 'should not appear after abort' },
+ ]);
+
+ await bus.enqueue({
+ uid: TEST_UID,
+ cid,
+ fromActorId: 'user',
+ text: `@${AGENT_NAME} long task`,
+ });
+
+ const start = Date.now();
+ while (Date.now() - start < 1000) {
+ const live = bus._cidStateForTest(TEST_UID, cid);
+ const worker = live?.workers.get(AGENT_ID);
+ if (worker?.running && worker.abortController) break;
+ await new Promise((r) => setTimeout(r, 10));
+ }
+ expect(bus._cidStateForTest(TEST_UID, cid)?.workers.get(AGENT_ID)?.abortController).toBeTruthy();
+
+ await bus.abort(TEST_UID, cid);
+ await waitForQuiescent(TEST_UID, cid, 2000);
+
+ const paths = await import('../../../../src/main/paths');
+ const storage = await import('../../../../src/main/storage');
+ const messages = await storage.readJsonl<any>(path.join(paths.userChatsDir(TEST_UID), `${cid}.jsonl`));
+ expect(messages.some((m: any) => m.text.includes('should not appear after abort'))).toBe(false);
+ const st = await state.readState(TEST_UID, cid);
+ expect(st.status).toBe('aborted');
+ });
+
it('a NEW user message after abort clears the sticky aborted flag', async () => {
const cid = newCid();
const state = await import('../../../../src/main/features/group_chat/state');
diff --git a/test/main/features/group_chat/bus.test.ts b/test/main/features/group_chat/bus.test.ts
index ca10a636..b9f63319 100644
--- a/test/main/features/group_chat/bus.test.ts
+++ b/test/main/features/group_chat/bus.test.ts
@@ -18,6 +18,7 @@ vi.mock('../../../../src/main/model/client', () => ({
yield { type: 'done' };
},
async chatWithModel() { return { ok: true, text: '', error: '', aborted: false }; },
+ abortActiveSessionsForConversation: vi.fn(() => 0),
}));
const cliRunMock = vi.hoisted(() => ({ calls: [] as any[] }));
diff --git a/test/main/features/group_chat/conv_title.test.ts b/test/main/features/group_chat/conv_title.test.ts
new file mode 100644
index 00000000..03751ecb
--- /dev/null
+++ b/test/main/features/group_chat/conv_title.test.ts
@@ -0,0 +1,66 @@
+/**
+ * conv_title.ts — fixture-lock the placeholder-title detection set so that
+ * adding a new UI language without updating `chat.default_title` in its
+ * locale file (or, worse, forgetting that the placeholder-set even exists)
+ * regresses noisily instead of silently. The bug this guards against
+ * shipped once in 2026-05 when Japanese landed `chat.default_title="新しい会話"`
+ * but the placeholder set was hardcoded to zh+en only — every new Japanese
+ * UI conversation lost its auto-title trigger.
+ *
+ * Set A: every supported language's `chat.default_title` MUST be detected,
+ * plus the historical capitalization aliases.
+ * Set B: a non-placeholder user-named title MUST NOT be detected.
+ */
+
+import { describe, it, expect } from 'vitest';
+
+import { SUPPORTED_LANGS, t } from '../../../../src/main/i18n';
+import { isPlaceholderTitle, PLACEHOLDER_TITLES } from '../../../../src/main/features/group_chat/conv_title';
+
+describe('conv_title › PLACEHOLDER_TITLES set membership', () => {
+ it('contains every supported language\'s chat.default_title (set A: locale-driven)', () => {
+ // Locked: zh / en / ja each have their default title pinned. If a new
+ // language is added but its locale file is missing this key, this test
+ // fails loudly instead of silently degrading auto-title for that UI.
+ for (const lang of SUPPORTED_LANGS) {
+ const value = t('chat.default_title', undefined, lang);
+ expect(value, `chat.default_title missing in ${lang}.json`).not.toBe('chat.default_title');
+ expect(
+ PLACEHOLDER_TITLES.has(value),
+ `${lang} default title "${value}" not recognised as placeholder`,
+ ).toBe(true);
+ expect(isPlaceholderTitle(value)).toBe(true);
+ }
+ });
+
+ it('contains the legacy capitalization aliases "New Conversation" / "New Chat"', () => {
+ expect(PLACEHOLDER_TITLES.has('New Conversation')).toBe(true);
+ expect(PLACEHOLDER_TITLES.has('New Chat')).toBe(true);
+ expect(isPlaceholderTitle('New Conversation')).toBe(true);
+ expect(isPlaceholderTitle('New Chat')).toBe(true);
+ });
+});
+
+describe('conv_title › isPlaceholderTitle (set B: real user-named titles must pass through)', () => {
+ it('returns false for an English user-named title', () => {
+ expect(isPlaceholderTitle('Quarterly review notes')).toBe(false);
+ });
+
+ it('returns false for a Chinese user-named title that is not the default', () => {
+ expect(isPlaceholderTitle('季度复盘')).toBe(false);
+ });
+
+ it('returns false for a Japanese user-named title that is not the default', () => {
+ expect(isPlaceholderTitle('週次レビューのメモ')).toBe(false);
+ });
+
+ it('returns false for a title that contains but does not equal the default ("新对话补充")', () => {
+ expect(isPlaceholderTitle('新对话补充')).toBe(false);
+ });
+
+ it('returns true for empty string / undefined / null (treat as unnamed)', () => {
+ expect(isPlaceholderTitle('')).toBe(true);
+ expect(isPlaceholderTitle(undefined)).toBe(true);
+ expect(isPlaceholderTitle(null)).toBe(true);
+ });
+});
diff --git a/test/main/features/group_chat/plan-recovery.test.ts b/test/main/features/group_chat/plan-recovery.test.ts
index 0d2a8216..04137b91 100644
--- a/test/main/features/group_chat/plan-recovery.test.ts
+++ b/test/main/features/group_chat/plan-recovery.test.ts
@@ -137,37 +137,78 @@ describe('plan recovery › retryStep', () => {
expect(r.error).toMatch(/not in failed state/);
});
- it('treats repeated retry after recovery starts as idempotent', async () => {
+ it('treats repeated retry after recovery starts as idempotent for mid-recovery states (pending / in_progress / blocked)', async () => {
const cid = newCid();
await seedPlan(TEST_UID, cid, {
steps: [
{ title: 'Step 1', assignee: A_NAME, wait_for: [] },
{ title: 'Step 2', assignee: A_NAME },
{ title: 'Step 3', assignee: A_NAME },
- { title: 'Step 4', assignee: A_NAME },
],
});
const plan = await import('../../../../src/main/features/group_chat/plan');
await plan.updateStep(TEST_UID, cid, 1, 'in_progress');
await plan.updateStep(TEST_UID, cid, 2, 'pending');
await plan.updateStep(TEST_UID, cid, 3, 'blocked');
- await plan.updateStep(TEST_UID, cid, 4, 'done');
const planExecutor = await import('../../../../src/main/features/group_chat/plan_executor');
const inProgress = await planExecutor.retryStep(TEST_UID, cid, 1);
const pending = await planExecutor.retryStep(TEST_UID, cid, 2);
const blocked = await planExecutor.retryStep(TEST_UID, cid, 3);
- const done = await planExecutor.retryStep(TEST_UID, cid, 4);
expect(inProgress.ok).toBe(true);
expect(pending.ok).toBe(true);
expect(blocked.ok).toBe(true);
- expect(done.ok).toBe(true);
const fresh = await plan.readPlan(TEST_UID, cid);
expect(fresh?.steps[0].status).toBe('in_progress');
expect(fresh?.steps[1].status).toBe('pending');
expect(fresh?.steps[2].status).toBe('blocked');
- expect(fresh?.steps[3].status).toBe('done');
+ });
+
+ it('clears pending_form_id when a failed user step is retried (so the re-dispatch stamps a fresh form_id)', async () => {
+ const cid = newCid();
+ await seedPlan(TEST_UID, cid, {
+ steps: [{ title: 'ask user', assignee: 'user', input: 'pick one', wait_for: [] }],
+ });
+ const plan = await import('../../../../src/main/features/group_chat/plan');
+ // Simulate a previously-dispatched-then-failed user step: pending_form_id
+ // is set (from the original dispatch), status is failed.
+ await plan.updateStep(TEST_UID, cid, 1, 'failed', {
+ failure_reason: 'user replied with an unrelated message',
+ pending_form_id: 'a1b2c3d4e5f60001',
+ });
+ let fresh = await plan.readPlan(TEST_UID, cid);
+ expect(fresh?.steps[0].pending_form_id).toBe('a1b2c3d4e5f60001');
+
+ const planExecutor = await import('../../../../src/main/features/group_chat/plan_executor');
+ const r = await planExecutor.retryStep(TEST_UID, cid, 1);
+ expect(r.ok).toBe(true);
+
+ fresh = await plan.readPlan(TEST_UID, cid);
+ // Retry re-arms the step and reconcile may have re-dispatched it (which
+ // would stamp a NEW pending_form_id). What we lock here is the
+ // invariant: the stale form_id from the failed run must NOT survive.
+ // Either the field is absent (no re-dispatch yet) OR it differs from
+ // the old one (re-dispatched with a fresh id) — never the old value.
+ expect(fresh?.steps[0].pending_form_id).not.toBe('a1b2c3d4e5f60001');
+ });
+
+ it('rejects retry on a done step (rail never shows retry button on done; reaching here would mis-attribute a retry signal)', async () => {
+ const cid = newCid();
+ await seedPlan(TEST_UID, cid, {
+ steps: [{ title: 'Step 1', assignee: A_NAME, wait_for: [] }],
+ });
+ const plan = await import('../../../../src/main/features/group_chat/plan');
+ await plan.updateStep(TEST_UID, cid, 1, 'done');
+
+ const planExecutor = await import('../../../../src/main/features/group_chat/plan_executor');
+ const r = await planExecutor.retryStep(TEST_UID, cid, 1);
+ expect(r.ok).toBe(false);
+ expect(r.error).toMatch(/not in failed state/);
+
+ // Step remains done — retry must NOT silently mutate a completed step.
+ const fresh = await plan.readPlan(TEST_UID, cid);
+ expect(fresh?.steps[0].status).toBe('done');
});
it('cascade-unblocks downstream skipped via `aborted by step N failure` reason', async () => {
diff --git a/test/main/features/group_chat/plan_executor.test.ts b/test/main/features/group_chat/plan_executor.test.ts
index fe48228c..69b9862f 100644
--- a/test/main/features/group_chat/plan_executor.test.ts
+++ b/test/main/features/group_chat/plan_executor.test.ts
@@ -446,6 +446,11 @@ describe('plan_executor › user-assignee step', () => {
}),
],
});
+ // Dispatch must stamp `pending_form_id` on the step itself so that
+ // `acceptsUserStepCompletion` can match the user reply in O(1) instead
+ // of re-scanning the conversation jsonl. If this regresses, the gate
+ // silently falls back to "accept any user reply" (legacy mode).
+ expect(snapshot?.steps[0].pending_form_id).toBe(question.form.form_id);
// A regular chat message is allowed to start a commander discussion,
// but it must not be mistaken for the pending user-form answer.
diff --git a/test/main/features/reflection-digest.test.ts b/test/main/features/reflection-digest.test.ts
deleted file mode 100644
index f9504bcc..00000000
--- a/test/main/features/reflection-digest.test.ts
+++ /dev/null
@@ -1,259 +0,0 @@
-import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
-import * as fs from 'node:fs';
-import * as os from 'node:os';
-import * as path from 'node:path';
-
-let tmpDir: string;
-let prevWs: string | undefined;
-const TEST_UID = 'u1';
-
-beforeEach(async () => {
- tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-digest-'));
- prevWs = process.env.ORKAS_WORKSPACE_ROOT;
- process.env.ORKAS_WORKSPACE_ROOT = tmpDir;
- vi.resetModules();
- const users = await import('../../../src/main/features/users');
- users.activateUser(TEST_UID);
-});
-
-afterEach(() => {
- process.env.ORKAS_WORKSPACE_ROOT = prevWs;
- fs.rmSync(tmpDir, { recursive: true, force: true });
-});
-
-async function loadModule() {
- return import('../../../src/main/features/reflection-digest');
-}
-
-// ── aggregateSession (pure) ─────────────────────────────────────────────
-
-describe('reflection-digest › aggregateSession', () => {
- it('counts tool_use blocks by name', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([
- { role: 'assistant', content: [{ type: 'tool_use', name: 'web_search', input: {} }], ts: 1 },
- { role: 'assistant', content: [{ type: 'tool_use', name: 'web_search', input: {} }], ts: 2 },
- { role: 'assistant', content: [{ type: 'tool_use', name: 'bash', input: {} }], ts: 3 },
- ], m);
- expect(m.toolCalls).toEqual({ web_search: 2, bash: 1 });
- expect(m.sessionsAnalyzed).toBe(1);
- });
-
- it('counts tool_result errors and captures up to 3 samples', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([
- { role: 'user', content: [{ type: 'tool_result', toolUseId: 'a', content: 'ENETUNREACH connection refused', isError: true }] },
- { role: 'user', content: [{ type: 'tool_result', toolUseId: 'b', content: 'rate limit', isError: true }] },
- { role: 'user', content: [{ type: 'tool_result', toolUseId: 'c', content: 'timeout', isError: true }] },
- { role: 'user', content: [{ type: 'tool_result', toolUseId: 'd', content: 'fourth — should be capped', isError: true }] },
- { role: 'user', content: [{ type: 'tool_result', toolUseId: 'e', content: 'success', isError: false }] },
- ], m);
- expect(m.errorCount).toBe(4);
- expect(m.errorSamples.length).toBe(3); // capped at 3
- expect(m.errorSamples[0]).toContain('ENETUNREACH');
- });
-
- it('truncates long error samples', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- const longErr = 'X'.repeat(500);
- mod.aggregateSession([
- { role: 'user', content: [{ type: 'tool_result', toolUseId: 'a', content: longErr, isError: true }] },
- ], m);
- expect(m.errorSamples[0].length).toBeLessThanOrEqual(160);
- });
-
- it('tracks skill_manage(read) skill ids in skillsLoaded', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([
- { role: 'assistant', content: [{ type: 'tool_use', name: 'skill_manage', input: { action: 'read', id: 'foo' } }] },
- { role: 'assistant', content: [{ type: 'tool_use', name: 'skill_manage', input: { action: 'read', id: 'foo' } }] },
- { role: 'assistant', content: [{ type: 'tool_use', name: 'skill_manage', input: { action: 'create', id: 'bar' } }] },
- ], m);
- expect(m.skillsLoaded).toEqual({ foo: 2 }); // create not tracked, only read
- expect(m.toolCalls.skill_manage).toBe(3);
- });
-
- it('tracks earliestTs and latestTs from message ts fields', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([
- { role: 'user', content: [{ type: 'text', text: 'a' }], ts: 100 },
- { role: 'assistant', content: [{ type: 'text', text: 'b' }], ts: 200 },
- { role: 'user', content: [{ type: 'text', text: 'c' }], ts: 50 },
- ], m);
- expect(m.earliestTs).toBe(50);
- expect(m.latestTs).toBe(200);
- });
-
- it('does not increment sessionsAnalyzed for empty messages array', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([], m);
- expect(m.sessionsAnalyzed).toBe(0);
- });
-
- it('accumulates across multiple aggregate calls (multi-session)', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([{ role: 'assistant', content: [{ type: 'tool_use', name: 'bash' }] }], m);
- mod.aggregateSession([{ role: 'assistant', content: [{ type: 'tool_use', name: 'bash' }] }], m);
- expect(m.sessionsAnalyzed).toBe(2);
- expect(m.toolCalls.bash).toBe(2);
- });
-
- it('handles malformed content blocks defensively', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- mod.aggregateSession([
- { role: 'assistant' }, // missing content
- { role: 'assistant', content: 'not an array' },
- { role: 'assistant', content: [null, undefined, { type: 'tool_use' /* no name */ }] },
- ], m);
- expect(m.sessionsAnalyzed).toBe(1);
- expect(m.toolCalls).toEqual({});
- });
-});
-
-// ── formatDigest (pure) ─────────────────────────────────────────────────
-
-describe('reflection-digest › formatDigest', () => {
- it('returns "no activity" when nothing was analyzed', async () => {
- const mod = await loadModule();
- const out = mod.formatDigest(mod.emptyMetrics(), Date.parse('2026-04-21'), Date.parse('2026-04-23'));
- expect(out).toMatch(/No new conversation activity/);
- });
-
- it('formats a populated digest with all sections', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- m.sessionsAnalyzed = 3;
- m.toolCalls = { bash: 5, web_search: 2 };
- m.errorCount = 2;
- m.errorSamples = ['rate limit', 'timeout'];
- m.skillsLoaded = { 'recover-foo': 3 };
- m.earliestTs = Date.parse('2026-04-21T10:00:00Z');
- m.latestTs = Date.parse('2026-04-23T10:00:00Z');
-
- const out = mod.formatDigest(m, m.earliestTs, Date.now());
- expect(out).toContain('3 session');
- expect(out).toContain('bash: 5');
- expect(out).toContain('web_search: 2');
- expect(out).toContain('Total errors**: 2');
- expect(out).toContain('rate limit');
- expect(out).toContain('recover-foo (3x)');
- });
-
- it('omits error section when no errors', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- m.sessionsAnalyzed = 1;
- m.toolCalls = { bash: 1 };
- const out = mod.formatDigest(m, Date.now() - 1000, Date.now());
- expect(out).not.toContain('Total errors');
- expect(out).not.toContain('Error samples');
- });
-
- it('caps tool list at top-N most-called', async () => {
- const mod = await loadModule();
- const m = mod.emptyMetrics();
- m.sessionsAnalyzed = 1;
- // 12 different tools — should only show top 8
- for (let i = 0; i < 12; i++) m.toolCalls[`tool${i}`] = 100 - i;
- const out = mod.formatDigest(m, Date.now() - 1000, Date.now());
- expect(out).toContain('tool0');
- expect(out).toContain('tool7');
- expect(out).not.toContain('tool8');
- });
-});
-
-// ── buildAgentReflectionDigest (I/O) ────────────────────────────────────
-
-describe('reflection-digest › buildAgentReflectionDigest', () => {
- function writeConv(uid: string, cid: string, agentId: string): string {
- // Write to chats/_index.json (newest-first list).
- const idxPath = path.join(tmpDir, uid, 'cloud', 'chats', '_index.json');
- fs.mkdirSync(path.dirname(idxPath), { recursive: true });
- let list: any[] = [];
- if (fs.existsSync(idxPath)) list = JSON.parse(fs.readFileSync(idxPath, 'utf8'));
- const sessionId = `${uid}-gconv-${cid}`;
- list.unshift({
- conversation_id: cid,
- title: `t-${cid}`,
- kind: agentId ? 'agent_run' : 'normal',
- agent_id: agentId,
- skill_id: '',
- session_id: sessionId,
- created_at: new Date().toISOString(),
- updated_at: new Date().toISOString(),
- });
- fs.writeFileSync(idxPath, JSON.stringify(list));
- return sessionId;
- }
-
- function writeSessionJsonl(uid: string, sessionId: string, lines: any[], mtime?: number): void {
- const file = path.join(tmpDir, uid, 'cloud', 'sessions', `${sessionId}.jsonl`);
- fs.mkdirSync(path.dirname(file), { recursive: true });
- fs.writeFileSync(file, lines.map((l) => JSON.stringify(l)).join('\n') + '\n');
- if (mtime !== undefined) fs.utimesSync(file, mtime / 1000, mtime / 1000);
- }
-
- it('returns "no activity" when no matching conversations exist', async () => {
- const mod = await loadModule();
- const out = await mod.buildAgentReflectionDigest(TEST_UID, '_default', Date.now() - 86400000);
- expect(out).toMatch(/No new conversation activity/);
- });
-
- it('aggregates only conversations matching the target agent_id', async () => {
- const sid1 = writeConv(TEST_UID, 'c1', ''); // normal → _default
- const sid2 = writeConv(TEST_UID, 'c2', 'agent-x'); // agent-x
- writeSessionJsonl(TEST_UID, sid1, [
- { role: 'assistant', content: [{ type: 'tool_use', name: 'bash' }], ts: Date.now() },
- ]);
- writeSessionJsonl(TEST_UID, sid2, [
- { role: 'assistant', content: [{ type: 'tool_use', name: 'web_search' }], ts: Date.now() },
- ]);
-
- const mod = await loadModule();
- const since = Date.now() - 86400000;
-
- const defaultDigest = await mod.buildAgentReflectionDigest(TEST_UID, '_default', since);
- expect(defaultDigest).toContain('bash');
- expect(defaultDigest).not.toContain('web_search');
-
- const xDigest = await mod.buildAgentReflectionDigest(TEST_UID, 'agent-x', since);
- expect(xDigest).toContain('web_search');
- expect(xDigest).not.toContain('bash');
- });
-
- it('skips session jsonls with mtime older than sinceMs', async () => {
- const sid = writeConv(TEST_UID, 'cold', '');
- const ancient = Date.now() - 30 * 86400000; // 30 days ago
- writeSessionJsonl(TEST_UID, sid, [
- { role: 'assistant', content: [{ type: 'tool_use', name: 'bash' }] },
- ], ancient);
-
- const mod = await loadModule();
- const since = Date.now() - 86400000; // last 24h
- const digest = await mod.buildAgentReflectionDigest(TEST_UID, '_default', since);
- expect(digest).toMatch(/No new conversation activity/);
- });
-
- it('tolerates malformed jsonl lines and missing files', async () => {
- const sid = writeConv(TEST_UID, 'malformed', '');
- const file = path.join(tmpDir, TEST_UID, 'cloud', 'sessions', `${sid}.jsonl`);
- fs.mkdirSync(path.dirname(file), { recursive: true });
- fs.writeFileSync(file, 'not json\n{"role":"assistant","content":[{"type":"tool_use","name":"ok"}]}\n}}}\n');
-
- // Also reference a conversation whose session file doesn't exist
- writeConv(TEST_UID, 'ghost', '');
-
- const mod = await loadModule();
- const digest = await mod.buildAgentReflectionDigest(TEST_UID, '_default', 0);
- expect(digest).toContain('ok'); // valid line still parsed
- // Ghost session simply skipped — no throw, and the malformed-but-valid one accounted for.
- });
-});
diff --git a/test/main/features/reflection-orchestrator.test.ts b/test/main/features/reflection-orchestrator.test.ts
new file mode 100644
index 00000000..cde1d50e
--- /dev/null
+++ b/test/main/features/reflection-orchestrator.test.ts
@@ -0,0 +1,216 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+
+// Same race as in reflection-transcript.test.ts: activateUser fires
+// sweepSessions fire-and-forget and would wipe our test session files.
+vi.mock('../../../src/main/features/sessions_sweep', () => ({
+ sweepSessions: vi.fn(async () => ({
+ scanned: 0, orphan_cid: 0, ephemeral_on_cloud: 0, legacy: 0,
+ local_aged_out: 0, errors: 0,
+ })),
+}));
+
+let tmpDir: string;
+let prevWs: string | undefined;
+const TEST_UID = 'u1';
+
+beforeEach(async () => {
+ tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-orchestrator-'));
+ prevWs = process.env.ORKAS_WORKSPACE_ROOT;
+ process.env.ORKAS_WORKSPACE_ROOT = tmpDir;
+ vi.resetModules();
+ const users = await import('../../../src/main/features/users');
+ users.activateUser(TEST_UID);
+});
+
+afterEach(() => {
+ process.env.ORKAS_WORKSPACE_ROOT = prevWs;
+ fs.rmSync(tmpDir, { recursive: true, force: true });
+});
+
+async function loadModule() {
+ return import('../../../src/main/features/reflection-orchestrator');
+}
+
+function writeReflectionState(uid: string, lastReflectedAt: Record<string, string>): void {
+ const dir = path.join(tmpDir, uid, 'local', 'config');
+ fs.mkdirSync(dir, { recursive: true });
+ fs.writeFileSync(path.join(dir, 'reflection-state.json'), JSON.stringify({ lastReflectedAt }));
+}
+
+// ── pickAgentsForCycle ──────────────────────────────────────────────────
+
+describe('reflection-orchestrator › pickAgentsForCycle', () => {
+ const NOW = Date.parse('2026-05-21T12:00:00Z');
+
+ it('skips agents within cooldown (< 4h since lastReflectedAt)', async () => {
+ const mod = await loadModule();
+ const state = { lastReflectedAt: { 'agent-x': new Date(NOW - 2 * 3600 * 1000).toISOString() } };
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ['agent-x'], state, NOW, async () => true);
+ expect(picked.length).toBe(0);
+ });
+
+ it('picks dirty agents past cooldown', async () => {
+ const mod = await loadModule();
+ const state = { lastReflectedAt: { 'agent-x': new Date(NOW - 6 * 3600 * 1000).toISOString() } };
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ['agent-x'], state, NOW, async () => true);
+ expect(picked.length).toBe(1);
+ expect(picked[0].reason).toBe('dirty');
+ });
+
+ it('forces reflection when past 7-day max gap regardless of dirty', async () => {
+ const mod = await loadModule();
+ const stale = new Date(NOW - 10 * 24 * 3600 * 1000).toISOString();
+ const state = { lastReflectedAt: { 'agent-x': stale } };
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ['agent-x'], state, NOW, async () => false);
+ expect(picked.length).toBe(1);
+ expect(picked[0].reason).toBe('max_gap');
+ });
+
+ it('treats never-reflected agents as eligible when dirty (default lookback)', async () => {
+ const mod = await loadModule();
+ const state = { lastReflectedAt: {} };
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ['agent-x'], state, NOW, async () => true);
+ expect(picked.length).toBe(1);
+ expect(picked[0].reason).toBe('never_reflected');
+ // sinceMs should be ~48h before now (DEFAULT_LOOKBACK_MS)
+ expect(NOW - picked[0].sinceMs).toBeCloseTo(48 * 3600 * 1000, -5);
+ });
+
+ it('skips never-reflected agents when not dirty', async () => {
+ const mod = await loadModule();
+ const state = { lastReflectedAt: {} };
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ['agent-x'], state, NOW, async () => false);
+ expect(picked.length).toBe(0);
+ });
+
+ it('caps at MAX_AGENTS_PER_CYCLE, picking earliest lastReflectedAt first', async () => {
+ const mod = await loadModule();
+ const lastReflectedAt: Record<string, string> = {};
+ for (let i = 0; i < 8; i++) {
+ // agent-0 oldest (NOW - 8h), agent-7 newest just past cooldown (NOW - 4.1h)
+ lastReflectedAt[`agent-${i}`] = new Date(NOW - (8 - i * 0.5) * 3600 * 1000).toISOString();
+ }
+ const ids = Object.keys(lastReflectedAt);
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ids, { lastReflectedAt }, NOW, async () => true);
+ expect(picked.length).toBe(mod.MAX_AGENTS_PER_CYCLE);
+ // First in result should be the oldest (agent-0)
+ expect(picked[0].agentId).toBe('agent-0');
+ });
+
+ it('processes a mix: cooldown / dirty / max_gap correctly', async () => {
+ const mod = await loadModule();
+ const state = {
+ lastReflectedAt: {
+ 'cool': new Date(NOW - 1 * 3600 * 1000).toISOString(), // within cooldown
+ 'dirty': new Date(NOW - 6 * 3600 * 1000).toISOString(), // past cooldown, dirty
+ 'stale': new Date(NOW - 10 * 24 * 3600 * 1000).toISOString(), // max_gap
+ 'idle': new Date(NOW - 6 * 3600 * 1000).toISOString(), // past cooldown, not dirty
+ },
+ };
+ const isDirty = async (_u: string, id: string) => id === 'dirty';
+ const picked = await mod.pickAgentsForCycle(TEST_UID, ['cool', 'dirty', 'stale', 'idle'], state, NOW, isDirty);
+ const ids = picked.map((p) => p.agentId).sort();
+ expect(ids).toEqual(['dirty', 'stale']);
+ });
+});
+
+// ── runOneCycle ──────────────────────────────────────────────────────────
+
+describe('reflection-orchestrator › runOneCycle', () => {
+ const NOW = Date.parse('2026-05-21T12:00:00Z');
+
+ it('returns 0 when no agents are eligible', async () => {
+ const mod = await loadModule();
+ const reflect = vi.fn(async () => { /* never called */ });
+ const completed = await mod.runOneCycle(TEST_UID, {
+ now: () => NOW,
+ reflect,
+ isDirty: async () => false,
+ });
+ expect(completed).toBe(0);
+ expect(reflect).not.toHaveBeenCalled();
+ });
+
+ it('failed reflection does not stamp lastReflectedAt (retry next cycle)', async () => {
+ const mod = await loadModule();
+ const reflect = vi.fn(async () => { throw new Error('provider down'); });
+ const completed = await mod.runOneCycle(TEST_UID, {
+ now: () => NOW,
+ reflect,
+ isDirty: async () => true,
+ });
+ expect(reflect).toHaveBeenCalled();
+ expect(completed).toBe(0);
+ const state = mod.readReflectionState(TEST_UID);
+ expect(state.lastReflectedAt).toEqual({}); // no stamp on failure
+ });
+
+ it('successful reflection stamps lastReflectedAt with `now`', async () => {
+ const mod = await loadModule();
+ const reflect = vi.fn(async () => { /* succeed */ });
+ const completed = await mod.runOneCycle(TEST_UID, {
+ now: () => NOW,
+ reflect,
+ isDirty: async () => true,
+ });
+ expect(completed).toBeGreaterThanOrEqual(1);
+ const state = mod.readReflectionState(TEST_UID);
+ expect(state.lastReflectedAt[mod.DEFAULT_AGENT_ID]).toBe(new Date(NOW).toISOString());
+ });
+
+ it('skips with debug log when feature flag is off', async () => {
+ process.env.ORKAS_METACOGNITION = '0';
+ try {
+ const mod = await loadModule();
+ const reflect = vi.fn();
+ const completed = await mod.runOneCycle(TEST_UID, { now: () => NOW, reflect, isDirty: async () => true });
+ expect(completed).toBe(0);
+ expect(reflect).not.toHaveBeenCalled();
+ } finally {
+ delete process.env.ORKAS_METACOGNITION;
+ }
+ });
+
+ it('skips with debug log when uid is empty', async () => {
+ const mod = await loadModule();
+ const reflect = vi.fn();
+ const completed = await mod.runOneCycle('', { now: () => NOW, reflect, isDirty: async () => true });
+ expect(completed).toBe(0);
+ expect(reflect).not.toHaveBeenCalled();
+ });
+});
+
+// ── readReflectionState / writeReflectionState round-trip ───────────────
+
+describe('reflection-orchestrator › state persistence', () => {
+ it('round-trips lastReflectedAt through disk', async () => {
+ const mod = await loadModule();
+ fs.mkdirSync(path.join(tmpDir, TEST_UID, 'local', 'config'), { recursive: true });
+ mod.writeReflectionState(TEST_UID, { lastReflectedAt: { 'agent-x': '2026-05-21T10:00:00Z' } });
+ const read = mod.readReflectionState(TEST_UID);
+ expect(read.lastReflectedAt['agent-x']).toBe('2026-05-21T10:00:00Z');
+ });
+
+ it('returns empty state when file is missing', async () => {
+ const mod = await loadModule();
+ expect(mod.readReflectionState(TEST_UID)).toEqual({ lastReflectedAt: {} });
+ });
+
+ it('returns empty state when file is malformed (defensive)', async () => {
+ const mod = await loadModule();
+ const dir = path.join(tmpDir, TEST_UID, 'local', 'config');
+ fs.mkdirSync(dir, { recursive: true });
+ fs.writeFileSync(path.join(dir, 'reflection-state.json'), '{not json');
+ expect(mod.readReflectionState(TEST_UID)).toEqual({ lastReflectedAt: {} });
+ });
+
+ it('filters out non-string values defensively', async () => {
+ const mod = await loadModule();
+ writeReflectionState(TEST_UID, { 'a': 'iso', /* @ts-expect-error */ 'b': 123 as any });
+ const read = mod.readReflectionState(TEST_UID);
+ expect(read.lastReflectedAt).toEqual({ a: 'iso' });
+ });
+});
diff --git a/test/main/features/reflection-transcript.test.ts b/test/main/features/reflection-transcript.test.ts
new file mode 100644
index 00000000..6a9afe50
--- /dev/null
+++ b/test/main/features/reflection-transcript.test.ts
@@ -0,0 +1,417 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+
+// activateUser fires sweepSessions fire-and-forget; in real boot it races
+// safely with subsequent writes, but in unit tests the sweep can wipe the
+// gmember-*.jsonl files we just created (cid not yet in _index.json when
+// the sweep snapshots cids). Stub it out for the whole file.
+vi.mock('../../../src/main/features/sessions_sweep', () => ({
+ sweepSessions: vi.fn(async () => ({
+ scanned: 0, orphan_cid: 0, ephemeral_on_cloud: 0, legacy: 0,
+ local_aged_out: 0, errors: 0,
+ })),
+}));
+
+let tmpDir: string;
+let prevWs: string | undefined;
+const TEST_UID = 'u1';
+
+beforeEach(async () => {
+ tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-transcript-'));
+ prevWs = process.env.ORKAS_WORKSPACE_ROOT;
+ process.env.ORKAS_WORKSPACE_ROOT = tmpDir;
+ vi.resetModules();
+ const users = await import('../../../src/main/features/users');
+ users.activateUser(TEST_UID);
+});
+
+afterEach(() => {
+ process.env.ORKAS_WORKSPACE_ROOT = prevWs;
+ fs.rmSync(tmpDir, { recursive: true, force: true });
+});
+
+async function loadModule() {
+ return import('../../../src/main/features/reflection-transcript');
+}
+
+// ── Test helpers ────────────────────────────────────────────────────────
+
+interface ConvSpec {
+ cid: string;
+ agentId: string;
+ title?: string;
+ createdAt?: string;
+}
+
+function writeConv(uid: string, spec: ConvSpec): { sessionId: string; gmemberSessionId: string } {
+ const idxPath = path.join(tmpDir, uid, 'cloud', 'chats', '_index.json');
+ fs.mkdirSync(path.dirname(idxPath), { recursive: true });
+ let list: any[] = [];
+ if (fs.existsSync(idxPath)) list = JSON.parse(fs.readFileSync(idxPath, 'utf8'));
+ const sessionId = `gconv-${spec.cid}`;
+ list.unshift({
+ conversation_id: spec.cid,
+ title: spec.title || `t-${spec.cid}`,
+ kind: spec.agentId ? 'agent_run' : 'normal',
+ agent_id: spec.agentId,
+ skill_id: '',
+ session_id: sessionId,
+ created_at: spec.createdAt || new Date().toISOString(),
+ updated_at: new Date().toISOString(),
+ });
+ fs.writeFileSync(idxPath, JSON.stringify(list));
+ return {
+ sessionId,
+ gmemberSessionId: `gmember-${spec.cid}-${spec.agentId}`,
+ };
+}
+
+function writeSessionJsonl(uid: string, sessionId: string, lines: any[]): void {
+ const file = path.join(tmpDir, uid, 'cloud', 'sessions', `${sessionId}.jsonl`);
+ fs.mkdirSync(path.dirname(file), { recursive: true });
+ fs.writeFileSync(file, lines.map((l) => JSON.stringify(l)).join('\n') + '\n');
+}
+
+function userMsg(text: string, ts: number, from = 'user', to = 'commander'): any {
+ return {
+ role: 'user',
+ content: [{ type: 'text', text: `<msg from="${from}" to="${to}">\n${text}\n</msg>` }],
+ ts,
+ };
+}
+
+function agentMsg(text: string, ts: number): any {
+ return {
+ role: 'assistant',
+ content: [{ type: 'text', text }],
+ ts,
+ };
+}
+
+function writeSignalsJsonl(uid: string, signals: any[], date?: Date): void {
+ const ymd = (date || new Date()).toISOString().slice(0, 10);
+ const file = path.join(tmpDir, uid, 'local', 'signals', `${ymd}.jsonl`);
+ fs.mkdirSync(path.dirname(file), { recursive: true });
+ fs.writeFileSync(file, signals.map((s) => JSON.stringify(s)).join('\n') + '\n');
+}
+
+function sig(overrides: any): any {
+ return {
+ id: `sig_${Math.random().toString(36).slice(2)}`,
+ ts: new Date().toISOString(),
+ source: 'event',
+ cid: 'c1',
+ aid: 'agent-x',
+ turn_id: 't1',
+ context_ref: { msg_ids: [] },
+ extractor_version: '[email protected]',
+ ...overrides,
+ };
+}
+
+// ── estimateTokens ──────────────────────────────────────────────────────
+
+describe('reflection-transcript › estimateTokens', () => {
+ it('uses ~4 chars/token for English text', async () => {
+ const mod = await loadModule();
+ // "hello world" = 11 chars / 4 ≈ 3 tokens
+ expect(mod.estimateTokens('hello world')).toBe(3);
+ });
+
+ it('uses ~0.7 token/char for CJK text', async () => {
+ const mod = await loadModule();
+ // "你好世界" 4 chars * 0.7 = 2.8 → 3
+ expect(mod.estimateTokens('你好世界')).toBe(3);
+ });
+
+ it('handles mixed Chinese / English correctly', async () => {
+ const mod = await loadModule();
+ // "hello 世界" = 6 ASCII chars / 4 + 2 CJK * 0.7 = 1.5 + 1.4 = 2.9 → 3
+ expect(mod.estimateTokens('hello 世界')).toBe(3);
+ });
+
+ it('returns 0 for empty string', async () => {
+ const mod = await loadModule();
+ expect(mod.estimateTokens('')).toBe(0);
+ });
+});
+
+// ── parseMsgWrapper / extractors (pure) ─────────────────────────────────
+
+describe('reflection-transcript › parseMsgWrapper', () => {
+ it('extracts from + inner text from a well-formed wrapper', async () => {
+ const { _internals } = await loadModule();
+ const r = _internals.parseMsgWrapper('<msg from="user" to="commander">\nhello\n</msg>');
+ expect(r).toEqual({ from: 'user', inner: 'hello' });
+ });
+
+ it('treats unwrapped text as user (defensive default)', async () => {
+ const { _internals } = await loadModule();
+ const r = _internals.parseMsgWrapper('raw legacy text');
+ expect(r.from).toBe('user');
+ expect(r.inner).toBe('raw legacy text');
+ });
+
+ it('extracts non-user from (e.g. commander dispatch)', async () => {
+ const { _internals } = await loadModule();
+ const r = _internals.parseMsgWrapper('<msg from="commander" to="agent-x">dispatch payload</msg>');
+ expect(r.from).toBe('commander');
+ });
+});
+
+describe('reflection-transcript › extractUserEntries', () => {
+ it('keeps role=user text messages where from=user', async () => {
+ const { _internals } = await loadModule();
+ const entries = _internals.extractUserEntries([
+ userMsg('hello', 100, 'user', 'commander'),
+ userMsg('dispatched payload', 200, 'commander', 'agent-x'), // not user
+ ]);
+ expect(entries.length).toBe(1);
+ expect(entries[0].text).toBe('hello');
+ expect(entries[0].kind).toBe('user');
+ });
+
+ it('skips messages without ts', async () => {
+ const { _internals } = await loadModule();
+ const entries = _internals.extractUserEntries([
+ { role: 'user', content: [{ type: 'text', text: '<msg from="user" to="commander">x</msg>' }] },
+ ]);
+ expect(entries.length).toBe(0);
+ });
+
+ it('drops tool_result content blocks (no false positives)', async () => {
+ const { _internals } = await loadModule();
+ const entries = _internals.extractUserEntries([
+ {
+ role: 'user',
+ content: [{ type: 'tool_result', toolUseId: 'a', content: 'output', isError: false }],
+ ts: 100,
+ },
+ ]);
+ expect(entries.length).toBe(0);
+ });
+});
+
+describe('reflection-transcript › extractAgentEntries', () => {
+ it('keeps role=assistant text only, skipping thinking/tool_use', async () => {
+ const { _internals } = await loadModule();
+ const entries = _internals.extractAgentEntries([
+ {
+ role: 'assistant',
+ ts: 100,
+ content: [
+ { type: 'thinking', thinking: 'noise' },
+ { type: 'tool_use', name: 'bash', input: {} },
+ { type: 'text', text: 'the actual reply' },
+ ],
+ },
+ ]);
+ expect(entries.length).toBe(1);
+ expect(entries[0].text).toBe('the actual reply');
+ });
+
+ it('truncates agent reply to MAX_AGENT_REPLY_CHARS', async () => {
+ const mod = await loadModule();
+ const long = 'X'.repeat(2000);
+ const entries = mod._internals.extractAgentEntries([
+ { role: 'assistant', ts: 100, content: [{ type: 'text', text: long }] },
+ ]);
+ expect(entries.length).toBe(1);
+ expect(entries[0].text.length).toBeLessThanOrEqual(mod.MAX_AGENT_REPLY_CHARS + 20);
+ expect(entries[0].text).toContain('truncated');
+ });
+
+ it('skips messages with only tool_use / thinking (no text)', async () => {
+ const { _internals } = await loadModule();
+ const entries = _internals.extractAgentEntries([
+ { role: 'assistant', ts: 100, content: [{ type: 'tool_use', name: 'bash' }] },
+ ]);
+ expect(entries.length).toBe(0);
+ });
+});
+
+// ── renderSignalEntry ───────────────────────────────────────────────────
+
+describe('reflection-transcript › renderSignalEntry', () => {
+ it('renders retry with step_index', async () => {
+ const { _internals } = await loadModule();
+ const e = _internals.renderSignalEntry(sig({ type: 'retry', metadata: { step_index: 3 } }));
+ expect(e?.kind).toBe('system');
+ expect(e?.text).toContain('retry');
+ expect(e?.text).toContain('step #3');
+ });
+
+ it('renders form_left_blank distinguishing required field', async () => {
+ const { _internals } = await loadModule();
+ const required = _internals.renderSignalEntry(sig({
+ type: 'form_left_blank',
+ metadata: { input_id: 'budget', was_required: true },
+ }));
+ expect(required?.text).toContain('required field');
+ expect(required?.text).toContain('budget');
+
+ const optional = _internals.renderSignalEntry(sig({
+ type: 'form_left_blank',
+ metadata: { input_id: 'note', was_required: false },
+ }));
+ expect(optional?.text).not.toContain('required');
+ });
+
+ it('returns null for non-system signal types', async () => {
+ const { _internals } = await loadModule();
+ expect(_internals.renderSignalEntry(sig({ type: 'correction' as any }))).toBeNull();
+ expect(_internals.renderSignalEntry(sig({ type: 'tool_failure' as any }))).toBeNull();
+ });
+});
+
+// ── buildTranscript (I/O) ───────────────────────────────────────────────
+
+describe('reflection-transcript › buildTranscript', () => {
+ it('returns empty when no matching conversations exist', async () => {
+ const mod = await loadModule();
+ const r = await mod.buildTranscript(TEST_UID, '_default', Date.now() - 86400000);
+ expect(r.text).toBe('');
+ expect(r.stats.convsIncluded).toBe(0);
+ });
+
+ it('joins gconv user msgs with gmember agent replies in time order', async () => {
+ const { sessionId, gmemberSessionId } = writeConv(TEST_UID, { cid: 'c1', agentId: 'agent-x' });
+ writeSessionJsonl(TEST_UID, sessionId, [
+ userMsg('first question', 100, 'user', 'commander'),
+ userMsg('follow up', 300, 'user', 'commander'),
+ ]);
+ writeSessionJsonl(TEST_UID, gmemberSessionId, [
+ agentMsg('first answer', 200),
+ agentMsg('second answer', 400),
+ ]);
+
+ const mod = await loadModule();
+ const r = await mod.buildTranscript(TEST_UID, 'agent-x', 0);
+
+ expect(r.stats.convsIncluded).toBe(1);
+ // Order: 100 user, 200 agent, 300 user, 400 agent
+ const i1 = r.text.indexOf('first question');
+ const i2 = r.text.indexOf('first answer');
+ const i3 = r.text.indexOf('follow up');
+ const i4 = r.text.indexOf('second answer');
+ expect(i1).toBeGreaterThan(0);
+ expect(i1).toBeLessThan(i2);
+ expect(i2).toBeLessThan(i3);
+ expect(i3).toBeLessThan(i4);
+ });
+
+ it('uses gconv assistant for _default agent (no gmember)', async () => {
+ const { sessionId } = writeConv(TEST_UID, { cid: 'c-default', agentId: '' });
+ writeSessionJsonl(TEST_UID, sessionId, [
+ userMsg('q', 100, 'user', 'commander'),
+ agentMsg('commander reply (no gmember)', 200),
+ ]);
+ // Deliberately do NOT write a gmember file.
+
+ const mod = await loadModule();
+ const r = await mod.buildTranscript(TEST_UID, '_default', 0);
+
+ expect(r.stats.convsIncluded).toBe(1);
+ expect(r.text).toContain('commander reply');
+ });
+
+ it('drops conversations whose newest msg predates sinceMs', async () => {
+ const { sessionId, gmemberSessionId } = writeConv(TEST_UID, { cid: 'c-old', agentId: 'agent-x' });
+ writeSessionJsonl(TEST_UID, sessionId, [
+ userMsg('old', 100, 'user', 'commander'),
+ ]);
+ writeSessionJsonl(TEST_UID, gmemberSessionId, [
+ agentMsg('old reply', 200),
+ ]);
+
+ const mod = await loadModule();
+ const r = await mod.buildTranscript(TEST_UID, 'agent-x', 10_000); // sinceMs > all msg ts
+
+ expect(r.stats.convsIncluded).toBe(0);
+ });
+
+ it('caps at MAX_CONVS and drops older convs first', async () => {
+ const mod = await loadModule();
+ // Create 7 convs all for agent-x
+ for (let i = 0; i < 7; i++) {
+ const cid = `c${i}`;
+ const { sessionId, gmemberSessionId } = writeConv(TEST_UID, { cid, agentId: 'agent-x' });
+ const baseTs = 1_000 + i * 1_000; // ascending by index
+ writeSessionJsonl(TEST_UID, sessionId, [userMsg(`q${i}`, baseTs, 'user', 'commander')]);
+ writeSessionJsonl(TEST_UID, gmemberSessionId, [agentMsg(`a${i}`, baseTs + 100)]);
+ }
+
+ const r = await mod.buildTranscript(TEST_UID, 'agent-x', 0);
+ expect(r.stats.convsConsidered).toBe(7);
+ expect(r.stats.convsIncluded).toBe(mod.MAX_CONVS);
+ expect(r.stats.convsTruncated).toBeGreaterThanOrEqual(2);
+
+ // Should have the most recent 5 (c2..c6), drop c0/c1
+ expect(r.text).not.toContain('q0');
+ expect(r.text).not.toContain('q1');
+ expect(r.text).toContain('q6');
+ });
+
+ it('injects retry / skip system events at the right cid + ts', async () => {
+ const { sessionId, gmemberSessionId } = writeConv(TEST_UID, { cid: 'c1', agentId: 'agent-x' });
+ writeSessionJsonl(TEST_UID, sessionId, [userMsg('q', 100, 'user', 'commander')]);
+ writeSessionJsonl(TEST_UID, gmemberSessionId, [agentMsg('a', 200)]);
+
+ const retryTs = new Date(150).toISOString();
+ const skipTs = new Date(250).toISOString();
+ writeSignalsJsonl(TEST_UID, [
+ sig({ type: 'retry', cid: 'c1', aid: 'agent-x', ts: retryTs, metadata: { step_index: 2 } }),
+ sig({ type: 'skip', cid: 'c1', aid: 'agent-x', ts: skipTs, metadata: { step_index: 3 } }),
+ sig({ type: 'correction', cid: 'c1', aid: 'agent-x', ts: retryTs }), // NOT inlined
+ ]);
+
+ const mod = await loadModule();
+ const r = await mod.buildTranscript(TEST_UID, 'agent-x', 0);
+
+ expect(r.text).toContain('retry on step #2');
+ expect(r.text).toContain('skip on step #3');
+ // Correction signal must NOT be inlined (deferred to future critic / weekly review).
+ expect(r.text).not.toMatch(/system event.*correction/);
+ // Time order: q(100) → retry(150) → a(200) → skip(250)
+ const iQ = r.text.indexOf('q\n') >= 0 ? r.text.indexOf('q\n') : r.text.indexOf('q');
+ const iRetry = r.text.indexOf('retry');
+ const iA = r.text.indexOf('a\n') >= 0 ? r.text.indexOf('a\n') : r.text.indexOf(']\na');
+ const iSkip = r.text.indexOf('skip');
+ expect(iQ).toBeLessThan(iRetry);
+ expect(iRetry).toBeLessThan(iA);
+ expect(iA).toBeLessThan(iSkip);
+ });
+
+ it('filters convs by agent_id (different agent gets nothing)', async () => {
+ const a = writeConv(TEST_UID, { cid: 'c1', agentId: 'agent-x' });
+ const b = writeConv(TEST_UID, { cid: 'c2', agentId: 'agent-y' });
+ writeSessionJsonl(TEST_UID, a.sessionId, [userMsg('for x', 100, 'user', 'commander')]);
+ writeSessionJsonl(TEST_UID, a.gmemberSessionId, [agentMsg('x reply', 200)]);
+ writeSessionJsonl(TEST_UID, b.sessionId, [userMsg('for y', 300, 'user', 'commander')]);
+ writeSessionJsonl(TEST_UID, b.gmemberSessionId, [agentMsg('y reply', 400)]);
+
+ const mod = await loadModule();
+ const rx = await mod.buildTranscript(TEST_UID, 'agent-x', 0);
+ expect(rx.text).toContain('for x');
+ expect(rx.text).not.toContain('for y');
+
+ const ry = await mod.buildTranscript(TEST_UID, 'agent-y', 0);
+ expect(ry.text).toContain('for y');
+ expect(ry.text).not.toContain('for x');
+ });
+
+ it('tolerates missing gmember session file (defensive)', async () => {
+ const { sessionId } = writeConv(TEST_UID, { cid: 'c1', agentId: 'agent-x' });
+ writeSessionJsonl(TEST_UID, sessionId, [userMsg('lone q', 100, 'user', 'commander')]);
+ // No gmember file written — agent never responded yet.
+
+ const mod = await loadModule();
+ const r = await mod.buildTranscript(TEST_UID, 'agent-x', 0);
+
+ // User entry still produces a section.
+ expect(r.stats.convsIncluded).toBe(1);
+ expect(r.text).toContain('lone q');
+ });
+});
diff --git a/test/main/features/reflection-trigger.test.ts b/test/main/features/reflection-trigger.test.ts
deleted file mode 100644
index dd4ab949..00000000
--- a/test/main/features/reflection-trigger.test.ts
+++ /dev/null
@@ -1,310 +0,0 @@
-import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
-import * as fs from 'node:fs';
-import * as os from 'node:os';
-import * as path from 'node:path';
-
-let tmpDir: string;
-let prevWs: string | undefined;
-const TEST_UID = 'u1';
-
-function stateFile(uid: string): string {
- return path.join(tmpDir, uid, 'local', 'config', 'reflection-state.json');
-}
-
-function customAgentsDir(uid: string): string {
- return path.join(tmpDir, uid, 'cloud', 'agents');
-}
-
-function writeAgent(uid: string, agentId: string): void {
- // Agent 目录形态:agents/<aid>/agent.json (详见 docs/plans/agent-as-directory.md)
- const aDir = path.join(customAgentsDir(uid), agentId);
- fs.mkdirSync(aDir, { recursive: true });
- fs.writeFileSync(
- path.join(aDir, 'agent.json'),
- JSON.stringify({ agent_id: agentId, name: agentId, description: '', workflow: '', updated_at: new Date().toISOString() }),
- );
-}
-
-beforeEach(async () => {
- tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-reflect-'));
- prevWs = process.env.ORKAS_WORKSPACE_ROOT;
- process.env.ORKAS_WORKSPACE_ROOT = tmpDir;
- vi.resetModules();
- const users = await import('../../../src/main/features/users');
- users.activateUser(TEST_UID);
-});
-
-afterEach(() => {
- process.env.ORKAS_WORKSPACE_ROOT = prevWs;
- fs.rmSync(tmpDir, { recursive: true, force: true });
-});
-
-async function loadModule() {
- return import('../../../src/main/features/reflection-trigger');
-}
-
-// ── pickEligibleAgents (pure) ──────────────────────────────────────────
-
-describe('reflection-trigger › pickEligibleAgents', () => {
- it('returns all agents when state is empty', async () => {
- const mod = await loadModule();
- const eligible = mod.pickEligibleAgents(['a', 'b', '_default'], { lastReflectedAt: {} }, Date.now());
- expect(eligible).toEqual(['a', 'b', '_default']);
- });
-
- it('filters out agents within cooldown', async () => {
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- const oneHourAgo = new Date('2026-04-23T11:00:00Z').toISOString();
- const eligible = mod.pickEligibleAgents(
- ['recent', 'old'],
- { lastReflectedAt: { recent: oneHourAgo } },
- now,
- );
- expect(eligible).toEqual(['old']);
- });
-
- it('keeps agents past cooldown', async () => {
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- const fiftyHoursAgo = new Date('2026-04-21T10:00:00Z').toISOString();
- const eligible = mod.pickEligibleAgents(
- ['ancient'],
- { lastReflectedAt: { ancient: fiftyHoursAgo } },
- now,
- );
- expect(eligible).toEqual(['ancient']);
- });
-
- it('treats invalid timestamps as "never reflected" (defensive)', async () => {
- const mod = await loadModule();
- const eligible = mod.pickEligibleAgents(
- ['broken'],
- { lastReflectedAt: { broken: 'not-a-date' } },
- Date.now(),
- );
- expect(eligible).toEqual(['broken']);
- });
-
- it('honors custom cooldownHours', async () => {
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- const twoHoursAgo = new Date('2026-04-23T10:00:00Z').toISOString();
- // 1-hour cooldown → eligible; 3-hour cooldown → not eligible
- expect(mod.pickEligibleAgents(['x'], { lastReflectedAt: { x: twoHoursAgo } }, now, 1)).toEqual(['x']);
- expect(mod.pickEligibleAgents(['x'], { lastReflectedAt: { x: twoHoursAgo } }, now, 3)).toEqual([]);
- });
-
- it('exact boundary: now - lastReflectedAt === cooldown is eligible', async () => {
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- const exactlyAtCooldown = new Date(now - 48 * 3600 * 1000).toISOString();
- expect(mod.pickEligibleAgents(['x'], { lastReflectedAt: { x: exactlyAtCooldown } }, now)).toEqual(['x']);
- });
-});
-
-// ── readReflectionState / writeReflectionState ─────────────────────────
-
-describe('reflection-trigger › state IO', () => {
- it('reads empty state when file missing', async () => {
- const mod = await loadModule();
- const state = mod.readReflectionState(TEST_UID);
- expect(state).toEqual({ lastReflectedAt: {} });
- });
-
- it('round-trips state through write+read', async () => {
- const mod = await loadModule();
- mod.writeReflectionState(TEST_UID, { lastReflectedAt: { foo: '2026-04-23T00:00:00Z' } });
- expect(fs.existsSync(stateFile(TEST_UID))).toBe(true);
- const state = mod.readReflectionState(TEST_UID);
- expect(state).toEqual({ lastReflectedAt: { foo: '2026-04-23T00:00:00Z' } });
- });
-
- it('treats malformed JSON as empty state (no throw)', async () => {
- const mod = await loadModule();
- // Pre-create the dir then write garbage.
- fs.mkdirSync(path.dirname(stateFile(TEST_UID)), { recursive: true });
- fs.writeFileSync(stateFile(TEST_UID), 'not json {{{');
- const state = mod.readReflectionState(TEST_UID);
- expect(state).toEqual({ lastReflectedAt: {} });
- });
-
- it('treats valid JSON without lastReflectedAt as empty state', async () => {
- const mod = await loadModule();
- fs.mkdirSync(path.dirname(stateFile(TEST_UID)), { recursive: true });
- fs.writeFileSync(stateFile(TEST_UID), JSON.stringify({ random: true }));
- const state = mod.readReflectionState(TEST_UID);
- expect(state).toEqual({ lastReflectedAt: {} });
- });
-
- it('filters non-string timestamp values', async () => {
- const mod = await loadModule();
- fs.mkdirSync(path.dirname(stateFile(TEST_UID)), { recursive: true });
- fs.writeFileSync(stateFile(TEST_UID), JSON.stringify({
- lastReflectedAt: { good: '2026-04-23T00:00:00Z', bad: 12345, also_bad: null },
- }));
- const state = mod.readReflectionState(TEST_UID);
- expect(state.lastReflectedAt).toEqual({ good: '2026-04-23T00:00:00Z' });
- });
-});
-
-// ── runStartupReflections (orchestrator with mocked reflect) ───────────
-
-describe('reflection-trigger › runStartupReflections', () => {
- it('no-ops when uid is empty', async () => {
- const mod = await loadModule();
- const reflect = vi.fn().mockResolvedValue(undefined);
- await mod.runStartupReflections('', { reflect });
- expect(reflect).not.toHaveBeenCalled();
- });
-
- it('reflects for _default plus every custom agent and stamps state', async () => {
- writeAgent(TEST_UID, 'agent-foo');
- writeAgent(TEST_UID, 'agent-bar');
- const mod = await loadModule();
- const reflect = vi.fn().mockResolvedValue(undefined);
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- await mod.runStartupReflections(TEST_UID, { reflect, now: () => now });
-
- // Called for each agent (order: _default first, then customs sorted).
- const calls = reflect.mock.calls.map((c) => c[1] as string);
- expect(calls).toContain('_default');
- expect(calls).toContain('agent-foo');
- expect(calls).toContain('agent-bar');
- expect(calls.length).toBe(3);
-
- // State file stamped for each.
- const state = mod.readReflectionState(TEST_UID);
- expect(state.lastReflectedAt['_default']).toBe(new Date(now).toISOString());
- expect(state.lastReflectedAt['agent-foo']).toBe(new Date(now).toISOString());
- expect(state.lastReflectedAt['agent-bar']).toBe(new Date(now).toISOString());
- });
-
- it('skips agents within cooldown', async () => {
- writeAgent(TEST_UID, 'agent-recent');
- writeAgent(TEST_UID, 'agent-old');
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- // Pre-stamp agent-recent within cooldown, agent-old long ago, _default never.
- mod.writeReflectionState(TEST_UID, {
- lastReflectedAt: {
- 'agent-recent': new Date(now - 1 * 3600 * 1000).toISOString(),
- 'agent-old': new Date(now - 100 * 3600 * 1000).toISOString(),
- },
- });
-
- const reflect = vi.fn().mockResolvedValue(undefined);
- await mod.runStartupReflections(TEST_UID, { reflect, now: () => now });
-
- const calls = reflect.mock.calls.map((c) => c[1]).sort();
- expect(calls).toEqual(['_default', 'agent-old']);
- // recent's timestamp untouched
- const state = mod.readReflectionState(TEST_UID);
- expect(state.lastReflectedAt['agent-recent']).toBe(new Date(now - 1 * 3600 * 1000).toISOString());
- expect(state.lastReflectedAt['agent-old']).toBe(new Date(now).toISOString());
- });
-
- it('does NOT stamp state when reflection throws', async () => {
- writeAgent(TEST_UID, 'agent-fail');
- const mod = await loadModule();
- const reflect = vi.fn(async (_uid: string, agentId: string) => {
- if (agentId === 'agent-fail') throw new Error('llm down');
- });
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- await mod.runStartupReflections(TEST_UID, { reflect, now: () => now });
-
- const state = mod.readReflectionState(TEST_UID);
- expect(state.lastReflectedAt['_default']).toBe(new Date(now).toISOString());
- // Failure → not stamped → eligible again next time.
- expect(state.lastReflectedAt['agent-fail']).toBeUndefined();
- });
-
- it('does not abort the batch when one agent fails', async () => {
- writeAgent(TEST_UID, 'agent-1');
- writeAgent(TEST_UID, 'agent-fail');
- writeAgent(TEST_UID, 'agent-2');
- const mod = await loadModule();
- const reflect = vi.fn(async (_uid: string, agentId: string) => {
- if (agentId === 'agent-fail') throw new Error('boom');
- });
- await mod.runStartupReflections(TEST_UID, { reflect });
- const calls = reflect.mock.calls.map((c) => c[1] as string);
- // All four (including the failing one) attempted.
- expect(calls.length).toBe(4);
- expect(calls).toContain('agent-1');
- expect(calls).toContain('agent-2');
- expect(calls).toContain('agent-fail');
- });
-
- it('per-uid state isolation', async () => {
- const otherUid = 'u2';
- writeAgent(TEST_UID, 'agent-shared-name');
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- // Stamp uid u2 (would skip everything if state were shared).
- mod.writeReflectionState(otherUid, {
- lastReflectedAt: { '_default': new Date(now).toISOString(), 'agent-shared-name': new Date(now).toISOString() },
- });
-
- const reflect = vi.fn().mockResolvedValue(undefined);
- await mod.runStartupReflections(TEST_UID, { reflect, now: () => now });
- // u1 never reflected → all eligible regardless of u2 state.
- expect(reflect.mock.calls.map((c) => c[1])).toContain('_default');
- expect(reflect.mock.calls.map((c) => c[1])).toContain('agent-shared-name');
-
- // u1 state stamped, u2 state untouched at 12:00:00Z (we wrote it ourselves above).
- const u1State = mod.readReflectionState(TEST_UID);
- const u2State = mod.readReflectionState(otherUid);
- expect(u1State.lastReflectedAt['_default']).toBe(new Date(now).toISOString());
- expect(u2State.lastReflectedAt['_default']).toBe(new Date(now).toISOString());
- expect(stateFile(TEST_UID)).not.toBe(stateFile(otherUid));
- });
-
- it('passes sinceMs from prior lastReflectedAt to reflect()', async () => {
- writeAgent(TEST_UID, 'agent-x');
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- const lastIso = new Date(now - 100 * 3600 * 1000).toISOString(); // outside cooldown
- mod.writeReflectionState(TEST_UID, { lastReflectedAt: { 'agent-x': lastIso } });
-
- const reflect = vi.fn().mockResolvedValue(undefined);
- await mod.runStartupReflections(TEST_UID, { reflect, now: () => now });
-
- const agentXCall = reflect.mock.calls.find((c) => c[1] === 'agent-x');
- expect(agentXCall).toBeDefined();
- // Third arg should be the parsed lastIso, NOT the default lookback.
- expect(agentXCall![2]).toBe(Date.parse(lastIso));
- });
-
- it('falls back to default lookback when no prior lastReflectedAt', async () => {
- const mod = await loadModule();
- const now = new Date('2026-04-23T12:00:00Z').getTime();
- const reflect = vi.fn().mockResolvedValue(undefined);
- await mod.runStartupReflections(TEST_UID, { reflect, now: () => now });
-
- const defaultCall = reflect.mock.calls.find((c) => c[1] === '_default');
- expect(defaultCall).toBeDefined();
- // First-ever reflection → 7 day lookback.
- const expectedSince = now - 7 * 24 * 3600 * 1000;
- expect(defaultCall![2]).toBe(expectedSince);
- });
-
- it('runs sequentially (next reflect awaited before previous returns)', async () => {
- writeAgent(TEST_UID, 'a1');
- writeAgent(TEST_UID, 'a2');
- const mod = await loadModule();
-
- const order: string[] = [];
- const reflect = vi.fn(async (_uid: string, agentId: string) => {
- order.push(`start:${agentId}`);
- await new Promise((r) => setTimeout(r, 10));
- order.push(`end:${agentId}`);
- });
-
- await mod.runStartupReflections(TEST_UID, { reflect });
- // Each agent's start/end pair must not interleave with another agent's pair.
- for (let i = 0; i < order.length; i += 2) {
- expect(order[i].replace('start:', '')).toBe(order[i + 1].replace('end:', ''));
- }
- });
-});
diff --git a/test/main/features/skills.test.ts b/test/main/features/skills.test.ts
index 3b23499a..be02443e 100644
--- a/test/main/features/skills.test.ts
+++ b/test/main/features/skills.test.ts
@@ -476,7 +476,7 @@ describe('skills › applySkillContainerFromCommander › edit', () => {
files: [{ path: 'SKILL.md', content: '---\nname: "shipped"\n---\n\ntampered\n' }],
});
expect(r.ok).toBe(false);
- expect(r.error).toMatch(/built-in|内置/i);
+ expect(r.error).toMatch(/marketplace|平台技能|マーケットプレイス/i);
// Original body preserved.
expect(fs.readFileSync(path.join(builtinSkillsDir(), 'shipped', 'SKILL.md'), 'utf8'))
.toContain('body');
@@ -646,7 +646,7 @@ describe('skills › createCustomSkill', () => {
const builtinDir = path.join(builtinSkillsDir(), 'fixed');
fs.mkdirSync(builtinDir, { recursive: true });
const s = await loadSkills();
- await expect(s.createCustomSkill('fixed', '')).rejects.toThrow(/conflicts with a built-in|内置技能冲突/);
+ await expect(s.createCustomSkill('fixed', '')).rejects.toThrow(/conflicts with a marketplace|与平台技能冲突|マーケットプレイスのスキルと競合/);
});
});
diff --git a/test/main/ipc/conversations-send-stream.test.ts b/test/main/ipc/conversations-send-stream.test.ts
new file mode 100644
index 00000000..11a432a0
--- /dev/null
+++ b/test/main/ipc/conversations-send-stream.test.ts
@@ -0,0 +1,162 @@
+import { describe, it, expect, beforeEach, afterEach, vi } from 'vitest';
+import * as fs from 'node:fs';
+import * as os from 'node:os';
+import * as path from 'node:path';
+
+type StreamStartFn = (
+ event: { sender: { isDestroyed: () => boolean; send: (channel: string, payload: unknown) => void } },
+ req: { requestId: string; channel: string; payload?: unknown },
+) => Promise<void>;
+
+let streamStartHandler: StreamStartFn | null = null;
+
+const groupChatMock = vi.hoisted(() => ({
+ subscribers: new Set<(ev: unknown) => void>(),
+ quiescent: false,
+ releaseSend: null as null | (() => void),
+ resolveSendStarted: null as null | (() => void),
+ resolveSendFinished: null as null | (() => void),
+ sendStarted: Promise.resolve(),
+ sendFinished: Promise.resolve(),
+}));
+
+vi.mock('electron', () => ({
+ ipcMain: {
+ handle: vi.fn(),
+ on: (channel: string, fn: StreamStartFn) => {
+ if (channel === 'orkas.streamStart') streamStartHandler = fn;
+ },
+ },
+ shell: { openExternal: vi.fn(async () => undefined), showItemInFolder: vi.fn() },
+ BrowserWindow: { getAllWindows: vi.fn(() => []), getFocusedWindow: vi.fn(() => null) },
+ dialog: { showOpenDialog: vi.fn(async () => ({ canceled: true, filePaths: [] })) },
+}));
+
+vi.mock('../../../src/main/features/group_chat', () => ({
+ subscribeBus: vi.fn((_userId: string, _cid: string, cb: (ev: unknown) => void) => {
+ groupChatMock.subscribers.add(cb);
+ return () => groupChatMock.subscribers.delete(cb);
+ }),
+ send: vi.fn(async () => {
+ groupChatMock.resolveSendStarted?.();
+ await new Promise<void>((resolve) => { groupChatMock.releaseSend = resolve; });
+ groupChatMock.resolveSendFinished?.();
+ return { ok: true };
+ }),
+ busIsQuiescent: vi.fn(() => groupChatMock.quiescent),
+ streamEvents: vi.fn(async function* () {}),
+}));
+
+let tmpDir: string;
+let prevWs: string | undefined;
+const TEST_UID = 'u1';
+
+beforeEach(async () => {
+ tmpDir = fs.mkdtempSync(path.join(os.tmpdir(), 'orkas-send-stream-ipc-'));
+ prevWs = process.env.ORKAS_WORKSPACE_ROOT;
+ process.env.ORKAS_WORKSPACE_ROOT = tmpDir;
+ streamStartHandler = null;
+ groupChatMock.subscribers.clear();
+ groupChatMock.quiescent = false;
+ groupChatMock.releaseSend = null;
+ groupChatMock.sendStarted = new Promise<void>((resolve) => { groupChatMock.resolveSendStarted = resolve; });
+ groupChatMock.sendFinished = new Promise<void>((resolve) => { groupChatMock.resolveSendFinished = resolve; });
+ vi.resetModules();
+
+ const users = await import('../../../src/main/features/users');
+ users.activateUser(TEST_UID);
+ const ipc = await import('../../../src/main/ipc/index');
+ ipc.register();
+});
+
+afterEach(() => {
+ process.env.ORKAS_WORKSPACE_ROOT = prevWs;
+ fs.rmSync(tmpDir, { recursive: true, force: true });
+});
+
+async function waitFor(predicate: () => boolean, timeoutMs = 500): Promise<void> {
+ const started = Date.now();
+ while (Date.now() - started < timeoutMs) {
+ if (predicate()) return;
+ await new Promise((resolve) => setTimeout(resolve, 5));
+ }
+ expect(predicate()).toBe(true);
+}
+
+describe('ipc › conversations.sendStream', () => {
+ it('relays group bus events before groupChat.send resolves', async () => {
+ if (!streamStartHandler) throw new Error('stream handler not registered');
+ const sent: Array<{ channel: string; payload: any }> = [];
+ const sender = {
+ isDestroyed: () => false,
+ send: (channel: string, payload: unknown) => sent.push({ channel, payload }),
+ };
+
+ const run = streamStartHandler(
+ { sender },
+ {
+ requestId: 'req1',
+ channel: 'conversations.sendStream',
+ payload: { cid: 'c123abc', content: 'go' },
+ },
+ );
+ await groupChatMock.sendStarted;
+
+ const liveEvent = {
+ type: 'process',
+ cid: 'c123abc',
+ actor: 'agent1',
+ data: { type: 'delta', text: 'live' },
+ };
+ for (const cb of groupChatMock.subscribers) cb(liveEvent);
+
+ await waitFor(() => sent.some((item) => item.channel === 'stream:req1' && item.payload?.event?.data === liveEvent));
+ expect(sent.some((item) => item.payload?.type === 'done')).toBe(false);
+
+ groupChatMock.quiescent = true;
+ groupChatMock.releaseSend?.();
+ await run;
+
+ expect(sent.at(-1)).toEqual({ channel: 'stream:req1', payload: { type: 'done' } });
+ });
+
+ it('keeps relaying group bus events after groupChat.send resolves while the bus is still active', async () => {
+ if (!streamStartHandler) throw new Error('stream handler not registered');
+ const sent: Array<{ channel: string; payload: any }> = [];
+ const sender = {
+ isDestroyed: () => false,
+ send: (channel: string, payload: unknown) => sent.push({ channel, payload }),
+ };
+
+ const run = streamStartHandler(
+ { sender },
+ {
+ requestId: 'req2',
+ channel: 'conversations.sendStream',
+ payload: { cid: 'c123abc', content: 'go' },
+ },
+ );
+ await groupChatMock.sendStarted;
+ groupChatMock.quiescent = false;
+ groupChatMock.releaseSend?.();
+ await groupChatMock.sendFinished;
+
+ await new Promise((resolve) => setTimeout(resolve, 20));
+ expect(sent.some((item) => item.payload?.type === 'done')).toBe(false);
+
+ const liveEvent = {
+ type: 'process',
+ cid: 'c123abc',
+ actor: 'agent1',
+ data: { type: 'event', event: { stream: 'tool', data: { phase: 'start', id: 't1', name: 'web_search' } } },
+ };
+ for (const cb of groupChatMock.subscribers) cb(liveEvent);
+
+ await waitFor(() => sent.some((item) => item.channel === 'stream:req2' && item.payload?.event?.data === liveEvent));
+ groupChatMock.quiescent = true;
+ for (const cb of groupChatMock.subscribers) cb({ type: 'state_changed', cid: 'c123abc', state: { status: 'idle', in_flight: [] } });
+ await run;
+
+ expect(sent.at(-1)).toEqual({ channel: 'stream:req2', payload: { type: 'done' } });
+ });
+});
diff --git a/test/main/ipc/produced-readText.test.ts b/test/main/ipc/produced-readText.test.ts
index ff65ff5b..bca22e79 100644
--- a/test/main/ipc/produced-readText.test.ts
+++ b/test/main/ipc/produced-readText.test.ts
@@ -39,20 +39,22 @@ async function callReadText(
const userWorkspace = await import('../../../src/main/features/user_workspace');
const { chatAttachmentDir } = await import('../../../src/main/paths');
const { safeId } = await import('../../../src/main/storage');
+ const { isPathAllowed } = await import('../../../src/main/util/path-sandbox');
const target = input?.path;
if (typeof target !== 'string' || !target) return { ok: false, error: 'missing path' };
const ws = userWorkspace.getWorkspacePath(userId);
const norm = path.resolve(target);
- const wsNorm = path.resolve(ws);
const cidRaw = input?.cid;
const attachmentScope = (typeof cidRaw === 'string' && cidRaw && safeId(cidRaw))
? path.resolve(chatAttachmentDir(userId, cidRaw))
: null;
- const inAttachment = !!attachmentScope &&
- (norm === attachmentScope || norm.startsWith(attachmentScope + path.sep));
- const inWorkspace = norm === wsNorm || norm.startsWith(wsNorm + path.sep);
- if (!inWorkspace && !inAttachment) {
+ // Mirrors `ipc/index.ts::produced.readText`: delegate the symlink-safe
+ // sandbox check to `util/path-sandbox.isPathAllowed`. Drift here = drift
+ // from the real handler.
+ const allowedRoots: string[] = [path.resolve(ws)];
+ if (attachmentScope) allowedRoots.push(attachmentScope);
+ if (!isPathAllowed(norm, allowedRoots)) {
return { ok: false, error: 'path is outside the user workspace' };
}
let st: fs.Stats;
diff --git a/test/main/ipc/workspace-reveal.test.ts b/test/main/ipc/workspace-reveal.test.ts
index 8b881ee6..58b8ebed 100644
--- a/test/main/ipc/workspace-reveal.test.ts
+++ b/test/main/ipc/workspace-reveal.test.ts
@@ -42,16 +42,19 @@ afterEach(() => {
async function callRevealPath(userId: string, input: unknown): Promise<{ ok: boolean; error?: string; path?: string }> {
// Bypass the full IPC runtime — call the feature directly with the same
- // guard logic the handler uses. This mirrors the handler body in ipc/index.ts.
+ // guard logic the handler uses. Mirrors `ipc/index.ts::workspace.revealPath`:
+ // it delegates the symlink-safe sandbox check to `util/path-sandbox.isPathAllowed`
+ // (via the `_ipcFileSandboxAllowedRoots` helper); the test reproduction tracks
+ // that — drift here = drift from the real handler.
const userWorkspace = await import('../../../src/main/features/user_workspace');
+ const { isPathAllowed } = await import('../../../src/main/util/path-sandbox');
const { shell } = await import('electron');
const p = (input as { path?: unknown })?.path;
if (!p || typeof p !== 'string') return { ok: false, error: 'missing path' };
const abs = path.resolve(p);
const wsRoot = path.resolve(userWorkspace.getWorkspacePath(userId));
- const rel = path.relative(wsRoot, abs);
- if (rel === '' || rel.startsWith('..') || path.isAbsolute(rel)) {
+ if (!isPathAllowed(abs, [wsRoot])) {
return { ok: false, error: 'path is outside the current workspace' };
}
if (!fs.existsSync(abs)) return { ok: false, error: 'file not found' };
@@ -107,6 +110,50 @@ describe('workspace.revealPath › validation', () => {
expect(showItemInFolder).not.toHaveBeenCalled();
});
+ it('rejects a symlink inside the workspace that points outside (symlink-escape)', async () => {
+ // The bug class the isPathAllowed migration closes: a symlink planted
+ // inside an allowed root (here, the user's workspace) that resolves to
+ // a target OUTSIDE the allowed root. The previous lexical
+ // `startsWith(wsRoot + sep)` check would let this through because the
+ // symlink's textual path IS inside the workspace; `isPathAllowed` calls
+ // `fs.realpathSync` on both sides and the rejection happens because the
+ // resolved target sits outside the root.
+ const ws = await import('../../../src/main/features/user_workspace');
+ const wsDir = path.join(tmpDir, 'ws-sym');
+ fs.mkdirSync(wsDir, { recursive: true });
+ ws.setWorkspacePath('u-sym', wsDir);
+
+ const outside = path.join(tmpDir, 'outside-target.txt');
+ fs.writeFileSync(outside, 'attacker-controlled');
+
+ const trap = path.join(wsDir, 'looks-inside.txt');
+ fs.symlinkSync(outside, trap);
+
+ const res = await callRevealPath('u-sym', { path: trap });
+ expect(res.ok).toBe(false);
+ expect(res.error).toMatch(/outside/);
+ expect(showItemInFolder).not.toHaveBeenCalled();
+ });
+
+ it('accepts a symlink inside the workspace that points to another file inside the workspace', async () => {
+ // The companion preservation case: a symlink whose target is also
+ // within the allowed root MUST be accepted (otherwise `isPathAllowed`
+ // would break legitimate `ln -s` use inside the workspace).
+ const ws = await import('../../../src/main/features/user_workspace');
+ const wsDir = path.join(tmpDir, 'ws-sym-ok');
+ fs.mkdirSync(wsDir, { recursive: true });
+ ws.setWorkspacePath('u-sym-ok', wsDir);
+
+ const target = path.join(wsDir, 'real.pdf');
+ fs.writeFileSync(target, '%PDF');
+ const link = path.join(wsDir, 'link.pdf');
+ fs.symlinkSync(target, link);
+
+ const res = await callRevealPath('u-sym-ok', { path: link });
+ expect(res.ok).toBe(true);
+ expect(showItemInFolder).toHaveBeenCalledWith(link);
+ });
+
it('rejects a non-string / missing path', async () => {
const res1 = await callRevealPath('u1', {});
expect(res1.ok).toBe(false);
@@ -128,16 +175,20 @@ describe('workspace.revealPath › validation', () => {
expect(showItemInFolder).not.toHaveBeenCalled();
});
- it('rejects the workspace root itself (empty relative path)', async () => {
- // `path.relative(root, root)` === '' — the guard should treat this as
- // "there's no file to reveal here".
+ it('accepts the workspace root itself (revealing the workspace folder is a legit action)', async () => {
+ // `isPathAllowed(root, [root])` returns true by design — revealing the
+ // workspace root in Finder/Explorer is a legitimate "open my workspace"
+ // request. The pre-refactor lexical guard also allowed this
+ // (`norm === wsNorm` short-circuits the `&& norm !== wsNorm` arm in the
+ // OR chain), so the behaviour is preserved across the isPathAllowed
+ // migration.
const ws = await import('../../../src/main/features/user_workspace');
const dir = path.join(tmpDir, 'ws');
fs.mkdirSync(dir, { recursive: true });
ws.setWorkspacePath('u5', dir);
const res = await callRevealPath('u5', { path: dir });
- expect(res.ok).toBe(false);
- expect(res.error).toMatch(/outside/);
+ expect(res.ok).toBe(true);
+ expect(showItemInFolder).toHaveBeenCalledWith(dir);
});
});
diff --git a/test/main/logger.test.ts b/test/main/logger.test.ts
index b022e85a..a0884254 100644
--- a/test/main/logger.test.ts
+++ b/test/main/logger.test.ts
@@ -58,6 +58,27 @@ describe('logger › redact', () => {
expect(out.list[1].token).toBe('***REDACTED***');
});
+ it('masks PII field names (phone / mobile / email / username) but leaves name passthrough', async () => {
+ const { redact } = await import('../../src/main/logger');
+ const out = redact({
+ phone: '13800138000',
+ mobile: '+8613800138000',
+ email: '[email protected]',
+ username: 'alice',
+ // `name` is intentionally NOT a redact key — too broad (agent.name etc).
+ name: 'Alpha Agent',
+ // PII-shaped fields nested in an array.
+ contacts: [{ email: '[email protected]', phone: '13900139000' }],
+ }) as any;
+ expect(out.phone).toBe('***REDACTED***');
+ expect(out.mobile).toBe('***REDACTED***');
+ expect(out.email).toBe('***REDACTED***');
+ expect(out.username).toBe('***REDACTED***');
+ expect(out.name).toBe('Alpha Agent');
+ expect(out.contacts[0].email).toBe('***REDACTED***');
+ expect(out.contacts[0].phone).toBe('***REDACTED***');
+ });
+
it('is case-insensitive on the key name', async () => {
const { redact } = await import('../../src/main/logger');
const out = redact({
diff --git a/test/main/util/log-sanitize.test.ts b/test/main/util/log-sanitize.test.ts
new file mode 100644
index 00000000..a7ed6846
--- /dev/null
+++ b/test/main/util/log-sanitize.test.ts
@@ -0,0 +1,110 @@
+/**
+ * `sanitizeLogTextForUpload` — fixtures pin both the shapes that MUST be
+ * masked (set A) and the look-alike shapes that MUST be preserved (set B).
+ * Follows PC/CLAUDE.md §9 "Hard rule for LLM-output text munging" — the
+ * same rule applies to any text-processing util whose accepted-input matrix
+ * could be widened over time.
+ */
+
+import { describe, it, expect } from 'vitest';
+import { sanitizeLogTextForUpload } from '../../../src/main/util/log-sanitize';
+
+describe('sanitizeLogTextForUpload › set A (must be masked)', () => {
+ it('masks Bearer token from raw Authorization header', () => {
+ const out = sanitizeLogTextForUpload('Authorization: Bearer abc.def-123_XYZ');
+ expect(out).toBe('Authorization: Bearer ***');
+ });
+
+ it('masks Bearer token inside a JSON-stringified blob', () => {
+ const out = sanitizeLogTextForUpload('{"authorization":"Bearer eyJhbGc"}');
+ expect(out).toBe('{"authorization":"Bearer ***"}');
+ });
+
+ it('is case-insensitive on the Bearer scheme name (RFC 6750)', () => {
+ const out = sanitizeLogTextForUpload('Cookie: bearer x.y.z');
+ expect(out).toBe('Cookie: Bearer ***');
+ });
+
+ it('masks a JWT-shaped 3-segment token', () => {
+ const jwt = 'eyJhbGciOiJIUzI1NiJ9.eyJzdWIiOiIxMjMifQ.signature-part';
+ const out = sanitizeLogTextForUpload(`token=${jwt} expired`);
+ expect(out).toBe('token=***JWT*** expired');
+ });
+
+ it('masks email to first-char + domain (Server mask_email parity)', () => {
+ const out = sanitizeLogTextForUpload('user contacted [email protected] today');
+ expect(out).toBe('user contacted a***@orkas.ai today');
+ });
+
+ it('masks CN mobile to 138****8000 (Server mask_phone parity)', () => {
+ const out = sanitizeLogTextForUpload('phone reported as 13800138000 in audit');
+ expect(out).toBe('phone reported as 138****8000 in audit');
+ });
+
+ it('applies multiple patterns in a single pass', () => {
+ const out = sanitizeLogTextForUpload(
+ 'Authorization: Bearer X; contact: [email protected]; phone 13800138000',
+ );
+ expect(out).toBe('Authorization: Bearer ***; contact: a***@b.io; phone 138****8000');
+ });
+});
+
+describe('sanitizeLogTextForUpload › set B (must be preserved)', () => {
+ it('leaves "BearerWord" (no whitespace after Bearer) untouched', () => {
+ // word-boundary follower required — "Bearer" without trailing whitespace
+ // is just a word in prose, not an auth header.
+ const out = sanitizeLogTextForUpload('class BearerWord extends Auth');
+ expect(out).toBe('class BearerWord extends Auth');
+ });
+
+ it('leaves a single base64url-looking segment (no JWT structure) untouched', () => {
+ // "eyJfoo" by itself is not a JWT — needs three dot-separated segments.
+ const out = sanitizeLogTextForUpload('payload prefix eyJfooBar after');
+ expect(out).toBe('payload prefix eyJfooBar after');
+ });
+
+ it('leaves a malformed email (1-char TLD) untouched', () => {
+ const out = sanitizeLogTextForUpload('placeholder [email protected] user note');
+ expect(out).toBe('placeholder [email protected] user note');
+ });
+
+ it('leaves a 10-digit number (not a CN mobile) untouched', () => {
+ const out = sanitizeLogTextForUpload('order id 1380013800 retrieved');
+ expect(out).toBe('order id 1380013800 retrieved');
+ });
+
+ it('leaves a 12-digit number (too long for a CN mobile) untouched', () => {
+ const out = sanitizeLogTextForUpload('measurement 138001380001 ms');
+ expect(out).toBe('measurement 138001380001 ms');
+ });
+
+ it('leaves a SHA-like 40-hex string untouched (avoids commit-sha false positives)', () => {
+ const sha = 'd24a12b33252f4c2f0b8c4c8cb790b610268bce7104dc531';
+ const out = sanitizeLogTextForUpload(`merge base ${sha} resolved`);
+ expect(out).toBe(`merge base ${sha} resolved`);
+ });
+
+ it('leaves the word "phone" or "email" in prose untouched (positional value-only matchers)', () => {
+ // The redact() layer handles named field MASKING for keys phone/email/
+ // username via REDACT_KEYS; this util only matches value-shaped text.
+ const out = sanitizeLogTextForUpload('field "phone" missing in payload');
+ expect(out).toBe('field "phone" missing in payload');
+ });
+});
+
+describe('sanitizeLogTextForUpload › edge cases', () => {
+ it('returns empty input unchanged', () => {
+ expect(sanitizeLogTextForUpload('')).toBe('');
+ });
+
+ it('handles multi-line text', () => {
+ const input = 'line 1: Bearer XYZ\nline 2: [email protected]\nline 3: 13800138000';
+ const out = sanitizeLogTextForUpload(input);
+ expect(out).toBe('line 1: Bearer ***\nline 2: a***@b.io\nline 3: 138****8000');
+ });
+
+ it('masks every occurrence in a single string (global flag)', () => {
+ const out = sanitizeLogTextForUpload('Bearer A then Bearer B');
+ expect(out).toBe('Bearer *** then Bearer ***');
+ });
+});
diff --git a/test/renderer/auto-title-parity.test.ts b/test/renderer/auto-title-parity.test.ts
new file mode 100644
index 00000000..a01dea4f
--- /dev/null
+++ b/test/renderer/auto-title-parity.test.ts
@@ -0,0 +1,89 @@
+/**
+ * Lock the regex/constant parity between
+ * `src/main/util/auto-title.ts` (backend persistence path —
+ * imported by features/chats.ts)
+ * `src/renderer/modules/auto-title.js` (renderer optimistic title path —
+ * loaded as a `<script>` and
+ * consumed via the `_autoTitle`
+ * global by agents.js /
+ * conversation.js / project-detail.js)
+ *
+ * The two copies exist because PC/CLAUDE.md §8 forbids `import` / `export`
+ * in renderer files — they have to mirror by hand. This fixture is the
+ * machinery that turns drift into a loud CI failure: adding a filler word
+ * to one side without the other immediately fails the test.
+ *
+ * Also includes a functional fixture set that runs the renderer
+ * `_autoTitle` on the same inputs `chats-autotitle.test.ts` pins for the
+ * backend `autoTitle`, asserting both surfaces produce the same stripped
+ * output. (The backend wraps the empty-result fall-through with
+ * `t('chat.default_title')`; the renderer returns '' and lets the caller
+ * fall back. The functional comparison runs on the pre-fallback transform
+ * only — we slice the comparison to inputs whose stripped form is
+ * non-empty, since the empty-input fallback is locale-dependent on the
+ * backend side and out of scope for the parity check.)
+ */
+
+import { describe, it, expect } from 'vitest';
+
+import * as mainAutoTitle from '../../src/main/util/auto-title';
+// eslint-disable-next-line @typescript-eslint/no-var-requires
+const rendererAutoTitle = require('../../src/renderer/modules/auto-title.js');
+import { autoTitle as backendAutoTitle } from '../../src/main/features/chats';
+
+const renderer: {
+ _autoTitle: (text: string) => string;
+ _AUTO_TITLE_ZH_FILLER: RegExp;
+ _AUTO_TITLE_EN_FILLER: RegExp;
+ _AUTO_TITLE_CLAUSE: RegExp;
+ _AUTO_TITLE_MAX: number;
+} = rendererAutoTitle;
+
+describe('auto-title parity › regex source / flags match across main + renderer', () => {
+ it('ZH_FILLER_RE.source matches', () => {
+ expect(renderer._AUTO_TITLE_ZH_FILLER.source).toBe(mainAutoTitle.ZH_FILLER_RE.source);
+ });
+
+ it('ZH_FILLER_RE.flags match', () => {
+ expect(renderer._AUTO_TITLE_ZH_FILLER.flags).toBe(mainAutoTitle.ZH_FILLER_RE.flags);
+ });
+
+ it('EN_FILLER_RE.source matches', () => {
+ expect(renderer._AUTO_TITLE_EN_FILLER.source).toBe(mainAutoTitle.EN_FILLER_RE.source);
+ });
+
+ it('EN_FILLER_RE.flags match (case-insensitive on the EN list)', () => {
+ expect(renderer._AUTO_TITLE_EN_FILLER.flags).toBe(mainAutoTitle.EN_FILLER_RE.flags);
+ });
+
+ it('CLAUSE_RE.source matches', () => {
+ expect(renderer._AUTO_TITLE_CLAUSE.source).toBe(mainAutoTitle.CLAUSE_RE.source);
+ });
+
+ it('TITLE_MAX matches', () => {
+ expect(renderer._AUTO_TITLE_MAX).toBe(mainAutoTitle.TITLE_MAX);
+ });
+});
+
+describe('auto-title parity › functional equivalence on representative inputs', () => {
+ // Shared inputs whose stripped form is non-empty (so the renderer ''
+ // fallback and the backend `t()` fallback don't diverge by design).
+ const cases = [
+ '看下本地还有哪些修改没提交',
+ '请帮我修复一下这个 bug',
+ '想问问这个怎么实现',
+ '请帮我看下数据库连接',
+ 'Can you help me debug this?',
+ 'Please review this PR',
+ 'I want to refactor the auth flow',
+ '帮我写一段 TypeScript 来解析 csv,谢谢',
+ '这是一个很长的对话标题,应该会被三十字符的上限截断到刚好显示',
+ 'AI,怎么样', // < 4-char clause floor; first clause not used
+ ];
+
+ for (const input of cases) {
+ it(`backend autoTitle and renderer _autoTitle agree on: ${JSON.stringify(input.slice(0, 30))}`, () => {
+ expect(backendAutoTitle(input)).toBe(renderer._autoTitle(input));
+ });
+ }
+});
diff --git a/test/renderer/conversation-info.test.ts b/test/renderer/conversation-info.test.ts
new file mode 100644
index 00000000..512c8791
--- /dev/null
+++ b/test/renderer/conversation-info.test.ts
@@ -0,0 +1,243 @@
+import { describe, it, expect } from 'vitest';
+import * as fs from 'node:fs';
+import * as path from 'node:path';
+import * as vm from 'node:vm';
+
+function renderTaskHtml(snapshot: {
+ history: any[];
+ plan: any;
+ members: any[];
+}): Promise<string> {
+ const elements = new Map<string, any>();
+ const getEl = (id: string) => {
+ if (!elements.has(id)) {
+ elements.set(id, {
+ id,
+ hidden: false,
+ innerHTML: '',
+ dataset: {},
+ classList: { toggle() {}, add() {}, remove() {} },
+ setAttribute() {},
+ addEventListener(type: string, fn: () => void) { this[`on${type}`] = fn; },
+ });
+ }
+ return elements.get(id);
+ };
+
+ const context: any = {
+ console,
+ setTimeout,
+ clearTimeout,
+ encodeURIComponent,
+ Date,
+ Map,
+ Array,
+ String,
+ Number,
+ RegExp,
+ createLogger: () => ({ warn() {}, info() {}, error() {} }),
+ t: (key: string) => ({
+ 'conversation_info.status.pending': 'Pending',
+ 'conversation_info.status.in_progress': 'Running',
+ 'conversation_info.status.done': 'Done',
+ 'chat.recipient_commander': 'Commander',
+ 'chat.from_user': 'User',
+ }[key] || key),
+ escapeHtml: (s: unknown) => String(s ?? '').replace(/[&<>"]/g, (c) => ({
+ '&': '&amp;',
+ '<': '&lt;',
+ '>': '&gt;',
+ '"': '&quot;',
+ }[c] || c)),
+ conversations: [{ conversation_id: 'c1', title: 'Current title' }],
+ apiFetch: async (url: string) => ({
+ json: async () => {
+ if (url.includes('/history')) return { ok: true, conversation: { title: 'Current title' }, history: snapshot.history };
+ if (url.includes('/plan')) return { ok: true, plan: snapshot.plan };
+ if (url.includes('/members')) return { ok: true, actors: snapshot.members };
+ if (url.includes('/attachments')) return { ok: true, items: [] };
+ return { ok: false, error: 'unknown' };
+ },
+ }),
+ document: {
+ readyState: 'complete',
+ getElementById: getEl,
+ querySelectorAll: () => [],
+ addEventListener() {},
+ },
+ window: {
+ addEventListener() {},
+ uiIconHtml: (name: string) => `[${name}]`,
+ fileKindIconHtml: () => '',
+ },
+ };
+ context.window.window = context.window;
+ vm.createContext(context);
+ const source = fs.readFileSync(path.join(__dirname, '../../src/renderer/modules/conversation-info.js'), 'utf8');
+ vm.runInContext(source, context);
+ context.window.ConversationInfo.bind('c1');
+ getEl('conversation-info-toggle').onclick();
+ return new Promise((resolve) => setTimeout(() => resolve(getEl('conversation-info-body').innerHTML), 0));
+}
+
+function statusTexts(html: string): string[] {
+ return Array.from(html.matchAll(/conversation-info-status[^>]*>([^<]+)/g)).map((m) => m[1]);
+}
+
+function renderFilesHtml(snapshot: {
+ history: any[];
+ plan?: any;
+ files: any;
+}): Promise<string> {
+ const elements = new Map<string, any>();
+ const getEl = (id: string) => {
+ if (!elements.has(id)) {
+ elements.set(id, {
+ id,
+ hidden: false,
+ innerHTML: '',
+ dataset: {},
+ classList: { toggle() {}, add() {}, remove() {} },
+ setAttribute() {},
+ addEventListener(type: string, fn: () => void) { this[`on${type}`] = fn; },
+ });
+ }
+ return elements.get(id);
+ };
+ const tabs = [
+ { dataset: { infoTab: 'tasks' }, classList: { toggle() {} }, addEventListener(type: string, fn: () => void) { (this as any)[`on${type}`] = fn; } },
+ { dataset: { infoTab: 'files' }, classList: { toggle() {} }, addEventListener(type: string, fn: () => void) { (this as any)[`on${type}`] = fn; } },
+ { dataset: { infoTab: 'attachments' }, classList: { toggle() {} }, addEventListener(type: string, fn: () => void) { (this as any)[`on${type}`] = fn; } },
+ ];
+
+ const context: any = {
+ console,
+ setTimeout,
+ clearTimeout,
+ encodeURIComponent,
+ Date,
+ Map,
+ Array,
+ String,
+ Number,
+ RegExp,
+ createLogger: () => ({ warn() {}, info() {}, error() {} }),
+ t: (key: string) => key,
+ escapeHtml: (s: unknown) => String(s ?? '').replace(/[&<>"]/g, (c) => ({
+ '&': '&amp;',
+ '<': '&lt;',
+ '>': '&gt;',
+ '"': '&quot;',
+ }[c] || c)),
+ conversations: [{ conversation_id: 'c1', title: 'Current title' }],
+ apiFetch: async (url: string) => ({
+ json: async () => {
+ if (url.includes('/history')) return { ok: true, conversation: { title: 'Current title' }, history: snapshot.history };
+ if (url.includes('/plan')) return { ok: true, plan: snapshot.plan || null };
+ if (url.includes('/members')) return { ok: true, actors: [] };
+ if (url.includes('/files')) return { ok: true, ...snapshot.files };
+ if (url.includes('/attachments')) return { ok: true, items: [] };
+ return { ok: false, error: 'unknown' };
+ },
+ }),
+ document: {
+ readyState: 'complete',
+ getElementById: getEl,
+ querySelectorAll: () => tabs,
+ addEventListener() {},
+ },
+ window: {
+ addEventListener() {},
+ uiIconHtml: (name: string) => `[${name}]`,
+ fileKindIconHtml: () => '',
+ },
+ };
+ context.window.window = context.window;
+ vm.createContext(context);
+ const source = fs.readFileSync(path.join(__dirname, '../../src/renderer/modules/conversation-info.js'), 'utf8');
+ vm.runInContext(source, context);
+ context.window.ConversationInfo.bind('c1');
+ (tabs[1] as any).onclick();
+ getEl('conversation-info-toggle').onclick();
+ return new Promise((resolve) => setTimeout(() => resolve(getEl('conversation-info-body').innerHTML), 0));
+}
+
+describe('ConversationInfo task status recovery', () => {
+ it('infers completed statuses for older plan-announcement cards after a re-plan', async () => {
+ const history = [
+ { id: 'u1', from: 'user', text: 'old request' },
+ {
+ id: 'p1',
+ from: 'commander',
+ to: ['user'],
+ plan_announcement: true,
+ ts: '2026-01-01T00:00:00',
+ text: 'Plan\n\n1. Collect candidates(@Researcher)\n2. Analyze candidates(@Analyst)\n3. Summarize(我自己)',
+ },
+ { id: 'd1', from: 'commander', to: ['a1'], dispatch: true, text: '@Researcher go' },
+ { id: 'd2', from: 'commander', to: ['a2'], dispatch: true, text: '@Analyst go' },
+ { id: 'a1out', from: 'a1', to: ['user'], text: 'collected' },
+ { id: 'a2out', from: 'a2', to: ['user'], text: 'analyzed' },
+ { id: 'cmdout', from: 'commander', to: ['user'], text: 'summary' },
+ { id: 'u2', from: 'user', text: 'new request' },
+ {
+ id: 'p2',
+ from: 'commander',
+ to: ['user'],
+ plan_announcement: true,
+ ts: '2026-01-02T00:00:00',
+ text: 'Plan\n\n1. New first(我自己)\n2. New second(@Analyst)',
+ },
+ ];
+ const plan = {
+ initial_message: 'new request',
+ updated_at: '2026-01-02T00:10:00',
+ steps: [
+ { index: 1, title: 'New first', assignee: 'commander', status: 'done' },
+ { index: 2, title: 'New second', assignee: 'Analyst', status: 'in_progress' },
+ ],
+ };
+ const members = [
+ { id: 'commander', name: 'Commander', kind: 'commander' },
+ { id: 'user', name: 'User', kind: 'user' },
+ { id: 'a1', name: 'Researcher', kind: 'agent' },
+ { id: 'a2', name: 'Analyst', kind: 'agent' },
+ ];
+
+ const html = await renderTaskHtml({ history, plan, members });
+
+ expect(statusTexts(html)).toEqual(['Done', 'Running', 'Done', 'Done', 'Done']);
+ });
+});
+
+describe('ConversationInfo files tab', () => {
+ it('renders the live workspace file listing and drops stale produced files under that root', async () => {
+ const html = await renderFilesHtml({
+ history: [
+ { produced: ['/tmp/workspace/deleted.md', '/tmp/outside.md'] },
+ ],
+ files: {
+ root: '/tmp/workspace',
+ rootExists: true,
+ truncated: false,
+ count: 1,
+ items: [
+ {
+ path: '/tmp/workspace/batch/skill_large-batch.md',
+ relPath: 'batch/skill_large-batch.md',
+ name: 'skill_large-batch.md',
+ bytes: 12,
+ mtime: 1700000000000,
+ },
+ ],
+ },
+ });
+
+ expect(html).toContain('batch');
+ expect(html).toContain('skill_large-batch.md');
+ expect(html).toContain('/tmp/outside.md');
+ expect(html).not.toContain('deleted.md');
+ expect(html).toContain('draggable="true"');
+ expect(html).not.toMatch(/<details[^>]*\sopen(?:\s|>|=)/);
+ });
+});
diff --git a/test/renderer/strip-structural-blocks.test.ts b/test/renderer/strip-structural-blocks.test.ts
index c284dd45..1078c111 100644
--- a/test/renderer/strip-structural-blocks.test.ts
+++ b/test/renderer/strip-structural-blocks.test.ts
@@ -26,6 +26,15 @@
// Mixed: real container + literal mention coexisting in one buffer — only
// the real container goes, the literal mention stays put.
//
+// Note on ```xml fences: starting `bc6e0156`, ```xml is treated as a
+// **structural** fence (its body IS the LLM's actual tag emission), not as
+// a preservation marker. The earlier set B1 wording "inside a ```xml block
+// must be preserved" is intentionally retired — the positive coverage for
+// the new semantic lives in the "explicit XML fences are treated as
+// structural blocks" describe blocks further down. ```xml as an example
+// container preservation marker is not coming back; LLM teaching surfaces
+// should use plain ``` for examples.
+//
// Adding a new guard / branch to any of these strippers? Add a fixture
// for the motivating shape AND keep the existing fixtures green. Don't
// rely on "I checked it manually" — that's the loop these tests exist
@@ -48,6 +57,7 @@ const {
_replaceKnownSkillIdsForDisplay,
_simplifyKnownSkillFollowPhrasesForDisplay,
_normalizeKnownSkillRefsForDisplay,
+ _collapseRepeatedStructuralPlaceholders,
} = strip as {
_splitMarkdownProseCode: (text: string) => Array<{ kind: 'prose' | 'code'; text: string; xmlFence?: boolean }>;
_findOuterTagRanges: (text: string, tagName: string) => Array<[number, number]>;
@@ -62,6 +72,7 @@ const {
_replaceKnownSkillIdsForDisplay: (text: string, skills: Array<{ id: string; name?: string }>) => string;
_simplifyKnownSkillFollowPhrasesForDisplay: (text: string, skills: Array<{ id: string; name?: string }>) => string;
_normalizeKnownSkillRefsForDisplay: (text: string, skills: Array<{ id: string; name?: string }>) => string;
+ _collapseRepeatedStructuralPlaceholders: (buf: string, placeholder: string) => string;
};
const PH = '⟨PLACEHOLDER⟩';
@@ -347,14 +358,15 @@ describe('<agent-input-submission> — explicit XML fences are structural', () =
describe('main-chat creation stream stripping', () => {
function streamDisplay(buf: string): string {
- return _stripSkillCreateContainer(
+ const skillPh = '[organizing skill config]';
+ return _collapseRepeatedStructuralPlaceholders(_stripSkillCreateContainer(
_replaceOuterTagBlocks(
- strip._replaceOuterSkillFileBlocks(buf, (path: string) => `\n[writing ${path || 'file'}]\n`),
+ strip._replaceOuterSkillFileBlocks(buf, () => `\n${skillPh}\n`),
'agent',
PH,
),
- PH,
- );
+ `\n${skillPh}\n`,
+ ), `\n${skillPh}\n`);
}
it('does not leak skill frontmatter or agent XML while bulk create output streams', () => {
@@ -387,7 +399,32 @@ describe('main-chat creation stream stripping', () => {
const display = streamDisplay(raw.slice(0, i));
expect(display).not.toMatch(/name: leio|description_zh|description_en|把研发工作组织|Organize engineering|# leio-sdlc-adapter|<workflow>|Plandex研发助手/);
}
- expect(streamDisplay(raw)).toContain('[writing SKILL.md]');
+ expect(streamDisplay(raw)).toContain('[organizing skill config]');
+ });
+
+ it('shows one skill placeholder for consecutive skill-file blocks', () => {
+ const raw = [
+ '<skill>',
+ '<<<skill-file path=SKILL.md',
+ '---',
+ 'name: first',
+ '>>>',
+ '<<<skill-file path=scripts/a.py',
+ 'print("a")',
+ '>>>',
+ '</skill>',
+ ].join('\n');
+ const display = streamDisplay(raw);
+ expect(display.split('[organizing skill config]').length - 1).toBe(1);
+ });
+});
+
+describe('stream placeholders', () => {
+ it('collapses consecutive repeated structural placeholders', () => {
+ expect(_collapseRepeatedStructuralPlaceholders(`lead\n${PH}\n\n${PH}\ntrail`, PH))
+ .toBe(`lead\n${PH}\ntrail`);
+ expect(_collapseRepeatedStructuralPlaceholders(`lead\n${PH}\nkeep\n${PH}`, PH))
+ .toBe(`lead\n${PH}\nkeep\n${PH}`);
});
});
@@ -478,6 +515,69 @@ describe('<artifact-result> — explicit XML fences are structural', () => {
});
});
+// `<marketplace-install-result>` — the machine tag the commander composes
+// when an install request resolves (accepted / declined / failed). Carried
+// in a user message so the next commander turn can read it; stripped from
+// the user bubble's display via `_stripUserStructuralBlocksForDisplay`.
+// Same whack-a-mole class as `<agent-input-submission>` / `<artifact-result>`
+// — added to `_stripSurvivingStructuralBlocks` allow-list in `2f325f90`
+// without fixtures, retroactively pinned here. Real shape (from
+// `group_chat/index.ts::resolveMarketplaceInstallRequest`):
+// `<marketplace-install-result request_id="..." kind="..." id="..." status="..."> body </marketplace-install-result>`
+describe('<marketplace-install-result> — set A (real result tags stripped from display)', () => {
+ it('A1. result tag with the 4 standard attributes + body', () => {
+ const buf = 'Confirming install:\n\n<marketplace-install-result request_id="r-1" kind="agent" id="abcd1234" status="accepted">\nuser approved\n</marketplace-install-result>';
+ expect(_stripOuterTagBlocks(buf, 'marketplace-install-result')).toBe('Confirming install:\n\n');
+ });
+
+ it('A2. body contains backticks / angle brackets — still atomic', () => {
+ const buf = 'pre\n\n<marketplace-install-result request_id="r" kind="skill" id="x" status="declined">\nreason: user said "use `<existing>` instead"\n</marketplace-install-result>\npost';
+ const out = _stripOuterTagBlocks(buf, 'marketplace-install-result');
+ expect(out).not.toContain('marketplace-install-result');
+ expect(out).not.toContain('`<existing>`');
+ expect(out).toBe('pre\n\n\npost');
+ });
+});
+
+describe('<marketplace-install-result> — set B (literal mentions survive)', () => {
+ it('B1. inside a fenced non-XML code block (e.g. docs quoting the protocol)', () => {
+ const buf = 'Wire format:\n```\n<marketplace-install-result request_id="X" kind="agent" id="Y" status="accepted">\n...\n</marketplace-install-result>\n```';
+ expect(_stripOuterTagBlocks(buf, 'marketplace-install-result')).toBe(buf);
+ });
+
+ it('B2. inside an inline backtick span', () => {
+ const buf = 'The commander posts a `<marketplace-install-result ...>` tag when the install resolves.';
+ expect(_stripOuterTagBlocks(buf, 'marketplace-install-result')).toBe(buf);
+ });
+
+ it('B3. UNCLOSED inline backtick (mid-stream code-explanation)', () => {
+ const buf = 'see `<marketplace-install-result';
+ expect(_stripOuterTagBlocks(buf, 'marketplace-install-result')).toBe(buf);
+ });
+});
+
+describe('<marketplace-install-result> — explicit XML fences are structural', () => {
+ it('strips a result block inside ```xml (parity with <artifact-result>)', () => {
+ const buf = 'Reply:\n```xml\n<marketplace-install-result request_id="X" kind="agent" id="Y" status="accepted">\nok\n</marketplace-install-result>\n```';
+ expect(_stripOuterTagBlocks(buf, 'marketplace-install-result')).toBe('Reply:\n```xml\n\n```');
+ });
+});
+
+describe('<marketplace-install-result> — round-trip through the two strip callers', () => {
+ it('_stripSurvivingStructuralBlocks removes assistant-text hallucinations of the tag', () => {
+ const buf = 'Sure, here is the install reply you asked for: <marketplace-install-result request_id="r" kind="skill" id="x" status="accepted">echo</marketplace-install-result> done.';
+ const out = _stripSurvivingStructuralBlocks(buf);
+ expect(out).not.toContain('marketplace-install-result');
+ expect(out).toContain('Sure, here is the install reply you asked for:');
+ expect(out).toContain('done.');
+ });
+
+ it('_stripUserStructuralBlocksForDisplay removes the user-emitted tag from the bubble', () => {
+ const buf = 'I accept.\n\n<marketplace-install-result request_id="r" kind="agent" id="x" status="accepted">ok</marketplace-install-result>';
+ expect(_stripUserStructuralBlocksForDisplay(buf)).toBe('I accept.');
+ });
+});
+
// --- Boundary check: tagName lookup must not bleed across siblings -------
describe('boundary check — `<agent>` lookup must not match `<agent-input-form>`', () => {