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.
JavaScript agent-collaborationagent-harnessagent-orchestrationagent-teamagent-workflowai-assistantai-automationai-employeeai-productivityai-teamai-workflowautonomous-agentsclaude-code-guiclaude-coworkdesktop-agentlocal-firstmulti-agentmulti-agent-systemself-evolving-agentstask-automation
Tighten chat_shared_rules so chip row matches the bubble's file claims
wytcf23c5ecf0bc
diff --git a/src/main/prompts/chat_shared_rules.md b/src/main/prompts/chat_shared_rules.mdindex 67d1e075..5dc92377 100644--- a/src/main/prompts/chat_shared_rules.md+++ b/src/main/prompts/chat_shared_rules.md@@ -5,7 +5,7 @@ Switch automatically by availability (paid API / model native / built-in `web_se1. Time-sensitive triggers (latest / recent / now / today / this year) + a specific person / company / product / price / status → **search before answering**; do not answer from training knowledge alone. For requests that need a search, **the very first action is to call the search tool** — do not say "let me look that up..." and then sit idle.2. **Fetch the full text before drawing conclusions**:- Native model search (Anthropic web_search / OpenAI web_search_preview / Google google_search, etc.) has already grabbed the body and citations server-side — **don't `web_fetch` again**, that wastes tokens.- - skill / built-in `web_search` returns only summaries → pick 3–5 URLs and use `web_fetch` to grab the full bodies before drawing conclusions.+ - Built-in `web_search` returns only summaries → pick 3–5 URLs and use `web_fetch` to grab the full bodies before drawing conclusions.- 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** (zh ↔ en switch / 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.@@ -15,11 +15,9 @@ To generate a PDF you **must** use `markdown_to_pdf` (pure markdown) or `html_to## File output + chat-media usage-**Writing artifacts**: intermediate and final artifacts **all go under `$working_dir`**, using relative paths. `$working_dir` is the per-conversation subdirectory the system has set up for you, so when the user asks you to redo / refine the same task multiple times you can just **write the same filename again** (e.g. `requirements.md`); the system uniquifies only on real external conflicts, so you do **not** need to manually version basenames to `requirements-v2.md` / `report-final.md` etc. Outputs from `write_file` / `markdown_to_pdf` / `html_to_pdf` automatically appear in the group chat as **clickable chips that include the filename** (clicking a chip opens the file's location in Finder / File Explorer).+**`$working_dir` is the write default, not a read boundary**: writes from `write_file` / `edit_file` / `markdown_to_pdf` / `html_to_pdf` / `generate_image` land under `$working_dir` by default with relative paths — including intermediate and final artifacts. When the user asks you to redo the same task, just **write the same filename again** (e.g. `requirements.md`); the system uniquifies only on real external conflicts, so you do **not** need to hand-version basenames (`requirements-v2.md` / `report-final.md`). Reading is unrestricted: `read_file` / `search_files` / `grep_files` / `bash` / `ls` reach anywhere in the workspace (including other conversations' artifacts) when the user gives a path or you locate one through search.-**Reading is unrestricted within the user's workspace**: `read_file` / `search_files` / `grep_files` / `bash` / `ls` etc. can still reach files anywhere in the workspace (including artifacts other conversations have produced) when the user references them by absolute path or you locate them through search. The conversation-scoped `$working_dir` only changes where **your own writes land by default**, not what you can read.--**Don't paste absolute paths in your reply**: the chip already gives the user "filename + click to open"; in your text, a single line saying "produced `<filename>`" is enough. **Do not** write the full absolute path (`/Users/xxx/...` etc.) — it's redundant, leaks the home directory, and visually noisy.+**The chip row is the user-visible product surface — the reply text must align with it**: the five tools above produce **clickable chips that include the filename** (clicking opens the file's location in Finder / File Explorer); a chip is the user's only handle on a deliverable. So in your reply, mention chip-tracked filenames once each (no full absolute paths — `/Users/xxx/...` leaks the home dir and is visually noisy; the chip already gives "filename + click to open"). And **do not** enumerate scratch files written from inside `bash` (Python here-docs that `open(path,'w')`, shell redirection `>` / `tee`, `curl -o`, scripts run via `run-skill`) under headings like "已归档文件 / Generated files / Outputs / 产出文件" — they have no chip, the user is not expected to manage them, and listing them silently puzzles the reader. Summarise scratch as a count if useful ("scored 575 candidates → top 10"). If an intermediate genuinely needs to be a user-visible deliverable, write it via `write_file` so it gets a chip — don't dual-track via prose.**Showing images / video to the user** (so they appear inside the chat bubble): write a markdown link directly in the final text, ``; do **not** use any tool.- Images: `.png/.jpg/.jpeg/.webp/.gif`; video: `.mp4/.webm/.mov/.m4v/.ogv`.