Solution for long term memory for agent coding CLIs and to facilitate handoff between different agent vendors
Rust
docs: clarify Codex shell hook payload (#974)
yubr-zzh691e3cc31701
diff --git a/CHANGELOG.md b/CHANGELOG.mdindex 53454e5e..c376849a 100644--- a/CHANGELOG.md+++ b/CHANGELOG.md@@ -36,13 +36,6 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0(`docs/lifecycle-ops.md#backup`) is unchanged. (#950)### Fixed-- Codex `exec_command` is now recognized as a shell tool by the capture- policy, so `ignore_paths` and the invalid-marker fail-closed stripping- apply to it. Previously, this Codex tool name bypassed the shell-command- path inspection, allowing commands that name an ignored path to be- captured. String and argv `cmd` forms, relative paths under `workdir`,- and malformed commands under an invalid marker now follow the existing- shell policy. (#974)- The `/web` page view keeps a leading H1 that is not the page title. Itdropped the body's first H1 whatever it said, as a duplicate of the titlein the header, but a frontmatter `title:` outranks the H1 and a setext H1diff --git a/crates/ai-memory-hooks/src/capture_policy.rs b/crates/ai-memory-hooks/src/capture_policy.rsindex 4a190198..c92f98b0 100644--- a/crates/ai-memory-hooks/src/capture_policy.rs+++ b/crates/ai-memory-hooks/src/capture_policy.rs@@ -795,7 +795,7 @@ fn family(name: &str) -> ToolFamily {"search" | "grep" | "glob" | "find" | "list" | "ls" | "list_files" | "read_dir"| "list_dir" | "grep_search" | "search_files" => ToolFamily::SearchList,"bash" | "shell" | "shell_command" | "exec" | "execute" | "run_command" | "web_search"- | "terminal" | "execute_bash" | "execute_cmd" | "exec_command" => ToolFamily::NonFile,+ | "terminal" | "execute_bash" | "execute_cmd" => ToolFamily::NonFile,_ => ToolFamily::Unknown,}}@@ -1872,99 +1872,21 @@ mod tests {),] {let decision = policy.inspect(agent, &raw, "/repo");- assert_eq!(- decision.protocol().tool_family(),- ToolFamily::NonFile,- "raw: {raw}",- );- assert_eq!(- decision.protocol().disposition(),- CaptureDisposition::Drop,- "raw: {raw}",- );+ if decision.protocol().tool_family() == ToolFamily::NonFile {+ assert_eq!(+ decision.protocol().disposition(),+ CaptureDisposition::Drop,+ "raw: {raw}"+ );+ } else {+ // `exec_command` is not a recognized shell tool name; it keeps+ // today's unknown-tool behavior rather than guessing.+ assert_eq!(raw["tool_name"], "exec_command", "raw: {raw}");+ assert_eq!(decision.protocol().tool_family(), ToolFamily::Unknown);+ }}}- #[test]- fn codex_exec_command_honors_ignored_paths() {- let policy = shell_policy();- // Regression for #974: Codex `exec_command` previously fell through- // to `ToolFamily::Unknown`, so the ignore_paths gate never matched- // the command line and the event kept flowing to the server. A- // command line that names an ignored path must be dropped locally,- // exactly like Claude `Bash` and Codex `shell`.- let ignored = json!({- "tool_name": "exec_command",- "tool_input": {"cmd": "cat docs/adr/x.md"},- });- assert_eq!(- policy- .inspect(AgentKind::Codex, &ignored, "/repo")- .protocol()- .disposition(),- CaptureDisposition::Drop,- "exec_command on an ignored path should be dropped",- );- // Legitimate control: the same shape against a path that is not- // ignored must still be captured, so the privacy gate does not- // become a blanket blackout for shell tools we now classify as- // recognized.- let allowed = json!({- "tool_name": "exec_command",- "tool_input": {"cmd": "cat src/lib.rs"},- });- assert_eq!(- policy- .inspect(AgentKind::Codex, &allowed, "/repo")- .protocol()- .disposition(),- CaptureDisposition::Keep,- "exec_command on an allowed path should still be captured",- );- }-- #[test]- fn codex_exec_command_array_form_honors_ignored_paths() {- let policy = shell_policy();- // Codex often wraps scripts as `["bash", "-lc", "..."]`; the argv- // form must be parsed the same way `shell` already is (#974).- let argv = json!({- "tool_name": "exec_command",- "tool_input": {"cmd": ["bash", "-lc", "cat docs/adr/x.md"]},- });- assert_eq!(- policy- .inspect(AgentKind::Codex, &argv, "/repo")- .protocol()- .disposition(),- CaptureDisposition::Drop,- "exec_command argv form on an ignored path should be dropped",- );- }-- #[test]- fn codex_exec_command_with_workdir_resolves_relative_ignored_paths() {- let policy = shell_policy();- // Codex `exec_command` carries its own `workdir`; relative- // arguments must be anchored to it, like OpenClaw `exec` and the- // other NonFile tools that already honour `workdir`.- let payload = json!({- "tool_name": "exec_command",- "tool_input": {- "cmd": "cat 0001.md",- "workdir": "docs/adr",- },- });- assert_eq!(- policy- .inspect(AgentKind::Codex, &payload, "/repo")- .protocol()- .disposition(),- CaptureDisposition::Drop,- "exec_command with workdir + relative ignored path should be dropped",- );- }-#[test]fn shell_matching_is_off_when_inactive_and_fails_closed_when_invalid_or_over_budget() {let raw = bash("cat docs/adr/x.md");@@ -2116,18 +2038,6 @@ mod tests {AgentKind::ClaudeCode,json!({"tool_name":"Bash","tool_input":{"cmd_line":"cat secret/x.md"}}),),- (- AgentKind::Codex,- json!({"tool_name":"exec_command","tool_input":{"cmd":7}}),- ),- (- AgentKind::Codex,- json!({"tool_name":"exec_command","tool_input":{"cmd":["cat", 7]}}),- ),- (- AgentKind::Codex,- json!({"tool_name":"exec_command","tool_input":{}}),- ),] {let decision = policy.inspect(agent, &payload, "/repo");assert_eq!(@@ -2138,36 +2048,6 @@ mod tests {}}- #[test]- fn invalid_marker_does_not_leak_codex_exec_command_body() {- // Security-boundary assertion for #974: even when a Codex- // `exec_command` payload cannot be parsed into a command line, the- // metadata-only body must never carry the raw argument. The fixed- // key set in `metadata_only_body` (session_id, cwd, tool_family,- // tool_name, tool_call_id, _ai_memory_capture) is the contract; if a- // future refactor widens it for shell tools, this test bites.- let policy = CapturePolicy::resolve(CaptureSource::Invalid, "/repo", None);- for payload in [- json!({"tool_name": "exec_command", "tool_input": {"cmd": 7}}),- json!({"tool_name": "exec_command", "tool_input": {"cmd": ["cat", 7]}}),- json!({"tool_name": "exec_command", "tool_input": {}}),- json!({"tool_name": "exec_command", "tool_input": {"cmd": "cat secret/x.md"}}),- ] {- let decision = policy.inspect(AgentKind::Codex, &payload, "/repo");- assert_eq!(- decision.protocol().disposition(),- CaptureDisposition::MetadataOnly,- "exec_command under an invalid marker must be metadata-only: {payload}",- );- let body =- serde_json::to_string(&metadata_only_body(Some("s"), Some("/repo"), &decision))- .unwrap();- assert!(!body.contains("cat"), "command body leaked: {body}");- assert!(!body.contains("secret"), "path-like body leaked: {body}");- assert!(!body.contains("\"cmd\""), "raw cmd key leaked: {body}");- }- }-#[test]fn long_bash_lc_script_in_argv_is_not_dropped_by_the_match_budget() {let ignore_paths = (0..40).map(|i| format!("/repo/private{i}/**")).collect();diff --git a/docs/marker-file.md b/docs/marker-file.mdindex a9a57c87..2459102f 100644--- a/docs/marker-file.md+++ b/docs/marker-file.md@@ -235,10 +235,11 @@ That form contains only bounded routing/tool/decision metadata, never paths,patterns, arguments, output, errors, titles, or nested payload. Unknown toolsretain current behavior.-Recognized shell tools (`Bash`, `shell`, `exec`, `exec_command`, `execute_bash`,-`terminal`, …) have no path field, so the command line is split into words-lexically, the way a POSIX shell quotes and separates them, without expanding-or running anything.+Recognized shell tools (`Bash`, `shell`, `exec`, `execute_bash`, `terminal`, …)+have no path field, so the command line is split into words lexically, the way+a POSIX shell quotes and separates them, without expanding or running anything.+Codex shell calls, including its `exec_command` path, reach hooks as `Bash`+with the command in `tool_input.command`.A command given as an argument vector keeps each element as one word (a pathwith spaces stays whole, up to 256 characters) and also splits each element onits own, so a `bash -lc "<script>"` script is read like any command line. Eachdiff --git a/docs/security-boundaries.md b/docs/security-boundaries.mdindex ee869d9b..b3fdf56a 100644--- a/docs/security-boundaries.md+++ b/docs/security-boundaries.md@@ -40,7 +40,7 @@ boundary not yet built.| 10 | Destructive-op live-process refusal + confirm flags (invariant #9) | `ai-memory-cli/src/commands/process_guard.rs` `sibling_processes` + confirm flags in `reset`/`restore`/`reindex`/`uninstall --purge-data`/`purge_project` | `admin_purge.rs` confirm→400; `removal.rs` — injected live-sibling makes each destructive command bail before touching the data dir | STRONG || 10b | Session purge is scope+owner-bound (no cross-session/project over-delete) | `ai-memory-store/src/ops.rs` `purge_session` — selection scoped to `(workspace_id, project_id)` and keyed on this session's own `summary_page_id` **or** `path='sessions/<sid>.md'` + `json_extract(frontmatter_json,'$.session_id')=<sid>` (frontmatter owner, not the recursive latest-chain); `in_scope==0 → NotFound` fail-closed; whole op in one transaction | `ops.rs` `purge_session_leaves_a_sibling_session_in_the_same_project_intact`, `…refuses_a_session_from_another_project_and_deletes_nothing`, `…refuses_a_session_from_another_workspace`, `…does_not_delete_an_identically_pathed_page_in_another_project`, `…removes_older_summary_versions_without_deleting_prior_manual_page` (#862) | STRONG || 11a | Hook backpressure (202/429) + bounded fan-out (invariant #5) | `ai-memory-hooks/src/router.rs` semaphore→429, 202 immediately, `MAX_HOOK_BATCH_ITEMS`, bounded LRU limiter | `router.rs` `handle_hook_returns_429_when_ingest_saturated`, `ingest_rate_limiter_is_bounded` | STRONG |-| 11b | Capture exclusions drop before storage | `ai-memory-hooks` `capture_policy.rs` `inspect`→`Drop` (before semaphore/spawn), including shell commands whose arguments name an ignored path (`match_command`; argv elements matched whole and tokenized); Codex `exec_command` (`tool_input.cmd`, string or string array, `workdir` relative-anchored) is recognized the same way as Codex `shell` so the gate cannot be bypassed by adapter/tool-name drift (#974); an invalid marker makes file and shell calls metadata-only, and the server admits a metadata-only shell body only under an invalid marker (`router.rs` `metadata_protocol_is_legal`); generated OpenCode/OMP/Pi/OpenClaw integrations mirror it in `render_shared.rs` `ts_capture_policy_v1` (`captureMatchCommand`) | `capture_policy.rs` per-agent `…honors_exclusions` tests, `shell_fixture_vectors` (shared `capture-policy.json` `shell` drop/keep vectors: tool aliases incl. OpenClaw/Devin `exec`, `workdir`, glob directories, Windows paths), `shell_tool_shapes_of_every_adapter_honor_exclusions`, `fixture_vectors` (incl. TS-adapter `bash`/`exec` vectors); `shell_matching_is_off_when_inactive_and_fails_closed_when_invalid_or_over_budget`; `capture_policy.rs` `codex_exec_command_honors_ignored_paths`, `codex_exec_command_array_form_honors_ignored_paths`, `codex_exec_command_with_workdir_resolves_relative_ignored_paths` (#974); `router.rs` `capture_protocol_shell_decisions_survive_server_reinspection`, `capture_protocol_invalid_marker_shell_is_metadata_only` (active metadata-only shell refused, older client's invalid-marker keep stripped, commandless control kept), `capture_protocol_unparseable_marker_strips_shell_events` (server fallback for an unparseable marker), `capture_protocol_invalid_shell_metadata_claim_must_be_canonical` (a stripped shell claim with a path count or non-`extracted` state is refused); `capture_policy.rs` `invalid_marker_strips_shell_calls_with_unparseable_commands` (now incl. Codex `exec_command` `cmd` int / mixed-argv / missing), `invalid_marker_does_not_leak_codex_exec_command_body` (security-boundary body-leak assertion, #974), `long_bash_lc_script_in_argv_is_not_dropped_by_the_match_budget`; `hook.rs` `shell_command_reading_an_ignored_path_is_dropped_before_spool`; `render_shared.rs` `generated_capture_policy_v1_node_runtime_evidence` (runs the same fixture sections against the emitted TypeScript; `#[ignore]` locally, run with `--ignored` under Node 24 by the Linux CI test job) | STRONG |+| 11b | Capture exclusions drop before storage | `ai-memory-hooks` `capture_policy.rs` `inspect`→`Drop` (before semaphore/spawn), including shell commands whose arguments name an ignored path (`match_command`; argv elements matched whole and tokenized); an invalid marker makes file and shell calls metadata-only, and the server admits a metadata-only shell body only under an invalid marker (`router.rs` `metadata_protocol_is_legal`); generated OpenCode/OMP/Pi/OpenClaw integrations mirror it in `render_shared.rs` `ts_capture_policy_v1` (`captureMatchCommand`) | `capture_policy.rs` per-agent `…honors_exclusions` tests, `shell_fixture_vectors` (shared `capture-policy.json` `shell` drop/keep vectors: tool aliases incl. OpenClaw/Devin `exec`, `workdir`, glob directories, Windows paths), `shell_tool_shapes_of_every_adapter_honor_exclusions`, `fixture_vectors` (incl. TS-adapter `bash`/`exec` vectors); `shell_matching_is_off_when_inactive_and_fails_closed_when_invalid_or_over_budget`; `router.rs` `capture_protocol_shell_decisions_survive_server_reinspection`, `capture_protocol_invalid_marker_shell_is_metadata_only` (active metadata-only shell refused, older client's invalid-marker keep stripped, commandless control kept), `capture_protocol_unparseable_marker_strips_shell_events` (server fallback for an unparseable marker), `capture_protocol_invalid_shell_metadata_claim_must_be_canonical` (a stripped shell claim with a path count or non-`extracted` state is refused); `capture_policy.rs` `invalid_marker_strips_shell_calls_with_unparseable_commands`, `long_bash_lc_script_in_argv_is_not_dropped_by_the_match_budget`; `hook.rs` `shell_command_reading_an_ignored_path_is_dropped_before_spool`; `render_shared.rs` `generated_capture_policy_v1_node_runtime_evidence` (runs the same fixture sections against the emitted TypeScript; `#[ignore]` locally, run with `--ignored` under Node 24 by the Linux CI test job) | STRONG || 11c | Capture hook ≤200ms budget (invariant #5) | `hooks/_lib.sh` capture path `curl --max-time 0.2` (context-fetch 1.0s and background drain 2.0s are separate, larger-budget paths) | none (shell-script timeout; hard to unit-test) — watch on any capture-path change | WATCH || 12 | Network/auth posture | `config.rs` loopback `DEFAULT_BIND`; `serve.rs` `validate_http_exposure`, `require_allowed_host`; `auth.rs` `require_bearer` | `serve.rs` host-guard (missing→400 / forged→403), non-loopback-requires-token; `auth.rs` wrong-token 401 | STRONG || 13 | Managed-run transcript attribution (concurrent launches in one checkout, invariant #16) | `ai-memory-store/src/workstream.rs` `link_native_session` stamps `native_session_linked_at` on its own run only, both `finish` updates drop the stamp when the session changes, `run_status` reports it; `ai-memory-cli/src/commands/run.rs` `resolve_native_session_after_run` takes a linked session only when `ai-memory-workstream` `native_session_in_checkout` holds it for this checkout (OpenCode by recorded directory), and never falls back to a link it set aside | `multi_session.rs` `a_session_linked_by_one_managed_run_is_not_another_runs`; `run.rs` `a_session_linked_during_the_run_wins_over_discovery` (concurrent newer session, another checkout's link refused, no fallback to it, unlinked control); `transcript.rs` `native_session_in_checkout_checks_the_opencode_directory`; `store/src/lib.rs` `managed_run_status_reports_a_link_made_during_the_run` | PARTIAL: a run whose child links nothing still falls back to discovering the newest session in the checkout |