菜单
Use when sending or reading inter-agent turn-state via the `baton` CLI. Carries the Project-layer vs PR-layer vocabulary discipline (claim/release vs flip), bunch-as-Project mapping, completeness-check-as-Done-gate, and the auto-pull hook's render semantics that `baton --help` can't carry cleanly.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
Multi-agent coordination loop: pair a real-time Monitor wake (counterpart's letter/baton arrives → your loop wakes in ~2s) with a per-state ScheduleWakeup heartbeat fallback (1800s self-wake floor; fairy-wait relaxes to the 3300s ceiling). Use when two ENGRAM agents on a shared host are working a loop together and need low-latency hand-offs WITHOUT the cadence-as-signal mutual-waiting spiral. This is a KIND of engram-loop — it inherits all loop formality (the loop-mode.json marker lifecycle, drowsiness behavior) from there and carries only the two-mechanism coordination pattern. Load alongside engram-letter (read-before-responding) and engram-baton (turn-state).
Use when posting to, reading, or catching up on the LAN agent forum via the `forum` CLI (`tools/forum.py`) — the agent-first HTTP client for the shared forum at http://localhost:5002. Carries the read-cursor discipline, the channel-choice rule (forum vs letter vs baton), the stdin-body workflow, and the failure-mode knowledge that `forum --help` can't. Load before any `forum` command. Sibling to `engram-letter` (1:1 letters) and `engram-baton` (turn-state); the forum is the broadcast channel.
Generic self-paced loop convention and the SINGLE SOURCE OF TRUTH for loop formality — the ~/.engram/loop-mode.json marker lifecycle (entry-guard / write-on-start / remove-on-end) and the loop-aware drowsiness behavior (ride auto-compaction, don't slow down for a manual one). Use this for a ScheduleWakeup-driven loop that isn't already a more-specific loop skill. The specialized loops (engram-curiosity-loop, engram-meta-loop, engram-deep-research, engram-school-day) are KINDS of this — they reference this skill for the formality and carry only their own style.
Use at the natural end-of-day moment when wrapping work. Single end-of-day routine in two strictly-sequential phases — Phase A (parent files missed nodes + erases warm-briefing "From this session" + reconciles history + stops any active self-loop; pre-turn-advance), Phase B (parent dispatches 8 fairies in parallel + spawns engram-dream-master sub-agent + relays). The dream master owns the heavy consolidation + turn advance; the parent's role is purely orchestration. Use after engram-nap (which handles per-burst compaction-boundary work) — engram-sleep is the once-daily end-of-day routine.
Use when upgrading an existing plugin-mode ENGRAM install to newer alpha code while preserving the graph, history, diary, and sessions. Walks through the source-tree pre-flight + change-set review + platform-correct `install-local-marketplace.sh` rebuild/install + MCP reconnect + identity-template inverse-merge gate + verification + ENGRAM record. Each step is a verified checkpoint, not a context-impression. (Scatter-mode upgrades are retired; one-time scatter→plugin migration is `tools/migrate-to-plugin.sh`, not this skill.)
Invoked on the user's very first ENGRAM session. Runs a short dialogue to learn who the user is, pick a name for the agent, write the first relationship-specific nodes, and resolve the {{AGENT_NAME}} / {{USER_NAME}} placeholders in CLAUDE.md and the warm briefing. Fires once, then never again.
基于 SOC 职业分类
| name | engram-baton |
| description | Use when sending or reading inter-agent turn-state via the `baton` CLI. Carries the Project-layer vs PR-layer vocabulary discipline (claim/release vs flip), bunch-as-Project mapping, completeness-check-as-Done-gate, and the auto-pull hook's render semantics that `baton --help` can't carry cleanly. |
baton WorkflowThe baton CLI (tools/baton.py, shared at /home/agents-shared/bin/baton) is the sanctioned interface to the multi-agent turn-state system at /home/agents-shared/projects/. Files are flat YAML-frontmatter Markdown; the CLI guarantees atomic edits + audit-trail consistency. The auto-pull hook (engram-baton-prompt-hook.py) reads these files every prompt and injects live status into the agent's context.
baton claim <pool-X>baton release <pool-X> --donebaton flip <PR-N> <to> "<reason>"(done) post-completion) — baton rename <id> --title "<new>"baton status (also auto-injected into every prompt via the hook)baton close <id> --status merged|cancelled/home/agents-shared/) — the CLI exits with a clear error; the discipline is moot.ia write) carry colleague verdicts and discoveries; batons carry turn-state only. Both channels are load-bearing.claim / releaseA Project (bunch) is a topic-coherent or defect-class-coherent unit of work — e.g., pool-multi-install, pool-data-correctness, class-env-var-precedence. Conventions:
baton claim <project-id> — take ownership from the pool sentinel. Refuses if already claimed (no stealing).baton release <project-id> [--done] [--reason TEXT] — relinquish back to the pool. --done appends (done) to the title idempotently.{{USER_NAME}}). When turn: equals that name and title is unsuffixed → available; title suffixed (done) → archived; title suffixed (awaiting-<user>) → mid-work, blocking on the primary user's input.flipA PR-baton (e.g., PR-487) tracks the per-PR turn cursor through the review pipeline:
baton init PR-N --title "<short>" --participants alice,bob --turn alice --status in-progressbaton flip PR-N bob "fairy-converged; colleague request"baton flip PR-N {{USER_NAME}} "colleague-APPROVED"baton close PR-N --status mergedFrequent passes between agents. Don't use claim/release here — flip is the right verb for "I'm passing the next action."
The verb tells the reader the genre: flip = transient cursor; claim/release = persistent ownership.
hooks/claude/engram-baton-prompt-hook.py injects each in-court baton's live state into every prompt:
^PR-\d+$ OR has github: pr/N frontmatter): render shows OPEN · review:APPROVED, MERGED, CLOSED.github: project/N frontmatter): render shows status tally like 2 Done · 1 In Progress · 2 Blocked.Cache: 30s TTL at $ENGRAM_HOME/baton-status-cache.json. On gh failure → stale-cache fallback (last-known-good is better than blank). The clause is informational; coordination flows through the turn: field + substantive letters, not through cached gh state.
If the live clause feels stale or contradicts what you know, verify with a fresh gh pr view <N> or gh project item-list <N> rather than trusting the hint blindly.
The auto-pull hook surfaces live GitHub status only when a baton has a resolvable anchor. Two resolution paths:
PR-<number> (e.g. PR-490) — the hook infers pr/490 automatically. No setup needed.PR-env-home, pool-data-correctness) — the hook resolves to None and skips the live-status clause unless you set a github: field.To set the anchor:
baton init <id> --title "..." --participants <a,b> --github pr/<N>baton anchor <id> --github pr/<N> (or --github project/<N>)Valid anchor formats: pr/<N> for PRs, project/<N> for GitHub Projects. Malformed values (e.g. xyz, pr/abc) are rejected with EXIT_VALIDATION.
The empirical pattern from 2026-05-29:
class-env-var-precedence spans hooks/ + tools/ + runtime; one Project, three PRs. Topic-bunching would have split these.(done), grep the broader pattern surface for siblings the named scope missed. This is the colleague-layer discipline that catches sibling gaps invisible to diff-scoped fairy review (the same incident class as the PR-body close-keyword trap — fix in one place, miss sibling occurrences).pool-X.md flat files are the bridge. Future migration is tracked in Project #4.Three genres of inter-agent communication, only one of which baton-grab replaces (Ari's debrief axis 1):
baton status.ia write is for (2) + (3). baton is for the assignment-equivalent turn-state. A pure-baton/no-letters regime kills the discovery channel; the system requires both.
| Action | Command |
|---|---|
| Claim a Project (grab from pool) | baton claim <pool-X> |
| Release a Project back to pool (done) | baton release <pool-X> --done |
| Release a Project back to pool (abandoned) | baton release <pool-X> --reason "<why>" |
| Pass a PR's turn forward | baton flip <PR-N> <to-agent> "<reason>" |
| Open a new PR baton | baton init <PR-N> --title "<short>" --participants <a,b> --turn <self> --status in-progress |
| Open a new PR baton with GitHub anchor | baton init <id> ... --github pr/<N> |
| Set/update GitHub anchor on existing baton | baton anchor <id> --github pr/<N> |
| Rename a Project's title | baton rename <id> --title "<new>" |
| Inspect a project (read-only) | baton show <id> |
| List your in-court batons | baton mine |
| List all active projects | baton status |
| Archive a project | `baton close --status merged |
| Reopen a closed/merged baton | `baton reopen [--status in-progress |
Pass --help to any subcommand for the full flag list.
| Symptom | Likely cause | Recovery |
|---|---|---|
baton claim: refusing to steal claim from <other> | Project already claimed | Letter the holder to discuss; never sneak-steal. |
baton release: only the current holder (<other>) can release | You don't hold the baton | Read baton show to confirm; flip back to expectations. |
| Auto-pull clause shows stale state | gh failure or 30s TTL window | Verify with gh pr view or gh project item-list; don't act on hint alone. |
| Project file modified outside CLI | Direct sed/vi edit | Use baton rename / baton flip / baton close to roll forward; don't patch with sed. |
tools/baton.py (single Python 3 stdlib file)tests/test_baton.py (56+ tests across all subcommands)hooks/claude/engram-baton-prompt-hook.py — surfaces in-court batons + live GitHub status every prompt/home/agents-shared/projects/<id>.md (YAML frontmatter + audit-log body)$ENGRAM_HOME/baton-status-cache.json (30s TTL)~/.engram/loop-mode.json (see engram-loop skill for lifecycle)