菜单
End-of-session — update memory, write journal entry, clean up
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。
基于 SOC 职业分类
Post-merge workflow — extract lessons and compile rules (manual sequence, wrap is retired)
Session-start — consume/derive orientation, poll mail since last signoff, re-derive carryforward gates, present next-steps for operator ruling
Drive the local pre-push review loop to settle — absorb findings locally before any external bot pass
Unified PR review triage — fetch, normalize, and batch-action bot comments
Compiled architectural rules and lessons from the Totem knowledge base. Load this when reviewing code for pattern violations or architectural drift.
Pre-work ritual — spec, search, and (when needed) a 1-page implementation design doc before touching code
| name | signoff |
| description | End-of-session — update memory, write journal entry, clean up |
End-of-session wrap-up. Post-Proposal-282 (ADR-106), journals + handoffs live in the per-repo .totem/orchestration/<agent-id>/ tree (gitignored) — NOT the substrate. Substrate stays as a frozen archive for forensic reads; the active surface is local.
Update memory. Update auto-memory files (e.g. MEMORY.md, topic memories) with any new state — version shipped, tickets closed, key decisions, banked feedback or doctrine signals.
Write a journal entry to the per-repo orchestration path. Filename convention: <model>-NNNN-<short-topic-slug>.md (e.g., claude-0057-phase-4-resolver-shipped.md).
Resolve the path two steps:
a. Identify your agent-id from the current repo's basename. The hardcoded map (Proposal 282 § Scope item 3 — keep in sync with the ADR-106 cohort list):
Repo (git rev-parse --show-toplevel basename) | Claude agent-id | Gemini agent-id | Kimi agent-id |
|---|---|---|---|
totem | totem-claude | totem-gemini | totem-kimi |
totem-strategy | strategy-claude | strategy-gemini | (not seated) |
liquid-city | lc-claude | lc-gemini | (not seated) |
arhgap11 | arhgap11-claude | arhgap11-gemini | (not seated) |
totem-status | status-claude | status-gemini | (not seated) |
totem-playground | (orphan stream — no native agent) | (orphan stream) | (orphan stream) |
Seat discovery is dir-derived (mmnto-ai/totem#2141): any .totem/orchestration/<agent-id>/ directory registers that seat for this repo, UNIONED with the basename map above so roster siblings stay visible on fresh clones where the gitignored tree is partial (precedence: TOTEM_SELF_AGENT env > config.json host_agents > seat dirs ∪ basename map). Override hook: a host_agents: string[] field in .totem/orchestration/config.json still replaces the derived answer — but omitting a PRESENT seat dir attaches a loud warning naming the omitted seat (the dir is the registration; config-exclusion is not a decommission mechanism). The returned list of agent-ids is used by consumers (e.g., totem mail) to filter cross-repo handoffs — messages addressed to any agent-id in the list belong to this repo's session.
Visiting case. If your row's Claude-agent-id column is _(no Claude variant)_ or _(orphan stream — no native agent)_, you are visiting a repo that doesn't natively host your agent. Resolve the journal path to <repoRoot>/.totem/orchestration/<your-home-agent-id>/journal/, where <your-home-agent-id> is your own agent-id (e.g., a strategy-claude session always writes as strategy-claude regardless of which repo it's visiting; concretely, strategy-claude visiting totem-status writes to totem-status/.totem/orchestration/strategy-claude/journal/). The journal records the visiting agent's session state — the host repo doesn't need a native Claude agent to be a valid write target.
b. Resolve the journal directory via resolveOrchestrationPaths(repoRoot, agentId).journal from @mmnto/totem. Returns the absolute path to <repoRoot>/.totem/orchestration/<agent-id>/journal/ when the tree exists. If source === 'none' (the tree does not exist yet in this repo) the resolver returns null for every path field — in that case, construct the path manually as <repoRoot>/.totem/orchestration/<agent-id>/journal/ and create the directory first via mkdir -p; the path is gitignored and safe to create.
No commit, no push. .totem/orchestration/ is gitignored — local filesystem write is the entire operation. No more substrate rebase-retry loops; the cross-agent write-collision class is eliminated by the single-writer-per-path invariant (you only ever write into your own <agent-id>/ subtree).
Clean up stale local branches:
git for-each-ref --format='%(refname:short) %(upstream:track)' refs/heads | while read -r branch track; do
[[ "$track" == "[gone]" ]] && git branch -D -- "$branch"
done
Prune + compact your own ECL cursor (retention + processed-mark GC). Delete your own outbox/ dispatches older than the retention window (N = 14 days) per ECL outbox-retention doctrine (mmnto-ai/totem-strategy:doctrine/ecl-discipline.md § 4.4), THEN compact your processed/ cursor per § 4.5 / ADR-106 § A2. The outbox is transport, not archive — a dispatch's durable content already lives in its home (rulings → ADRs / issues, work-state → the GH board, session history → journal/), so the aged courier file is disposable (gitignored + local). The processed/ cursor is the read-side twin: a mark whose inbound dispatch its sender already swept shadows nothing, so it is safely collectable. The operator should never have to janitor the mail substrate.
Mechanism: totem ecl-gc --apply --compact — self-resolves your agent-id (same precedence as step 2a: TOTEM_SELF_AGENT env > config.json host_agents > seat-dir ∪ basename map). It prunes only <repoRoot>/.totem/orchestration/<your-agent-id>/outbox/ (a self-resolving binary structurally cannot prune a peer), then compacts only your own processed/ marks that shadow nothing. Compaction is cursor-coupled, not age-based, and deletes ONLY against a provably-complete poll — full expected cohort roster present, zero scan warnings, not truncated — else it retains everything (uncertain ⇒ retain). Dry-run by default; --apply deletes. Neither phase touches journal/. Report the pruned + collected counts. Exit codes: 0 clean · 1 some deletes failed (janitorial sensor) · 2 usage/agent-unresolvable · 3 compaction ABORTED loudly (fail-loud, never a silent skip) — no cohort roster declared, the roster is incomplete on this machine, or its A2.4 re-poll check tripped. Do not block the seal on 1 or 3 — the gate-red arms retain the whole cursor (uncertain ⇒ retain); only note them. The gc is a janitorial sensor, not a gate (Tenet 13).
Report: what shipped, what's pending, what's next.
Cross-repo handoffs (when you need to dispatch a message to another agent) write to your own <repoRoot>/.totem/orchestration/<agent-id>/outbox/<YYYY-MM-DDTHHMMZ>-<your-agent-id>.md with to: <recipient-agent-id> in the frontmatter. Recipients discover inbound handoffs by polling the single-level glob <workspace>/*/.totem/orchestration/*/outbox/*.md filtered by their own to: frontmatter match.
Substrate (legacy) is read-only. Do NOT write new content to mmnto-ai/totem-substrate:.handoff/ or :.journal/. The substrate stays mounted as a frozen archive accessible via resolveSubstratePaths(cwd) for forensic reads; the cutover broadcast (when it lands) will confirm the final substrate-write cutoff.