菜单
Operate the Chaplain automation system. Use when: starting the chaplain, checking pipeline status, submitting proposals to inbox, debugging pipeline failures, retrying failed topics, cleaning worktrees, running the philosopher or inquisitor.
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
| name | chaplain-ops |
| description | Operate the Chaplain automation system. Use when: starting the chaplain, checking pipeline status, submitting proposals to inbox, debugging pipeline failures, retrying failed topics, cleaning worktrees, running the philosopher or inquisitor. |
| argument-hint | 'start', 'status', 'retry gh-XXX', or 'clean' |
Operate the FSM-based automation runtime for YAMLGraph development lifecycle. Canonical source: docs/context/chaplain-system.md.
GitHub Issue (chaplain label)
↓ inbox_sync.sh
.chaplain/inbox/gh-XXX.md
↓ Dispatcher FSM
.chaplain/processing/gh-XXX.md
↓ Pipeline FSM (plan → judge → enforce → validate → merge)
├→ SUCCESS → .chaplain/done/gh-XXX.md
└→ FAILURE → .chaplain/failed/gh-XXX.md
Two FSMs:
.chaplain/config/watcher-dispatcher.yaml): Polls inbox every 10s, spawns pipeline worker per topic.chaplain/config/watcher-pipeline-v2.yaml): Plan → Judge → Enforce → Micro-Remediate → Validate → MergeThree-model separation: Plan (gpt-5.3-codex) → Judge (claude-sonnet-4, fresh eyes) → Enforce (gpt-5.3-codex)
.chaplain/scripts/start-system.sh
Starts: cleanup → validate → diagrams → web UI (port 3001) → dispatcher.
.chaplain/scripts/pipeline-status.sh # Overview
.chaplain/scripts/pipeline-status.sh 339 # Detailed for gh-339
Write a markdown file to .chaplain/inbox/:
cat > .chaplain/inbox/refactor-state-builder.md << 'EOF'
Problem: State builder has grown to 450 lines.
Task: Split into state_builder.py and state_reducer.py.
EOF
Remote submission: Open a GitHub Issue with the chaplain label. inbox_sync.sh imports it automatically.
rm .chaplain/failed/gh-XXX.md
git worktree remove tmp/worktrees/feat/watcher2-gh-XXX --force
git branch -D feat/watcher2-gh-XXX && git push origin --delete feat/watcher2-gh-XXX
gh issue edit XXX --add-label chaplain
.chaplain/scripts/clean-worktree.sh 339 340 341
| State | Type | Timeout | Purpose |
|---|---|---|---|
setup | bash | 30s | Branch + worktree creation |
plan | yamlgraph | 600s | FR draft + research + tests |
judge | yamlgraph | 600s | FR evaluation (different model) |
enforce_session | yamlgraph | 3600s | Implement changes |
micro_changelog | python | 30s | Changelog fragment remediation |
micro_title | bash | 30s | Commit title repair |
validate_fix | yamlgraph | 600s | Pre-commit/pytest remediation |
sanity_check | yamlgraph | 1200s | Post-validate review + diary |
validate_gate | python | — | Deterministic gate (max 5 retries) |
done | bash | 300s | Push → PR → merge |
Judge can emit: APPROVE → proceed, AMEND/SPLIT → revise, REJECT → fail.
| Directory | Purpose |
|---|---|
.chaplain/inbox/ | Awaiting pipeline admission |
.chaplain/processing/ | Currently being processed |
.chaplain/done/ | Completed topics |
.chaplain/failed/ | Failed topics |
.chaplain/drafts/ | Plan-phase drafts |
Scans docs/diary/ for recurring patterns, proposes graduations to Scripture:
.chaplain/philosopher.sh
Audits recent commits against the Scripture:
.chaplain/inquisitor.sh
.chaplain/inquisitor.sh --force # Override worktree gate
.chaplain/inquisitor.sh --propose # Write fix proposals to inbox
logs/
├── fsm-dispatcher.log
├── fsm-pipeline-gh-XXX-*.log
└── philosopher-YYYY-MM-DD.log
statemachine-validate .chaplain/config/watcher-dispatcher.yaml
statemachine-validate .chaplain/config/watcher-pipeline-v2.yaml
| Component | Location |
|---|---|
| Dispatcher config | .chaplain/config/watcher-dispatcher.yaml |
| Pipeline config | .chaplain/config/watcher-pipeline-v2.yaml |
| Shell library | .chaplain/lib/watcher/ |
| Custom actions | .chaplain/actions/ |
| Graphs | .chaplain/graphs/ |
| ID registry | .chaplain/id-registry.yaml |
| Author allowlist | .chaplain/allowed-authors.txt |
Author YAMLGraph prompt YAML files. Use when: writing prompt templates, configuring inline schemas or output_schema, using Jinja2 templates, adding system/user messages, defining structured output fields, or using system_segments for prompt caching.
Author YAMLGraph graph YAML files. Use when: creating or editing graph YAML, adding nodes or edges, configuring routing and conditions, setting up state keys, variables, data_files, tools, loop limits, error handling, or verification gates.
Create and manage YAMLGraph feature requests. Use when: writing a feature request, planning a feature, following the plan-judge-enforce workflow, submitting a proposal to the chaplain inbox, understanding the FR lifecycle, or writing acceptance criteria.
Release a new YAMLGraph version. Use when: bumping version, creating a release, tagging, pushing a release, freezing changelog, or handling pre-commit hook cascades during release.
Run code quality analysis on YAMLGraph. Use when: running ruff, bandit, radon, vulture, pytest coverage, checking code quality, finding dead code, security scanning, or generating improvement recommendations. For autonomous analysis, use the code-analysis agent instead.
Check LangSmith traces for YAMLGraph runs. Use when: debugging graph execution, reviewing LLM latency or token usage, investigating failed runs, verifying tracing config, querying recent traces, troubleshooting LangSmith connectivity.