メニュー
Orchestrates the complete planning pipeline: product direction (ceo-review) -> architecture (eng-review) -> implementation plan (start) -> validation (validate) -> execution (execute). Run stages individually or let the orchestrator coordinate the full flow.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Autonomous improvement loop: scan codebase metrics, scaffold experiment files, run agent-driven iterations until metric improves
Audit Claude Code agents defined in .claude/agents/ for description specificity, model tier appropriateness, tools scoping, and system prompt quality. Detects dispatch ambiguity between agents, flags over-permissive tool grants, and checks for human-in-the-loop patterns that break programmatic orchestration. Use when onboarding to a project with existing agents, after adding new agents to a fleet, or when an orchestrator consistently selects the wrong agent.
Post-deploy monitoring: watch production after a deploy and alert on regressions
Clean up stale git worktrees with merged branch detection and disk usage report
Safely remove a git worktree with branch cleanup and safety checks
Create isolated git worktrees for feature development without switching branches
| name | plan-pipeline |
| description | Orchestrates the complete planning pipeline: product direction (ceo-review) -> architecture (eng-review) -> implementation plan (start) -> validation (validate) -> execution (execute). Run stages individually or let the orchestrator coordinate the full flow. |
| allowed-tools | Read Write Bash Task |
| effort | high |
Orchestrates the complete plan-to-execution pipeline. Can run the full pipeline or a single isolated stage.
| Stage | Skill | Purpose |
|---|---|---|
| 1 | /plan-pipeline:ceo-review | Challenge the brief, lock product direction |
| 2 | /plan-pipeline:eng-review | Lock architecture, diagrams, and test matrix |
| 3 | /plan-pipeline:start | 5-phase planning: PRD, research, ADRs, task list |
| 4 | /plan-pipeline:validate | 2-layer validation before any code is written |
| 5 | /plan-pipeline:execute | Worktree isolation, parallel agents, quality gate, PR |
/plan-pipeline # full pipeline, asks for context
/plan-pipeline --from=start # skip gates, start from planning phase
/plan-pipeline --from=validate # validate an existing plan
/plan-pipeline --from=execute # execute a validated plan
ceo-review: use before any significant feature when the direction is not locked. Especially valuable when the request is specific (specificity signals collapsed solution space).
eng-review: use after direction is locked. Required for features with async components, external dependencies, or multi-step flows.
start: use for any non-trivial feature touching more than 2 files or involving architecture decisions.
validate: always before execute. The cost of validation is negligible against the cost of discovering issues mid-execution.
execute: after validate confirms all issues are resolved.
--from=eng-review or later)--from=start or later)docs/plans/plan-{name}.md ceo-review
|
eng-review
|
start
|
validate
|
execute
Each stage writes its output to disk before the next stage begins. If the pipeline is interrupted, resume with --from=<stage> using the correct stage name. All decisions are recorded in docs/plans/plan-{name}.md and the corresponding ADRs in docs/adr/.