菜单
Initialize .claude/spec.md from the latest plan's
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
Use when the user describes a multi-step feature or refactor and you need to decide whether to handle it solo, dispatch a parallel `/fan-out`, or run a `/worktree-team` planner→generator→reviewer pipeline. Outputs a routing recommendation with the reasoning behind it.
Use whenever the user asks to apply the same operation to many independent files or components — bulk migration, batch refactor across controllers, parallel exploration of unrelated subsystems. Dispatches one subagent per file with a shared prompt template, then collects the results.
Use to toggle the safe-mode flag at .claude/safe-mode. The flag is written automatically by consecutive-failure-guard once tool failures hit FORGE_SAFE_MODE_THRESHOLD (default 5), and block-destructive.sh denies every Bash/Write/Edit while it exists. /safe-mode off clears the flag, logs the exit, and prompts /postmortem; /safe-mode on enters manually; /safe-mode status reports current state.
Run the 5-stage Token Transformation Pipeline (TRAE §5.2.2) over the current session state — Collection, Ranking, Compression, Budgeting, Assembly. Emits a concrete next-action recommendation (/compact, /lean-md, or /progress-log + fresh session) instead of a generic warning.
Audit or scaffold a CLAUDE.md against the Karpathy 4-section structure (Think Before Coding · Simplicity First · Surgical Changes · Goal-Driven Execution). Companion to /lean-md — structure audit, not size trim.
Print every policy enforcement point declared in plugins/diagnostics/registry/policies.json — id, verdict, plugin, hook event, severity, bypass — grouped by verdict. Single discoverable index of what the harness blocks, anchors, nudges, or logs at runtime.
| name | living-spec |
| description | Initialize .claude/spec.md from the latest plan's |
| when_to_use | Reach for this immediately after a plan is approved (ExitPlanMode), before the generator dispatches, and re-run only if the plan itself changes. Do NOT use for session progress logging — use `/progress-log` instead. |
| paths | [".claude/spec.md"] |
| disable-model-invocation | true |
| allowed-tools | ["Read","Write","Glob","Grep","Bash"] |
| logical | .claude/spec.md initialized (or delta-appended) with contract items from the latest plan |
Write .claude/spec.md from the latest plan's ## Contract. Unlike /contract (which just re-reads static criteria to fight context decay), the living spec is continuously updated by after-subagent.sh as each phase (planner → generator → reviewer) completes. The reviewer compares code vs the spec — not just vs the original plan.
Find the active plan. Run bash plugins/workflow/skills/orchestrate/scripts/find-active-plan.sh (the single source of truth — sort -V numeric-prefix order, skips gate-complete plans). If it prints empty, stop: No plan found. Run Plan mode first.
Extract ## Contract. If missing, stop: Plan has no ## Contract section. /living-spec expects one.
If .claude/spec.md already exists — read it. Check the Plan: header line. If it matches the current plan file, report spec.md already initialized for this plan. Use after-subagent updates instead. and stop.
Write .claude/spec.md with this exact structure:
# Living Spec
Plan: .claude/plans/<file>.md
Initialized: <UTC ISO8601>
## Contract (from plan)
<verbatim copy of the ## Contract section>
## Deltas
<!-- after-subagent.sh appends blocks here, newest last:
### <UTC> — <agent>
Completed:
- ...
Pending:
- ...
-->
Report: spec.md initialized with N contract items. after-subagent.sh will append deltas as agents complete.
bash plugins/workflow/skills/orchestrate/scripts/find-active-plan.sh (abort if it prints empty)## Contract (abort if missing).claude/spec.md exists and its Plan: header matches, report spec.md already initialized for this plan. and stop.claude/spec.md with Plan / Initialized / ## Contract (from plan) / ## Deltas sections (verbatim contract copy)spec.md initialized with N contract items. after-subagent.sh will append deltas as agents complete.Writers:
/living-spec (this skill) — initializesplugins/workflow/hooks/after-subagent.sh — appends delta blocks on each SubagentStopReaders:
/verify — cross-references deltas with features.jsonsurface-progress.sh (SessionStart) — surfaces the last delta/session-resume — includes spec tail in briefing/rest-audit Reliability axis — checks spec presence + delta densityPaired artifacts:
.claude/plans/*.md — the input.claude/features.json — the parallel testable view (from /feature-list)## Contract is empty → write spec.md with ## Contract section empty + warn..claude/spec.md is a directory → fail loudly..claude/ → fail loudly; do NOT fall back to writing elsewhere.claude-progress.txt — those are different artifacts (spec = current plan state, progress = session-to-session history).Input:
plan file: .claude/plans/s7-cache-rewrite.md
## Contract excerpt:
- Cache layer behind a single `CacheService` interface
- All call sites use the interface; no direct Redis client imports outside `app/Cache/`
- Failure path: Redis unavailable → in-memory fallback for ≤60s, then bubble error
spec.md status: does not exist
Output:
# Living Spec
Plan: .claude/plans/s7-cache-rewrite.md
Initialized: 2026-04-28T15:00:00Z
## Contract (from plan)
- Cache layer behind a single `CacheService` interface
- All call sites use the interface; no direct Redis client imports outside `app/Cache/`
- Failure path: Redis unavailable → in-memory fallback for ≤60s, then bubble error
## Deltas
Reported: spec.md initialized with 3 contract items.
Input:
spec.md status: exists
Plan: header matches the current latest plan
Output:
spec.md already initialized for this plan. Use after-subagent updates instead.
(no file write)