菜单
Turn a spec into a design.md (the 'how' — architecture decisions). Use when Malik says 'plan the implementation', 'how should we build this', or 'design this'. Second SDD step.
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
Health-check the memory bank, rules, and index — flag stale, contradictory, or redundant entries and spec↔code drift. Use when Malik says 'audit the bank', 'is the memory stale', 'check spec harness health', or 'clean up the rules'.
Run the SDD pipeline over tasks.md: implement → test → VERIFY → review, one task at a time, under the ratchet. Use when Malik says 'build it', 'run the pipeline', 'implement the tasks', or 'ship this feature'. Each task gated by the separate verifier; lessons harvested by learn.
Bind the 6 SDD agents to THIS project's rules — fill each agent's GEN:rules region with its concern's real rules (implementer←Coding, tester←Testing+commands, reviewer←Reviewing, researcher/planner←Architecture, verifier←goal). Use when Malik says 'generate the agents', 'make the agents project-specific', 'bind agents to the rules', or right after rules are created/updated. If no rules exist yet, it routes to rules-generation (3 sources: structure + project best-practice + the internet) first. Idempotent; re-run after learn updates a rule.
Add or refresh Spec Harness in a repo — greenfield (new) or existing (integrate), idempotent. Use when Malik says 'install spec harness', 'set this repo up with spec harness', 'integrate spec harness into X', or 'start a new spec-harness project'. Runs the interview (mode, stack, constraints, work-intake source, learning goals, regression guards) and scaffolds the bank + agents + commands + skills without clobbering existing CLAUDE.md/AGENTS.md.
The learning loop — turn feedback, a correction, or a verifier-FAIL root cause into one enforced rule injected into the ratchet (AGENTS.md/ai_rules), dated. Use when Malik says 'learn this', 'remember not to do X', 'capture this lesson', 'don't do that again', or right after he corrects you. Closes the loop the verifier opens; emits an optional human lesson too.
Per-directory rules — resolve the nearest RULES.md for a path (coding/architecture/packages/testing/reviewing), or generate it from a directory's real code when missing. Use when Malik says 'different directories have different rules', 'add rules per directory', 'each package needs its own rules', 'what rules apply here', or when working in a monorepo with mixed stacks. Each section is owned by one SDD agent; deepest-wins cascade.
| name | spec-harness-plan |
| description | Turn a spec into a design.md (the 'how' — architecture decisions). Use when Malik says 'plan the implementation', 'how should we build this', or 'design this'. Second SDD step. |
Produce the 'how': architecture decisions in design.md, honoring constitution.md and the existing system (no rebuilding what the index shows already exists).
.claude/commands/spec-harness/plan.md (this repo) or the spec-harness source commands/plan.md..claude/agents/: planner..memory/) · Spec-Driven Development · Harness (ratchet AGENTS.md + verifier + learning loop). Read SPEC-HARNESS.md for how this repo is wired.goal.md.spec-harness-learn → a dated rule.