Governance Skill

Enforces engineering standards and code quality policies. The review runs in one delegated reviewer subagent, which keeps intermediate output out of the main conversation.

Standards Document: standards.md (starter default — replace with your own) Audit Script: npm run audit:standards Full Protocol: agent-prompt.md

Execution Model (v2.2)

Type: Delegated Execution — delegates to ONE built-in reviewer subagent.

This skill uses the reviewer subagent type (available in all Claude Code installations — no custom agent required):

1. Spawn a single reviewer agent
2. Delegate the governance task, telling it to read agent-prompt.md in full first
3. Await completion
4. Report the structured summary to the user

Why Delegate

The reviewer runs commands, reads files and applies fixes in its own context. Only the structured summary returns to the main conversation, so the main thread does not carry the review's intermediate output.

What's Checked

The full rules live in agent-prompt.md. The reviewer covers:

• Approach validation — is the chosen approach sound before the details are reviewed
• Type safety — no unjustified any, explicit types at boundaries
• Constant and mock consistency — one canonical home per shared constant; mocks match real shapes
• External-API contracts — and anti-false-green mocks that pass whatever the code does
• Shared-state and concurrency audit — five checks: global reads outside the producer, re-firing lifecycle listeners, missing re-fire tests, double reads of a module-level singleton across an await, and unguarded async-completion writes to shared state
• Duplicate security gates and sibling implementations — the same guard or logic must not diverge across copies
• Observation and gate correctness in verification code — a check must be able to fail, and must examine what it claims to
• Zero deferral — every finding is fixed now, with narrow, documented exceptions
• String-replacement sweeps — scripted edits are re-checked for dropped or mangled text
• Severity guide — how findings are classified
• Retrospectives — structured post-merge review

How to Invoke

When this skill is triggered, read agent-prompt.md for the full protocol, then delegate:

Code Review

Task({
  description: "Governance code review",
  prompt: `Read <skill-dir>/agent-prompt.md in full before doing anything, and follow it.
(<skill-dir> is the directory containing this SKILL.md, e.g. ~/.claude/skills/governance.)
Review the following files for standards compliance:
- [list files or PR description]

Execute full code review workflow:
1. Run standards audit
2. Check each file against standards.md
3. Fix ALL issues immediately (zero deferral)
4. Return structured summary`,
  subagent_type: "reviewer"
})

Pre-Commit Check

Task({
  description: "Pre-commit governance",
  prompt: `Read <skill-dir>/agent-prompt.md in full before doing anything, and follow it.
(<skill-dir> is the directory containing this SKILL.md, e.g. ~/.claude/skills/governance.)
Run pre-commit governance checks:
1. typecheck, lint, format:check, test
2. Check for untracked source files
3. Return pass/fail with specific failures`,
  subagent_type: "reviewer"
})

Retrospective

Task({
  description: "Run retrospective",
  prompt: `Read <skill-dir>/agent-prompt.md in full before doing anything, and follow it.
(<skill-dir> is the directory containing this SKILL.md, e.g. ~/.claude/skills/governance.)
Run retrospective for [topic/sprint]:
1. Analyze completed issues
2. Gather metrics
3. Write report
4. Return summary`,
  subagent_type: "reviewer"
})

Quick Reference

Pre-Commit Checklist

[ ] npm run typecheck
[ ] npm run lint
[ ] npm run test
[ ] npm run audit:standards
[ ] No console.log statements
[ ] No hardcoded secrets

Compliance Check

npm run audit:standards
node scripts/governance-check.mjs

Full agent protocol, anti-pattern tables, and output format in agent-prompt.md Version 2.2.0 — Expanded review rules, code-review and retro templates, dispatcher passes the protocol path