Harness Kit

skill

code-review

Parallel multi-agent code review. Launch reviewer team, synthesize findings, auto-fix blocking issues, loop until clean. Use when: "review this", "code review", "is this ready to ship", "check this code", "review my changes", "autoreview", "Codex review", "Claude review", "second-model review". Trigger: /code-review, /review, /autoreview.

/code-review

Source: skills/code-review/SKILL.md

What it does

Parallel multi-agent code review. Launch reviewer team, synthesize findings, auto-fix blocking issues, loop until clean. Use when: "review this", "code review", "is this ready to ship", "check this code", "review my changes", "autoreview", "Codex review", "Claude review", "second-model review". Trigger: /code-review, /review, /autoreview.

Workflow role

Review primitive in the Harness Kit operating loop.

Source contract preview

This generated excerpt gives readers the beginning of the live primitive contract before they jump to GitHub.

Multi-provider, multi-harness code review. You are the marshal — read the diff,
select reviewers, craft prompts, dispatch everything in parallel, synthesize
results, fix blockers, loop until clean.

## Delegation Floor

Delegation floor applies: probe the roster first; dispatch two or more
providers for substantive work; direct solo only for mechanical, emergency,
user-forbidden, or fewer-than-two-providers cases. See
`harnesses/shared/AGENTS.md` (Roster).

Local lane guidance: Use independent adversarial review lanes calibrated to what would embarrass us in production; give reviewers the diff, acceptance criteria, and risk lens, not author reasoning.

## Completion Evidence

Completion evidence core applies: use `harnesses/shared/AGENTS.md`
(Completion Evidence) as the universal evidence shape, then fill the local
fields below for review verdicts.

## Marshal Protocol

1. **Read the diff.** `git diff $BASE...HEAD` (default base: `main` or `master`).
   Classify: what changed? (API, UI, visual artifact, docs/report surface,
   tests, infra, security, perf, data model, etc.)

2. **Optionally run the structured autoreview helper.** When the human asks for
   `autoreview`, Codex review, Claude review, or second-model review, or when a
   frozen local/branch/commit bundle would sharpen synthesis, use
   `references/autoreview-helper.md`. The helper is a leaf review engine, not a
   substitute for roster evidence, verdict refs, executable-path verification,
   or `/deliver` clean loop.

3. **Load review pattern catalogs.** If the target repo has a local
   `references/review-patterns.md` or equivalent repo-local pattern catalog,
   load it before dispatching reviewers. When a local entry links to shared
   Harness Kit references such as
   `references/bounded-payload-discipline.md`, include that reference in the
...

What to verify