How do I install the Review skill?

Run npx skillmds@latest add honerlaw/review in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

Is Review safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Review?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Review free to use?

Yes. Installing skills from SkillMD is free, and the skill stays under its author's original license.

Who published Review?

honerlaw (@honerlaw) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Review by honerlaw — Agent Skill

Name: Review
Author: honerlaw

Review

Reviews a changeset against both the minerva spec/knowledge lenses and code quality. With a work unit in context it runs a spec/knowledge audit alongside the code quality review, presenting both result sets before unified triage; if a GitHub PR exists for the branch it delegates code quality to `code-review:code-review`, otherwise it performs a check via a fresh-context subagent in the same finding format. Triage state persists to the scratchpad so re-runs pre-fill prior dispositions. Use when implementation on a work unit has just finished and the diff is unreviewed, when the user asks to review or audit a changeset or to verify shipped code matches what was designed, or when they invoke `minerva:review`.

honerlaw 923c824 2 files · 15.8 KB Updated Sep 17, 2026

File contents

Review the active changeset for both design compliance and code quality. Works against the local diff — no PR required. When a minerva work unit is found, runs a spec/knowledge audit alongside the code quality review and presents both result sets in parallel before triage. When no minerva context exists, runs the code quality review alone.

Usage

minerva:review — audits the work unit inferred from current-session context, or the most-recently-modified if context is ambiguous; falls back to a plain code review if no work unit is found
minerva:review 005-add-payments — audit the named unit explicitly
minerva:review <date-slug> --auto=<orchestrator> — orchestrated mode; see Orchestrated mode below

Orchestrated mode (`--auto`)

Mode argument: --auto

--auto=<orchestrator> is an observable signal that an autonomous orchestrator has substituted its own adjudication for this skill's user gates. Act on the argument — never on a judgment about who is calling (2026-06-07-decision-phase-handoff-rides-observable-intake). Absent it, every gate behaves exactly as written.

It satisfies the triage hard gate in references/protocol.md.

Protocol

The full step protocols live verbatim in references/protocol.md — read it now, before executing: Target resolution → Worktree entry → Diff resolution → Minerva audit (spec fidelity + knowledge compliance; only when minerva context exists) → Code review invocation (delegates to code-review:code-review when a PR exists, fresh-context subagent check otherwise) → Parallel presentation → Interactive triage (FIX / SUGGEST / IGNORE) → Triage persistence (scratchpad pre-fill for re-runs) → On approval — file writes → Report.

Lifecycle ordering

Canonical order: minerva:work → minerva:review → minerva:promote → minerva:ship. Review runs before promote so review-derived scratchpad notes flow through the promote partition. If review surfaces durable knowledge that promote already ran on, run promote again. The skills are idempotent — cycle as needed.

Idempotency

Review is stateless across runs except for the ## Review triage YYYY-MM-DD blocks it writes to scratchpad for resume. Re-running on the same diff produces the same findings; dispositions are pre-filled from the most recent triage block.

minerva:review  →  fixes applied  →  minerva:promote  →  minerva:review  →  zero findings  →  minerva:ship

Out of scope

Writing to .minerva/knowledge/ directly. All durable knowledge goes through minerva:promote — one writer, one set of conventions.
A review.md log file. FIX outcomes are visible in git; SUGGEST and IGNORE outcomes flow through scratchpad → promote. Triage state lives in scratchpad. A standalone log duplicates what those already capture.
Scoped review (e.g. minerva:review src/api/). Deferred until the no-scope default proves noisy.

---name: reviewdescription: Reviews a changeset against both the minerva spec/knowledge lenses and code quality. With a work unit in context it runs a spec/knowledge audit alongside the code quality review, presenting both result sets before unified triage; if a GitHub PR exists for the branch it delegates code quality to `code-review:code-review`, otherwise it performs a check via a fresh-context subagent in the same finding format. Triage state persists to the scratchpad so re-runs pre-fill prior dispositions. Use when implementation on a work unit has just finished and the diff is unreviewed, when the user asks to review or audit a changeset or to verify shipped code matches what was designed, or when they invoke `minerva:review`.---Review the active changeset for both design compliance and code quality. Works against the local diff — no PR required. When a minerva work unit is found, runs a spec/knowledge audit alongside the code quality review and presents both result sets in parallel before triage. When no minerva context exists, runs the code quality review alone.## Usage- `minerva:review` — audits the work unit inferred from current-session context, or the most-recently-modified if context is ambiguous; falls back to a plain code review if no work unit is found- `minerva:review 005-add-payments` — audit the named unit explicitly- `minerva:review <date-slug> --auto=<orchestrator>` — orchestrated mode; see **Orchestrated mode** below## Orchestrated mode (`--auto`)**Mode argument**: `--auto``--auto=<orchestrator>` is an **observable** signal that an autonomous orchestrator hassubstituted its own adjudication for this skill's user gates. Act on the argument — never on ajudgment about who is calling (`2026-06-07-decision-phase-handoff-rides-observable-intake`).Absent it, every gate behaves exactly as written.It satisfies the triage hard gate in `references/protocol.md`.## ProtocolThe full step protocols live verbatim in `references/protocol.md` — **read it now, before executing**: **Target resolution** → **Worktree entry** → **Diff resolution** → **Minerva audit** (spec fidelity + knowledge compliance; only when minerva context exists) → **Code review invocation** (delegates to `code-review:code-review` when a PR exists, fresh-context subagent check otherwise) → **Parallel presentation** → **Interactive triage** (FIX / SUGGEST / IGNORE) → **Triage persistence** (scratchpad pre-fill for re-runs) → **On approval — file writes** → **Report**.## Lifecycle orderingCanonical order: `minerva:work → minerva:review → minerva:promote → minerva:ship`. Review runs **before** promote so review-derived scratchpad notes flow through the promote partition. If review surfaces durable knowledge that promote already ran on, run promote again. The skills are idempotent — cycle as needed.## IdempotencyReview is stateless across runs except for the `## Review triage YYYY-MM-DD` blocks it writes to scratchpad for resume. Re-running on the same diff produces the same findings; dispositions are pre-filled from the most recent triage block.```minerva:review → fixes applied → minerva:promote → minerva:review → zero findings → minerva:ship```## Out of scope- **Writing to `.minerva/knowledge/` directly.** All durable knowledge goes through `minerva:promote` — one writer, one set of conventions.- **A `review.md` log file.** FIX outcomes are visible in git; SUGGEST and IGNORE outcomes flow through scratchpad → promote. Triage state lives in scratchpad. A standalone log duplicates what those already capture.- **Scoped review** (e.g. `minerva:review src/api/`). Deferred until the no-scope default proves noisy.

Frequently asked questions