agentleFS
junhjangjunhjang/claude-harness/.claude/skills/critic-review/SKILL.md

critic-review

Skill0 starsChanged 6 months agoMITView on GitHub

Adversarial principle review — dispatches the `critic` agent against the current diff (or named target) to surface violations of `docs/harness/principles.md` (P1–P12 + retired commitments). Use when asked "audit principles", "is this drifting from the principles", "principle review", "is this consistent with our locked-in commitments", "is this re-introducing a retired pattern", or before merging substantial architecture / harness changes. The agent is Opus, read-only, costs apply — confirm before fan-out.

2 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

2 rules. About 0 are common, 1 is distinctive, and 1 is very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Operator-attended: Never auto-confirm step 2.

    line 59
  2. Very specific:

    No double-charging: If /review-changes was already run on this HEAD sha and produced a critic-style finding, this slash is redundant — …

    line 61

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

/critic-review

Adversarial review wrapper that fires the critic agent against the diff (or a named target) to find principle violations.

This skill is the auto-loadable companion to the critic agent. Without it, the LLM has to remember to dispatch critic via the Agent tool with subagent_type: critic — which in practice nobody does. The wrapper makes the intent explicit and surfaceable through the same auto-load path the rest of the harness uses.

Per docs/harness/principles.md §12, this skill is locked because (a) it dispatches an Opus agent — real LLM cost, and (b) on principle audits at architectural boundaries, the operator should attend the spend.

When invoked: /critic-review [<target>]

<target> defaults to the current branch's diff vs main (mirrors /review-changes). Otherwise an explicit path or PR number.

1. Determine the review target

TARGET="${1:-}"
if [ -z "$TARGET" ]; then
  BASE=$(git merge-base origin/main HEAD)
  TARGET_DESC="diff vs origin/main since $BASE"
else
  TARGET_DESC="$TARGET"
fi

Refuse on main itself (no diff to review):

'/critic-review' has nothing to review on main; check out a feature branch first.

2. Operator confirmation gate

Cost estimate: one Opus invocation against the diff scope. Typical: $0.50–$2.00 depending on diff size + how much of principles.md the agent reads. Surface the estimate and wait for operator --yes or explicit confirmation.

3. Dispatch the critic agent

Invoke via the Agent tool with subagent_type: critic. Pass:

  • The full diff (git diff origin/main..HEAD).
  • A pointer to docs/harness/principles.md (the agent reads it itself).
  • A pointer to any audits relevant to the diff scope (e.g., recent .claude/audits/<date>-*).

The agent's output is a structured critique with cited evidence — see .claude/agents/critic.md for the exact contract.

4. Surface the result

Print verdict + per-principle findings to the operator. Findings are advisory at this layer (the operator decides whether to address before merge); they do NOT become a gate state file. /review-changes is the cost-class peer that produces the merge-blocking artifact.

Optional follow-up: for any P3-critical violation surfaced, suggest the operator open a CHANGE-REQUEST against the offending commit before invoking /review-changes.

Rules

  • Operator-attended. Never auto-confirm step 2. Even when the auto-trigger fires the slash, the operator must confirm before dispatch.
  • Read-only. This skill never edits the diff or the principles file. Critic-agent output is surfaced; operator decides.
  • No double-charging. If /review-changes was already run on this HEAD sha and produced a critic-style finding, this slash is redundant — surface the existing state and ask whether to re-run.
  • Per-PR scope. Critique focuses on the diff. For repo-wide principle audits, use a planning audit (a manual write under .claude/audits/).

Forbidden

  • Auto-applying fixes from the critic's findings.
  • Skipping the operator-confirmation step.
  • Dispatching the agent against main itself (no diff scope).

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.