Passed 15 of 20 test cases with the skill, and 6 of 20 without it.
Without skill30%
With skill74%
Checks it passed that the model alone missed
The $claude-print skill is run in read-only mode.
The review prompt invokes the $claude-print skill.
The review workflow executes the $claude-print skill.
and 9 more in the full report
10 fixed7 passed either way6 still failing0 broke
Pass-rate lift
+43 pp
Tokens
+33%
Agent turns
0%
Response time
-29%
Model gemini-3.6-flashMeasured 8/27/2026v717 of 23 cases passed with skill · 20 comparable
Tokens, turns and time describe resource use, not a dollar price.
See it in action
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
1 of 10
Prompt
We have constructed the peer review prompt for `src/auth/oauth.py`. Which tool skill should be invoked to run this prompt and in what mode?
What was checked, without and with the skill
Checked
Without skill
With skill
The review workflow executes the $claude-print skill.
Missed
Passed
The $claude-print skill is run in read-only mode.
Missed
Passed
case-21 · gemini-3.6-flash · 8/27/2026 · answers are capped and redacted
What's inside
The instructions your model reads when the skill loads.
SKILL.md52 linessingle file
Peer Review
Independent peer review via Claude. Translates a natural-language review request into a Claude-specific prompt so invocations stay implementation-agnostic.
Step 1: Determine Review Request
Extract from the current context:
Material — code diff, files, plan, feedback, or other artifact to review
Review dimensions — correctness, security, API usage, consistency, simplicity, coverage, plan readiness, feedback interpretation, or explicit user criteria
Criteria — reference file paths Claude should read directly, inline criteria text, or the material's own domain conventions
Prior adjudications — findings already judged, one line each with its verdict and the recorded reason; optional
Output format — findings schema, verdict labels, priority scale, and any required metadata fields
If no reviewable material is available, stop and state that material is required.
Step 2: Build the Claude Prompt
Assemble a prompt for Claude:
State that Claude is acting as an independent peer reviewer for Codex.
Identify the exact material and scope. Pass text you did not author as file paths and diff commands rather than pasted content.
When the review revisits material an earlier round already covered, rebuild the review subject from the material's current content and state that any earlier version of it is superseded.
Instruct Claude to read referenced criteria files directly.
When prior adjudications were identified in Step 1, state them inline in the prompt, labeled as decisions already reached on proposed changes rather than as established properties of the material, and state that a listed finding is raised again only on evidence its recorded reason does not already account for.
Bound Claude's reads to the material under review, the sources needed to verify claims about it, and the criteria identified in Step 1. Exclude documents unrelated to those three.
Preserve independent dimensions. If multiple dimensions are requested and Codex sub-agent fan-out is unavailable, ask Claude to review each dimension in a separately labeled section. When the request instead asks explicitly for a single-pass review covering all dimensions, keep it to one pass with each dimension in its own labeled section rather than fanning out.
Require the exact output format expected by the calling skill.
Require evidence for every actionable finding.
Tell Claude not to modify files.
Tell Claude to perform the review itself rather than delegating to another peer review skill or back to Codex. The prompt has already crossed the tool boundary; further forwarding would loop.
Step 3: Run $claude-print Skill
Run the $claude-print skill in read-only mode with the assembled prompt.
Step 4: Validate Output
Compare Claude's output against the reviewed material and the dimensions and structure requested in Step 2, then classify it into one of three branches: