review-panel
Skill2 starsChanged 3 days agoMITView on GitHub
Have an artifact reviewed by reviewers whose context is separate from the authoring session, then judge every finding against the artifact by its evidence rather than by how many reviewers agree, apply the valid ones, and re-check. Also checks review comments from people. Use for "cross-check this PR with another model", "get a second opinion on this design", or "교차 검토".
13 other projects use a version of this.
Adopted by
Most recently taken up by
- daryaKuto/claude-skillstheir own version, first seen Nov 15, 2026
- JeffreyFerreiras/ai-skillstheir own version, first seen Nov 1, 2026
- rosslevinsky/portable-agent-skillstheir own version, first seen Oct 15, 2026
- gigio1023/agent-skillsword for word, first seen Oct 13, 2026
- raine/consult-llm-mcptheir own version, first seen Oct 8, 2026
- HACK-WU/skillstheir own version, first seen Sep 25, 2026
Other versions13 projects use a version of this.
Word for word · 1
- gigio1023/agent-skills★ 2100% same
Edited · 0
None found.
Wrote their own version · 12
- raine/consult-llm-mcp★ 148same name
- 9thLevelSoftware/legion★ 76same name
- iamneilroberts/claude-skills★ 26same name
- rosslevinsky/portable-agent-skills★ 12same name
- HACK-WU/skills★ 3same name
- josherau/claude-operating-core★ 3same name
- kbichave/skills★ 2same name
- JeffreyFerreiras/ai-skills★ 1same name
- patmagee/skills★ 1same name
- daryaKuto/claude-skills★ 0same name
2 more not listed
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository53 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- 1password-cliSkill
- codex-delegateSkill
- credo-journalSkill
- credo-paper-planSkill
- credo-releaseSkill
- credo-tasteSkill
- cursor-cli-delegationSkill
- english-prompt-reviewSkill
- mermaid-diagramsSkill
- pdf-page-countSkill
- pr-review-commentSkill
- read-agent-sessionsSkill
- reddit-searchSkill
- release-notesSkill
- small-model-handoffSkill
- spot-gpu-evaluationSkill
- toss-portfolio-stateSkill
- vuln-reportSkill
- zcode-cliSkill
- bug-diagnosisSkill
- coding-standardsSkill
- commit-and-pushSkill
- copydeskSkill
- design-docSkill
- design-explorationSkill
- draft-prSkill
- evaluation-analysisSkill
- evaluation-auditSkill
- evaluation-designSkill
- evaluation-operationsSkill
- find-unknownsSkill
- frontend-designSkill
- git-worktree-setupSkill
- idea-critiqueSkill
- insight-dashboardSkill
- install-skill-packSkill
- internal-source-researchSkill
- interviewSkill
- literature-researchSkill
- load-testingSkill
- ml-experimentSkill
- model-routingSkill
- orchestrate-subagentsSkill
- praxis-directionSkill
- prompting-guideSkill
- session-handoffSkill
- skill-builderSkill
- terminologySkill
- work-planSkill
- write-issueSkill
- write-notionSkill
The file
We split this skill into the rules it gives an agent
3 rules. 0 are common, 3 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 11 edited versions we read keep these.
3
Distinctive
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 37
Keep the lead's conclusions out of the packet.
- Distinctive: line 41
Verify each material finding in the artifact: open the cited location, read the governing rule or intent source, and run the cheap check …
- Distinctive: line 49
Read each reviewer's declined list as a coverage map: an item one reviewer declined and no other judged is a part nobody reviewed.
“Common” is measured: kept by at least 6 in 10 of the 11 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Review Panel
Have an artifact reviewed by reviewers who did not write it and do not share the authoring session's context, then decide what each finding is worth by checking it in the artifact.
Findings are judged by evidence in the artifact, never by how many reviewers raised them. Reviewers who agree may share one premise: the same packet, the same stale document, the same misleading comment, or similar training. A single reviewer's finding whose evidence holds outranks two that agree without it.
Reviewers
An explicit user choice of reviewers, models, or count governs. Otherwise the lead picks the reviewers and their number from what is reachable and what a missed defect would cost. Reviewers from different model families are less likely to share blind spots; a separate subagent on the session's own model is also acceptable when it receives only the packet and the artifact, not the conversation. Say which reviewers ran.
Independence is what the panel buys. A reviewer that saw the authoring conversation is not independent, and neither is a review the session performs itself; if no separate context can be started, say so and label the review as not independent. If a requested model cannot be reached, say so rather than substituting silently.
Reviewers are read-only: no edits, commits, posts to a forge or tracker, state-changing or budget-spending runs, or network unless granted for this review. A reviewer writes only the report path its packet names. The packet and the files a reviewer reads go to that model's provider, so for an external provider include only what the user's data policy allows.
Sizing and Axes
Size the review by asking whether a wrong version of this change would fail loudly where it is made or silently somewhere else. A crash, type error, or failing check at the changed line shows itself on the next run, so light coverage usually suffices. A silent defect passes every check at the change site and surfaces later: a guard that still passes but no longer guards, an auth, money, or data boundary, a public interface or serialized format other code depends on, a reported number, or retry, timeout, ordering, and concurrency behavior. Give silent or uncertain changes more coverage, and have the packet name where the failure would surface (callers, consumers of a format, the metric pipeline) so reviewers read those too. Designs and evaluations are mostly silent until results arrive.
For code and designs it often helps to review two axes in separate contexts: fidelity (does it do what the request, issue, or design asked) and standards (does it meet the applicable rules). Kept apart, a reviewer who finds the code clean does not take that as evidence it is correct. Fidelity needs an intent source at a named revision; if none exists, state the intent as the lead understands it and mark it as a hypothesis to test. Criteria by artifact has criteria to quote into a packet.
Packet
Send the same packet to every reviewer on an axis, so differences come from the reviewers. The packet template is a starting point. A packet says what to review (paths, revision or diff range, out of scope), why the artifact exists and for whom, the criteria, current rules that override older records (so compliance with a newer decision is not flagged), the limits, and the response wanted: findings ranked by impact with path:line evidence and a concrete fix, parts checked and found sound, a short list of what the reviewer declined to judge and why, and a verdict for the stated purpose.
Keep the lead's conclusions out of the packet; a suspicion goes in only as a hypothesis to test. Do not pass one reviewer's findings to another in the first round.
Adjudication
Verify each material finding in the artifact: open the cited location, read the governing rule or intent source, and run the cheap check that would show the problem. A reviewer's citation is a lead, not proof.
- Valid: the problem exists as described. Apply the proposed fix or a better one.
- Partially valid: the problem is real but its cause, scope, or fix is wrong. Record what holds and apply the corrected fix.
- Rejected: the problem does not exist at this revision, the cited location says otherwise, or it conflicts with a newer user decision. Record the reason and evidence.
A finding that needs a measurement or access the review lacked stays open with the check that would settle it. A finding arguing that a current rule is wrong goes to the user as a question.
Read each reviewer's declined list as a coverage map: an item one reviewer declined and no other judged is a part nobody reviewed. Cover it with a check of your own, another reviewer, or an open item, and never read silence as approval.
Merge duplicates into one finding listing every reviewer who raised it; its judgment comes from the check alone. Rank by impact on the artifact's purpose, not by agreement. Keep rejected findings with their reasons so a later round does not reopen them without new evidence. A judgment table (finding, raised by, evidence checked, judgment, action) works well; the worked example shows its depth and the line between partially valid and rejected.
Apply and Re-check
Show the judgment before applying when the user asked to see it or a fix is consequential: it changes behavior beyond the reviewed change, an interface, a design decision, or published text, or is hard to reverse. Otherwise apply valid fixes within the scope already granted and re-run the checks that cover them. A review-only request, or a review of someone else's artifact, returns the judgment and proposed fixes without editing. A second round is worth it when fixes changed structure or behavior broadly.
Review Comments From People
Comments from a human reviewer follow the same rule: verify each against the artifact before implementing it. Ask about unclear comments before implementing related ones. When a comment is wrong for this artifact, answer with the evidence. A comment that conflicts with an earlier user decision goes to the user. Reply with the fix or the evidence, without performative agreement. Posting replies or resolving threads on a forge or tracker needs the user's approval.
Report
Lead with what the review means for the artifact's next step, then the judgments, the changes applied, the checks re-run, open items and questions for the user, and the panel as run, including declined items no reviewer judged and any loss of independence.
Adapted sources and licenses are listed in sources.
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
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.

