cross-model-review
Skill1 starsChanged 5 months agoMITView on GitHub
Who wrote it does not review it — a different model reviews plans and diffs and returns a structured, cited verdict. Activates when a plan or a diff is ready for review, before any fix or merge, whenever you would otherwise self-review your own output.
5 other projects use a version of this.
Adopted by
Most recently taken up by
- makedirectory/ModelPeertheir own version, first seen Nov 8, 2026
- hyperb1iss/hyperskillstheir own version, first seen Nov 1, 2026
- NoobyGains/godmodetheir own version, first seen Nov 1, 2026
- github/levertheir own version, first seen Nov 1, 2026
- hiteshbandhu/skills-i-usetheir own version, first seen Jun 26, 2026
Other versions5 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 5
- github/lever★ 279same name
- NoobyGains/godmode★ 118same name
- hyperb1iss/hyperskills★ 36same name
- hiteshbandhu/skills-i-use★ 1same name
- makedirectory/ModelPeer★ 1same name
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 repository5 other files it gives its agents.
- branch-guardSkill
- empirical-firstSkill
- orchestrator-verifySkill
- review-loop-autopilotSkill
- worktree-isolateSkill
The file
Files
Cross-model-review
The author never reviews their own work. A different model — ideally a different vendor — reviews against an explicit checklist and returns a structured, cited verdict.
Contract
The engine that authored a plan or diff does not review it. A different engine reviews it against an explicit checklist and returns a structured verdict (APPROVE / NEEDS REVISION / NEEDS WORK) with file:line citations. Complementary blind spots between models catch a strictly larger bug class than self-review. The verdict is INPUT to the orchestrator, never final authority.
When to trigger
- A plan is drafted and about to advance — review it before locking it.
- A diff is implemented and about to be marked done — review it before any fix loop or merge.
- You are about to accept your own output without a second model seeing it — don't.
- The same model that wrote the artifact is the one you were going to ask to review it — pick a different one.
Protocol
- The author engine (engine A) produces the plan or diff.
- A different engine (engine B, ideally a different vendor) reviews it against an explicit checklist — correctness, scope fidelity, missing coverage, regressions.
- Engine B returns a structured verdict: one of
APPROVE/NEEDS REVISION/NEEDS WORK, with concretefile:linecitations for each finding, graded Critical / Important / Suggestion. - Hand the verdict to orchestrator-verify — never straight to a fix loop. The orchestrator re-reads each finding before anything mutates.
- On the next iteration keep roles fixed (A authors, B reviews) for stability — switch to alternating roles only late and deliberately (see flipped-flow). Local rule if that doc is unavailable: stay classic for iterations 1–3; only after ~3 iterations, when the remaining findings are all small / document-class (no Critical / architectural), swap so the prior reviewer authors the next fix.
Anti-patterns
- The same model reviewing its own output (no complementary blind spot — defeats the purpose).
- A free-form verdict with no
file:linecitations (un-verifiable; invites churn). - Treating the reviewer verdict as authority instead of input (skips the orchestrator gate).
- Downshifting reviewer effort silently — calibrate it to budget, but know that you did.
Worked micro-example
A structured verdict block the reviewer returns:
Verdict: NEEDS WORK
- Critical: newproject.mjs:282 — `ours-stale` branch deletes user edits on restart.
- Important: cli.mjs:1027 — descendant symlink escape not re-checked before write.
Skeleton CONFIRMED SOUND; failures are 1 data-loss path + 1 confinement gap.The orchestrator does not act on this directly — it opens each cited line, confirms or rejects, and only then forwards the verified scope.
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.