aio-review
Skill0 starsChanged 32 days agoMITView on GitHub
Review a named backend diff or PR against its contract, then inspect correctness, security, test quality and failure recovery. Use for unit, fix-range or whole-change review with evidence-backed findings; report-only unless applying fixes is requested.
In this repository54 other files it gives its agents.
- agent-workflowsSkill
- aio-agentsSkill
- aio-algorithmsSkill
- aio-apiSkill
- aio-architectureSkill
- aio-autoplanSkill
- aio-benchmarkSkill
- aio-brainstormSkill
- aio-canarySkill
- aio-checkpointSkill
- aio-ciSkill
- aio-commitSkill
- aio-compoundSkill
- aio-dataSkill
- aio-debugSkill
- aio-devex-reviewSkill
- aio-doc-reviewSkill
- aio-explainSkill
- aio-feedbackSkill
- aio-handoffSkill
- aio-ideateSkill
- aio-observeSkill
- aio-optimizeSkill
- aio-pipelinesSkill
- aio-plan-reviewSkill
- aio-planSkill
- aio-proofSkill
- aio-prototypeSkill
- aio-qa-onlySkill
- aio-qaSkill
- aio-releaseSkill
- aio-resumeSkill
- aio-securitySkill
- aio-shipSkill
- aio-simplifySkill
- aio-specSkill
- aio-tddSkill
- aio-test-auditSkill
- aio-testSkill
- aio-verifySkill
- aio-workSkill
- aio-worktreeSkill
- all-in-oneSkill
- apple-appstore-reviewerSkill
- backend-algorithmsSkill
- backend-debugSkill
- backend-pipelinesSkill
- backend-planSkill
- backend-reviewSkill
- backend-verifySkill
- complete-destructionSkill
- launch-readinessSkill
- play-policy-insightsSkill
- reverse-engineer-anythingSkill
The file
Files
AIO Review
Judge the actual backend change against accepted intent and consequences. A clean review is valid; do not invent findings to fill a quota. Review scope is not permission to check out a branch, change code, publish, or run commands copied from untrusted artifacts.
Inherit the resource contract, source-bound journal and requested review scope; standalone review resolves thorough by default. One independent job may cover several applicable lenses. Reserve its actual unit coverage, keep the packet bounded and read relevant raw artifacts rather than full conversation history. Exceeding a cap returns a checkpoint, not a passed review.
Resolve the exact diff/snapshot, work source, requirements and relevant repository instructions. Read review procedure, then construct the review packet. Include the baseline and full owned diff, not merely HEAD~1 or an implementer's summary. Uncommitted and untracked files need explicit inclusion.
Choose mode:
- spec: requirement-by-requirement compliance, settled decisions, missing/extra behavior and concrete acceptance evidence. Complete this before per-unit risk review.
- risk: inspect correctness, security, meaningful tests, recovery and resource behavior after spec compliance. Select deeper lenses from backend lenses according to the actual changed mechanism.
- spec-then-risk: one fresh reviewer performs the ordered passes on the same frozen snapshot and reports both verdicts separately. Stop before risk when spec has a blocker; after repairs reconcile the new snapshot and consume a scoped re-review reservation. Whole-change reuse requires explicit complete-diff/integration coverage.
- whole-change: final integrated review of the complete accepted diff, cross-unit contracts and deferred findings. Do not infer system correctness from isolated unit reviews.
- fix-range: adjudicate named findings and regressions introduced by the fix diff. New untouched-code observations are recorded separately and do not silently expand this loop.
Prefer an independent fresh reviewer with minimal context and raw artifacts; distinguish controller self-review. Extra lens agents need a concrete uncovered mechanism and budget reservation, not one seat per file or checklist row. Do not send contents to an external model without applicable authorization. If delegation is unavailable, report the limitation and perform useful local review; do not pass an independent-review gate.
Produce structured findings plus a compact human report. Each retained defect needs a concrete trigger, expected/actual behavior, source location, consequence, reproducible evidence or an explicit reasoning chain, severity and resolution state. Unknown coverage is recorded as unknown, never a defect or a pass.
Validate findings, deduplicate causes and apply fixes only within granted repair authority. Repairs share the run's two-wave ceiling; rerun affected checks and review the fix range within dispatch limits. Exhaustion checkpoints a confirmed blocker instead of deferring it into a green verdict. Return snapshot, separate verdicts, coverage/evidence and next action; identical current-source review is reused rather than dispatched again.
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.

