review
Skill2 starsChanged 3 days agoMITView on GitHub
Use for independent reviews of plans, code, UI, backend and releases, including review-loop participants and think's stop-condition check. Code review must differ in lineage from the code authors.
192 other projects use a version of this.
Adopted by
Most recently taken up by
- dzhng/skillstheir own version, first seen Nov 15, 2026
- muxy-app/muxytheir own version, first seen Nov 15, 2026
- macro-inc/macrotheir own version, first seen Nov 15, 2026
- yagizdo/quivertheir own version, first seen Nov 15, 2026
- markmhendrickson/neotomatheir own version, first seen Nov 15, 2026
- felipeam86/pipe-skillstheir own version, first seen Nov 15, 2026
Used with
Projects with it also use close-spec, design-with-images and 10 more.
Other files people use alongside this one.
Files found in the same public projects as a version of this one, more often than they turn up anywhere else. Each project counts once, large collections count for less, and this file's own repository is left out.
dzhngSkillclose-spec
When a spec under specs/<feature>/ is done shipping (a write-spec build, or any planned task that produced a spec), archive it to specs/done/ and rewrite it from a build-plan into a durable rationale — the why, the principles, the invariants — pointing back to the real code for the how. Use when implementation has landed and the plan no longer matches what shipped, or the user says a feature/spec is finished. Pairs with [write-spec](../write-spec/SKILL.md) (the plan this closes) and [review](../review/SKILL.md) (run before closing).
Also in 5 projects that use this
dzhngSkilldesign-with-images
Explore UI and visual design with image generation, then iterate the real implementation against the user-selected concept. Use when brainstorming visual directions, offering mockup variants, or implementing an approved generated design; do not stop at the mockups or the first code approximation.
Also in 5 projects that use this
LambenthanSkilldaily-arxiv
每日从 arXiv 拉取新论文,过滤相关性,auto-ingest 高优先级论文,检测 SOTA 更新
Also in 5 projects that use this
LambenthanSkillexp-eval
实验判决门:Review LLM 独立评判实验结果 → 4 种判决路径 → 自动更新 claims confidence、ideas status、graph edges
Also in 5 projects that use this
- +3 more from dzhng
- +5 more from Lambenthan
Other versions192 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 192
- garrytan/gstack★ 142ksame name
- ansible/ansible★ 71ksame name
- docling-project/docling★ 71ksame name
- traefik/traefik★ 66ksame name
- ClickHouse/ClickHouse★ 51ksame name
- temporalio/temporal★ 24ksame name
- ccch1mneyyy/dsh-Go★ 24ksame name
- tisfeng/Easydict★ 15ksame name
- juicedata/juicefs★ 15ksame name
- elie222/inbox-zero★ 13ksame name
182 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 repository17 other files it gives its agents.
- AGENTS.mdAGENTS.md
- agent-tuneupSkill
- babysitterSkill
- backlog-burnSkill
- challengeSkill
- choose-designSkill
- design-loopSkill
- focus-groupSkill
- github-writingSkill
- implementSkill
- plan-reportSkill
- raise-issueSkill
- raise-prSkill
- review-loopSkill
- shipSkill
- test-writerSkill
- thinkSkill
The file
We split this skill into the rules it gives an agent
10 rules. 0 are common, 10 are distinctive, and 0 are very specific to this project.
0
Common
Most of the 157 edited versions we read keep these.
10
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 12
Read project settings at start.
- Distinctive: line 29
Run the required parallel companions on the same frozen revision.
- Distinctive: line 60
If required code-author identities cannot be verified, no eligible reviewer is available, or required dispatch is prohibited, report the …
- Distinctive: line 95
When reviewing visible behavior:
- Distinctive: line 97
Use the host's supported browser or device tools.
- Distinctive: line 102
When verifying a fix, retain the original capture and take a fresh capture of the same affected state on the fixed revision.
- Distinctive: line 113
If a required state cannot be captured, finish independent checks and identify the missing state, reason and remaining verification.
- Distinctive: line 120
Use the shared incident SEV scale for review findings, including pre-release bugs.
- Distinctive: line 131
Do not force stylistic preferences into bug findings or inflate severity because a fix is easy.
- Distinctive: line 172
If cross-thread access fails, report it and arrange an authorized handoff.
“Common” is measured: kept by at least 6 in 10 of the 157 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 against the goal
Review the requested scope against its requirements. Cross-lineage restrictions apply only to written code and its material model authors. The orchestrator's lineage is irrelevant unless it also wrote that code. Plans, prose and visual judgments have no lineage exclusion. Read project settings at start. This skill owns review evidence and severity; it grants no repair, filing or release authority.
Obtain an independent review
The coordinator obtains review; assigned reviewers never delegate. Apply T3 dispatch preflight before T3 dispatch.
- Identify the scope, exact revision or decision brief, acceptance criteria, every model that materially authored the reviewed code, including repairs. Distinguish human authors from unknown model identities.
- Discover supported host tasks and exact model identities. Code reviewers, including OCR backends, must differ in lineage from every material code author. Versions, effort settings and provider aliases do not change lineage. Record identities and lineage evidence. Apply reviewer model selection to every review, recheck and additional opinion, including inside implement. Preserve named choices and apply its authorised credit/lineage fallbacks and holds. Run the required parallel companions on the same frozen revision. Extra plan or visual judges may share any lineage; their judgments cannot replace independent review of authored code.
- Reuse completed reviews only when eligible against all current code authors,
including repairers, and covering the current head or decision brief with all
required companion reports. Obtain missing companion coverage before reuse.
Otherwise obtain a fresh read-only review through authorized host orchestration.
Supply the request, scope, source/version, evidence, criteria and model identities.
Apply the reviewer brief for fixed source,
permitted checks, scratch ownership and cleanup. Give the reviewer the
absolute path of the assigned-reviewer card
and require reading it; a reviewer does not need this skill.
In T3 Code, discover models with
orchestrator_capabilities, use a newdelegate_taskfor every review or recheck that the coordinator dispatches, record itstaskIdand follow its returnednext. Use native orchestration elsewhere. Respect caller delegation limits. OCR runs through its library CLI, not host-agent model dispatch. External reviews must meet these identity, scope and reporting rules; otherwise report the specific review gate. - For adopted
review_rounds: 1, apply the single-round procedure before repairs. Otherwise follow the caller's stopping rule across repairs and resumes. Review-loop and design-loop use score targets and two-round stagnation without a required cap; their evaluations do not inherit the standalone limit. A standalone review allows at most three rounds, including failed attempts. On stalling or exhaustion, apply the budget diagnosis. Recheck material changes or unresolved findings; do not seek unanimity or restart budgets by changing reviewer. Receive the complete report before treating the review as completed. A timeout, missing reply or task status alone is not a review.
If required code-author identities cannot be verified, no eligible reviewer is available, or required dispatch is prohibited, report the specific review gate and hold work that depends on it. Finish independent preparation. Self-review cannot satisfy the code gate. Matching the orchestrator alone never creates a gate or needs a waiver. Required human approval remains a separate gate.
Reviewer and parent responsibilities
Children read a role card, not this skill. Each card restates these rules for one role and links to nothing else:
- An assigned reviewer follows the assigned-reviewer card. For code review it checks code-author identities against its own lineage, reports a missing or conflicting identity instead of claiming eligible review, reviews directly and never dispatches agents. It returns findings with severity and evidence; it does not fix code, decide deferrals, file issues or change labels.
- An implementation participant follows the implementation-participant card. It applies the parent's selected dispositions and never claims independent review or dispatches reviewers.
A later explicit repair assignment is separate work. For separately authorised GitHub reviews, apply GitHub writing.
The parent validates findings and decides whether to fix, defer, reject or raise issues within its existing authorization, under feedback dispositions. Validating findings and repairing code does not make the parent an independent reviewer. For a direct user request, the coordinator obtains the independent review and reports it to the user.
Capture what was evaluated
Bind evidence to the reviewed revision, relevant dirty changes and environment. Identify the application build or deployment when it differs from the source. Record the flow, state, viewport or device and conditions needed to interpret it.
When reviewing visible behavior:
- Use the host's supported browser or device tools. Capture and inspect screenshots of each reviewed flow's key states, including passing states. Include relevant loading, empty, error or responsive states when in scope.
- Capture every visible finding. Explain the expected behavior, what the image shows and why it matters. A screenshot alone does not prove a functional cause.
- When verifying a fix, retain the original capture and take a fresh capture of the same affected state on the fixed revision. Do not reuse an old image as proof.
- Give captures stable evidence IDs. Store them in the host-approved artifact location, using a verified ignored directory when repository-local storage is required. Avoid secrets and personal data in captures and reports.
For backend/code findings use reproductions, tests, logs or file/line evidence, not terminal screenshots. Apply proof quality when adopting bundles, claiming HTTP access or diagnosing unavailable tools. Plan-only reviews need no screenshots.
If a required state cannot be captured, finish independent checks and identify the missing state, reason and remaining verification. Mark visual review incomplete and carry that gap into the release recommendation. Do not substitute a mockup for observed behavior or claim that passing checks cover the gap.
Assign severity
Use the shared incident SEV scale for review findings, including pre-release bugs.
| Severity | Impact | Default release treatment |
| SEV1 | Essential functionality unavailable or comparably severe impact, including serious security exposure or irreversible data loss | Blocks release; an exception requires the user's explicit approval |
| SEV2 | Degraded functionality with a workable alternative or workaround | Parent evaluates release requirements and the workaround |
| SEV3 | Minor defect with low or no user impact | Normally allows release |
Justify severity using impact, affected users, triggering conditions and existing mitigations. Rare failures can still be severe. Separate confirmed observations from hypotheses; uncertainty is an evidence gap, not a reason to lower impact. Do not force stylistic preferences into bug findings or inflate severity because a fix is easy. When a consequential defect has no workaround, assess whether essential functionality is unavailable rather than automatically calling it minor.
Priority describes scheduling and remains separate from severity. Deferral or
issue creation does not lower severity. Report the proposed severity even if
the repository lacks a matching label. Canonical target names are
severity:SEV1, severity:SEV2 and severity:SEV3; verify they exist before use.
The parent checks current labels and
maps equivalent meanings; do not equate SEV1 with priority:P1 or invent a label
mapping. Label creation and migration belong to separately authorized work.
Track deferred bugs
For every confirmed bug intentionally left unfixed, the coordinator must obtain a verified issue URL before declaring an execution workflow complete. Apply deferred-bug tracking for deduplication, issue contents and read-back verification. This includes non-blocking SEV3 bugs and approved release exceptions. Issue creation does not lower severity or waive a blocking finding. Reviewers report findings; the coordinator files within its authority. If filing is blocked, preserve the draft and report tracking as incomplete. Rejected findings and stylistic preferences require no issue. Already non-blocking bugs may ship while filing is pending under that procedure; the workflow stays open until required issue URLs are verified.
Return a usable report
Include scope, revision, checks, reviewed flows and coverage gaps, even when no bugs are found. Record material code-author and reviewer model identities, verified lineages and any unknowns. For each finding provide:
- Stable finding ID, SEV level and rationale.
- Location, expected versus observed behavior and user impact.
- Reproduction steps or supporting evidence, with uncertainty stated.
- Evidence IDs, screenshot captions and accessible artifact paths or links.
Dispatched reviewers return a complete plain-text report with artifact references the parent can actually access. Follow host presentation rules; never leave the only findings in a child-thread visualisation. The parent must show the user the relevant screenshots with captions through supported image or presentation tools, including key passing states. A directory path alone does not show the evidence. If cross-thread access fails, report it and arrange an authorized handoff.
The parent retains finding IDs and original evidence, and appends dispositions, reasons, issue URLs and verification results. Keep artifacts available for that handoff; before task cleanup, preserve any evidence needed for deferred issues in an authorized durable destination. Do not upload private captures merely because an issue may eventually be filed.
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.

