Review Checklist logo

Review Checklist

Organization
devcodex-labs
review-checklist

复审清单整理与审查规范 — 创建、冻结、证据执行、状态更新与收敛关闭

Overview

Publisherdevcodex-labs
Repositorydevcodex
Skill namereview-checklist
Stars
433
Forks
44
Bundled files
1
LicenseAGPL-3.0
Links
  • Markdown instructions

    A SKILL.md file the model loads on demand, so it only costs tokens when a request actually matches.

  • Works with any LLM

    AI skills are plain Markdown, not provider-specific code, so this works with GPT, Claude, Gemini, Grok, or a local model.

  • 1 bundled files

    Scripts, templates, and references the model can read while it works. Files are read-only and never executed.

  • Open source

    Published by devcodex-labs on GitHub. Read the source before you install it.

Installation

Install the Review Checklist AI skill in TypingMind to use it with any LLM, or drop it into another agent that reads SKILL.md.

1

Install in TypingMind

TypingMind installs a skill straight from its GitHub folder — it reads SKILL.md, bundles the resource files, and stores the result locally.

  1. Open the app and go to Plugins → Skills.
  2. Choose "Install from GitHub".
  3. Paste the skill folder URL below and confirm.
    https://github.com/devcodex-labs/devcodex/tree/main/content/skills/review-checklist
  4. Enable the skill in any chat where you want it available.
Plugins → Skills → Add skill → From GitHub URL, then paste the folder URL and press Continue.
2

Install in another agent

Any agent that reads the Agent Skills format can use this skill — copy the folder into that agent's skills directory.

Claude Code — .claude/skills
git clone --depth 1 https://github.com/devcodex-labs/devcodex.git /tmp/devcodex
mkdir -p .claude/skills
cp -r /tmp/devcodex/content/skills/review-checklist .claude/skills/review-checklist
Restart Claude Code after copying so it picks up the new skill.

Use it in TypingMind

Enable Review Checklist in any TypingMind chat and the model takes it from there. Its name and description sit in the system prompt, and the moment a request matches, the model loads the full instructions itself — you never invoke it by hand, and it costs no tokens until it is actually used.

The model loads Review Checklist on its own as soon as a request matches it.

Works with any AI model

AI skills are plain Markdown instructions rather than provider-specific code, so Review Checklist is not tied to the model it was written for. Install it once in TypingMind and use it with GPT-5, Claude, Gemini, Grok, DeepSeek, Mistral, Llama, or a local model you run yourself — all on your own API keys.

  • Loaded only when it is needed

    The system prompt carries just the name and description. The instructions are fetched on the first matching request, so an idle skill costs nothing.

  • Switch models mid-chat

    Because the skill is instructions rather than code, changing model does not break it — the next model reads the same SKILL.md.

Skill instructions

This is the SKILL.md content the model loads. Read it before installing — a skill is instructions your model will follow.

Review Checklist Skill

职责

当任务需要正式复审、ECR、发布前复审、多轮收敛审查、外部 finding 批次处理或用户要求“复审直至收敛 / 按清单复审”时,本 Skill 负责创建和维护复审清单。

复审清单是执行真相源之一,不是报告里的临时段落。没有清单文件、没有逐项证据、没有状态更新或没有遗漏逃逸分析时,不得宣告复审已收敛。

触发条件

场景是否触发
用户要求复审、再次复审、收敛复审、全维度复审必须
dev/fix 的 ECR 执行闭环复审涉及控制面、多文件、发布、模板、validate 或部署副本必须
外部审查报告、AI review finding、audit issue、代码评审发现进入批量处理必须
发布前审查、release readiness、tag/publish 前风险复查必须
CP 确认后命中高风险、多模块、公开契约、配置、安全、package、docs consumer、控制面或多真相源场景必须
低风险单文件 typo 且无正式复审要求N/A + skipReason

清单文件要求

字段要求
reviewScope本轮复审对象、目标版本、active requirement/task/bug id
sourceAnchors需求、技术方案、用户文档、审查报告、finding 来源或 release 目标
frozenChecklist冻结后的检查项,不允许复审中静默删除
dimensionSet本轮维度集合;R2+ 必须体现 ReviewDimensionDeltaGate
evidence每项绑定命令、代码落点、文档路径、页面、截图、构建产物或反向缺席扫描
statustodo / running / passed / failed / blocked / N/A
skipReasonN/A 或未执行项必须写明原因、残余风险和替代证据
escapeRecords复审或实施中发现原清单遗漏时的追加记录,字段见 ReviewEscapeRecordGate
closure收敛结论、未关闭项、下一步和报告引用
evidenceLedger多批次或矩阵验证命中 BatchEvidenceLedgerStateGate 时,冻结 baseline、actualSources、commands、status、finding / skipReason
progressCard多批次命中 BatchProgressCardGate 时,记录 totalScope、completed、currentBatch、nextBatch、remaining、blockers、evidenceLinks
coverageClaims按 inventory/machine/manual/sample/executed 分级;强覆盖声明绑定 FileEvidenceLedger

必执行门禁

ReviewExecutionPlanV1、ReviewEvidenceReceiptV1、ReviewStateSnapshotV1、EvidenceSaturationResultV1 与 StageTimingV1 的确定性行为由 hooks/_runtime/review-execution-contract.cjs 拥有。本 Skill 负责人读清单与流程;checklist、report、memory、progress 和 final 状态只能投影同一个 snapshotDigest,不得分别手写状态。

PR-1 的机器投影写入任务内 .memory/review-execution-pr1.json(ReviewExecutionEvidenceV1),包含 plan、receipts[]、saturationInput、snapshot,以及 candidate:{path,digest}、review:{path,digest}、artifacts:[{ref,path,digest}]。路径均相对唯一 task-root,摘要来自实际文件回读;candidate 绑定当前 02 方案,review 绑定独立 03 复审。先完成实际语义复审、写入报告与证据,再由上述 owner 函数生成 plan/receipt/saturation/snapshot;不得用示例值、历史结论或自填通过状态替代实际复审。Stop 验证所有摘要、回执及相关证据;正文长度、固定章节词和通过字样不产生资格。未知或陈旧状态继续诊断为 UNVERIFIED,保持 advisory。

  • ReviewChecklistPrecreationGate:正式复审前先创建清单文件或明确复用已有清单。
  • ChecklistFreezeFileGate:开始执行前冻结清单范围、维度和来源锚点;新增项只能追加,不得静默改写已冻结项。
  • ReviewChecklistCompletenessGate:每个清单项都要有状态、证据或 skipReason。
  • EvidenceExecutionGate:不能只按审查报告文字验收;关键结论必须做本地代码、测试、构建、页面、文档或配置证据核验。
  • BlockerSnapshotCompletenessGate:同一复审阶段存在多个安全独立检查时,先完成该阶段并冻结完整 blocker 快照,再统一修正;若因 invalid premise、破坏性副作用或证据污染提前停止,必须记录 stopReason、skippedChecks 和恢复入口。
  • RequiredCandidateEvidenceGate:CP1/CP2 候选进入确认或复审清单时,记录 CandidateReviewBundleV1 分类与缺失项;CP1 缺 RQMatrix / DomainRealityMatrix / ClaimEvidenceMatrix / EscapeAbsorptionQueue,或 CP2 缺 TDMatrix / BlockerSnapshot / ClaimEvidenceMatrix,不得标本轮复审收敛。
  • SampleIssueExpansionGate:用户给出样例问题时,样例只能作为 seed evidence;正式复审必须先展开全维度图,标明样例覆盖 / 未覆盖维度,再冻结清单。
  • ReviewAnchorMaterializationGate:PR / TD / CP2 / 发布前审查锚点必须物化为可 grep 的清单项、章节或表格,不能只写“已语义覆盖”。
  • RequirementDimensionBindingGate / RequirementPriorityAndPhaseGate:需求维度进入复审清单时,必须绑定 CP2、批次计划、验收证据和阶段关闭规则;多阶段项写 entry / exit / carryOver / closeRule。
  • ValidationLifecycleTraceabilityGate(GR-044):分阶段/状态机需求必须证明独立验证阶段(ValidationPlanV1 / BatchValidationResultV1 / GlobalValidationResultV1);验收矩阵不能替代 validate→accept 边界;无结果或 fail/inconclusive 不得 final。
  • PhaseDeliverySemanticGate:多阶段/路线图复审冻结 phaseKind、planningCoverage、sourceDelivery 与 OriginalIntentReverseTrace;不得把规划覆盖误报为源码交付。
  • ChecklistEscapeAnalysisGate:发现遗漏或返修时,分析为什么上轮清单、维度或探针没覆盖。
  • ReviewEscapeRecordGate:二次复审、返修或实施过程中发现新问题逃逸时,必须先追加 escape record,再补清单和重跑证据。
  • ChecklistStateFreshnessGate:最终报告前核对清单状态、报告结论、audit-state、sessions、SUMMARY 和 dirty 边界一致。
  • ReviewDimensionDeltaGate:R2+ 复审不得机械重复同一维度;重复维度必须有阻断项回归、高风险锚点、新证据或抽样理由。
  • PostConfirmationReviewScopeGate:CP1 / CP2 / CP3 确认后先按 C19↔R 映射判定强度(轻量=R1 / 标准=R2 / 全面=R3 / 发布安全=R4),输出 ReviewGradeCard(reviewScope·stage·cpPhase·riskClass·riskFlags·reviewClass·c19Label·contentPack·blockers·result·skipReason);命中 R3/R4 或高风险场景时必须创建或复用本清单,R1 降级写 N/A + skipReason。默认 post-confirmation/ECR 不得无理由落在 R1。
  • FeatureChecklistEvidenceMatrixGate:需求维度、功能清单或公开能力进入复审时,必须把 capability group × evidence surface 绑定到当前证据;复审清单记录验证状态,不替代稳定 Profile feature inventory。
  • BatchEvidenceLedgerStateGate:多批次、矩阵验证、长链路吸纳或发布前检查必须冻结 EvidenceLedger,区分 baseline-confirmed、executed-passed、partial、failed、not-started,且每项有 actualSources、commands、status、finding 或 skipReason。
  • BatchProgressCardGate:多批次最终报告、记忆和回复必须同步 Progress Card,覆盖总范围、已完成、当前批、下一批、剩余项、阻塞/风险和证据链接。
  • ChecklistStateMaterializationGate:每轮 clean、streak 增加或 closed 声明前,必须重开当前清单,以同一个 ChecklistStateSnapshot 原子核对 header、冻结项、轮次表、Evidence Ledger、Progress Card、Closure 六区块的 currentRound / zeroFindingStreak / currentBatch / remaining / blockers / openFindings / closureState。任一区块 stale、缺字段或互相冲突时,本轮无效且 streak 不增加;修正后必须从受影响轮次重新执行,不能只改文案。
  • ReviewExecutionPlanningGate:复审执行前以 workflow/stage/userIntent/candidate/changed+affected/risk/claims/dimensions/evidence/exclusions 形成 ReviewExecutionPlanV1;未知 stage/risk/scope、changed 不属于 affected closure 或身份无效时 fail closed 为 R4/full-required。文件少不得降低控制面、公共契约、安全或发布风险。
  • ReviewEvidenceFreshnessGate:receipt 必须绑定 plan/candidate/stage/scope/dimension/claim/lens、rule/skill/probe、impact graph、command/environment/runId、dependency/consumer/intent/risk digest;failed/inconclusive/open/blocker 或任一绑定变化均 stale 且不可复用。清单结论含“已验证 / 推荐 / 可确认 / 完整 / 采纳外部 finding”等 strong claim 时,追加 evidence-freshness:生成或引用 ClaimEvidenceIndexV1、EvidenceFreshnessReceiptV1、StaleEvidenceLintDecisionV1;summary-only 只能作为 navigation hint。
  • EvidenceSaturationGate:适用 dimensions/claims/fresh evidence 完整、UnreviewedRelatedSet 为空或逐项具备 reason/authority/upgradeCondition、open/blocker/stale=0、独立/负向/fallback/dirty boundary 达标才通过。稳定 receipt 按内容身份抽样 5%(最少 3、最多 20);任一 mismatch 立即 full-required。
  • fallbackEvidencePassed 核验本阶段实际需要的回退证据:CP1/CP2/CP3计划审查应核对回退/安全停止设计、允许范围、触发条件、现有恢复点及可行性,不执行未来尚未授权的实现/数据回滚;版本化文档修订可用实际拒收错误候选、旧确认和业务字节保持等已发生的安全停止证据。implementation/ECR或release命中的运行/数据变更,仍需该阶段适用的实际回退/恢复验证,不得以计划或未来承诺代执行。每个passed必须绑定对应当前主张和真实证据,不能只改布尔值。
  • 无实现、无运行状态的只读文档 audit:不得伪造回滚执行或把 fallbackEvidencePassed 改为 true。确实不适用时,向 evaluateEvidenceSaturation 传 fallbackApplicability: ReviewFallbackApplicabilityV1:status=not-applicable、kind=read-only-document-audit,精确 planId/candidateDigest/scopeDigest,四项 implementationPresent/runtimeStatePresent/mutableConsumerStatePresent/sourceMutationAuthorized=false,并填写 reason/residualRisk/upgradeCondition。计划须事先包含 dimension=fallback-applicability、claim=no-runtime-fallback-target;evidenceReceiptDigests 引用 1~8 条同候选、同范围、当前新鲜且通过的对应 receipt,其 evidenceRefs 必须指向实际只读现状证据。Owner 只允许 audit/audit、changedSet=[]、low/normal、非 R4、风险 flags 仅 public-contract/docs-consumer 或空;有实现/运行/可变消费者、Source mutation、控制面、未知风险及 full/security/release 均不得豁免。普通 exclusions、N/A 标签或空证据不满足此门禁;独立、负向、dirty boundary、轮数及其他质量要求保持适用。
  • RepairCollaborationAcceptanceGate:repair task 的清单必须绑定 repairClass / contractState / authorizationEvidence;full 合同逐项核对 findingToPatchMap、handoffIntegrity、independentReReview 和 acceptanceMatrix,禁止补丁产出者以唯一证据关闭高风险项。
  • RepairPreventionAssessmentReviewGate:所有 repair 的清单在 accepted 前核对 RepairPreventionAssessmentV1 的 mode/decision、双根因、regression seeds、negative cases、Owner/consumers、immediate closure、prospective status 与 rollback/sunset;repeat escape 的旧控制复用、无理由 no-new-control、当前重跑晋级 prevention 均为 blocker。
  • ReviewCoverageClaimIntegrityGate:逐文件/逐服务/全量深读声明必须有 FileEvidenceLedger;抽样必须公开 sampledSet、unreadSet、sampleMethod 与 inferenceBoundary。
  • ReworkPreventionHandoffGate:escape 属于原确认范围且已越过目标门禁时,交给 rework-prevention-engineering 分类 ReworkEvent/cluster,并把 prevention 注册到后续可比较任务;本任务重跑通过不能单独证明预防有效。
  • CandidateDiffCompletenessGate:commit/tag/publish 前,复审清单必须把授权范围物化为 staged candidate snapshot,并记录 cached diff check、name-status、secret-shape scan 与 intended scope 对账;普通 working diff 不得作为未跟踪文件已覆盖的证据。
  • PostStageDerivedArtifactFreshnessGate:存在 portfolio、索引、生成代码、文档清单或其他 tracked consumer 派生资产时,清单必须在 staged snapshot 物化后执行 Owner 提供的 candidate check,并在 commit 后对 clean target tree replay;记录 input digest、candidate identity、staged/post-commit 两次结果,禁止用“生成器刚运行过”代替提交态新鲜度。
  • ReleaseEfficiencyControlGate:发布清单记录 candidate generation/freeze、关键路径预算模式、evidence reuse/invalidations 与 release rework incident;无基线预算只能 advisory。
  • ConsumerDesignFitnessRepairGate:独立消费者验证命中时,清单同时核对 DesignFitnessMatrix 和 ValidationFindingRepairLoop;source mutation 后旧 evidence 必须 stale。

执行步骤

  1. 识别复审目标和 active 范围,列出来源锚点。
  2. 创建或定位清单文件,并写入 reviewScope、sourceAnchors、dimensionSet。
  3. 冻结 frozenChecklist,标记每项初始状态为 todo 或 N/A。
  4. 按项执行证据核验:命令、grep、代码落点、生成产物、页面、API、pack/install、registry/tag 或项目等价证据。
  5. 每完成一项立即更新状态;失败项写修复建议和阻断级别。
  6. 返修或发现遗漏时先执行 ReviewEscapeRecordGate,追加 escape record,再执行 ChecklistEscapeAnalysisGate,说明遗漏原因、补充清单和补充探针。
  7. 收敛前执行 ChecklistStateFreshnessGate,确认清单、报告、记忆、SUMMARY 和台账状态一致。
  8. 从唯一 ReviewStateSnapshotV1 投影 checklist/report/memory/progress/final,并记录互斥 StageTimingV1;模型 reasoning 不可观察时写 N/A。
  9. 最终报告引用清单文件路径,未关闭项不得被隐藏。

ReviewEscapeRecordGate

逐条件证据与双状态

复合要求必须从原始已确认来源拆出全部独立条件,先冻结分母,再逐条件核对实际消费者、设计落点、拒收与执行证据。一个证据可支持多个条件,但不得把条件合并消失;标题、文件存在或一个成功样例不能证明all-of满足。语义Owner负责检查分母完整性、证据真实性与当前性;机器结构通过不能替代此判断。

需要机器消费时使用同一可选 ConditionEvidenceMatrixV1:schemaVersion / applicability(applicable|not-applicable) / skipReason / expectedConditionIds / rows。适用时ID列表非空且唯一,每个ID恰有一行,不得遗漏、孤儿或重复;整体不适用时ID和rows均为空并给理由。每行含 conditionId / sourceRef / applicable(boolean) / skipReason / consumer / designEvidenceRefs / designStatus / executionStatus / executionEvidenceRefs / verificationResponsibilities / executionContext / limitation。

designStatus=covered|partial|unreviewed|not-applicable与executionStatus=not-run|running|passed|failed|unknown|not-applicable独立:covered须有当前设计证据与真实consumer,passed须有真实执行引用与含候选、环境、版本、run的上下文及限制;covered+not-run合法,运行passed也不自动补齐设计partial。applicable=false要求两状态均not-applicable且有skipReason。未运行写清原因与限制,不把N/A当未运行。责任是 test-router定义的D/A/I/H,可多选,不是证据等级。

对象声明只允许自有顶层属性 conditionEvidence;Markdown只认围栏外唯一保留H2 ## ConditionEvidenceMatrixV1,其section内恰有一个json围栏,至下一真实H2结束。其他章节的教学JSON、嵌套属性及围栏中的示例标题不声明本次记录。显式坏声明、坏JSON、未知版本、重复声明必须拒收,不能退回旧路径。旧候选没有扩展时保持原分类,但不获得新的逐条件保证,不迁移历史文档。

scripts/lib/candidate-review-bundle.js仅在原候选issues通道校验显式记录的结构与引用是否具备,不判断自然语言分母或证据真假;原ReviewExecution与freshness继续负责执行与新鲜度。关闭主张分别核算设计与实际执行全部适用条件;任一partial/unreviewed/not-run/unknown不得被笼统“全部已验证”覆盖。

当复审、再次复审、ECR、发布前检查、实施中验证或外部 finding 处理发现“原清单没有覆盖但本轮必须处理”的问题时,必须在同一个复审清单文件追加 escape record,不能只在最终报告里口头说明。

escape record 至少包含:

字段说明
escapedItem逃逸问题或遗漏项名称
detectedAt发现时间、轮次、Run ID 或触发命令
previousChecklistGap原冻结清单缺了哪一项、哪一类维度或哪条证据
whyMissed为什么一开始没发现:范围遗漏、消费者漏扫、探针旧口径、只信报告、样例未扩维、历史镜像误判等
missingDimensionOrProbe缺失的审查维度、grep 反查、测试、validate、页面验证或人工证据
prevention下次如何避免同类问题,包括新增/调整清单项、探针、模板字段或 ownerSkill
checklistPatch新增到 frozen checklist 的项或追加清单编号
rerunEvidence补清单后重新执行的命令、代码落点、页面、构建或报告证据
ledgerRoute是否需要写 VL / PF / GAP / PI / ISSUE;不需要时写 record.none + skipReason

外部审查报告暴露的新问题先写入 EscapeAbsorptionQueue 或 escape record,再决定采纳/部分采纳/拒绝/延后;不得把外部报告正文直接吸收入需求/方案后跳过本地证据。

发现遗漏后处理顺序固定为:append escape record -> patch checklist -> 执行补充验证 -> 更新状态 -> 判断是否写台账 -> 再次收敛复审。若只是修复问题但没有记录逃逸原因和防复发策略,不得宣告“复审直至收敛”。

清单项分类

分类说明
requirement-alignment对照需求、用户文档、契约文档或技术方案
code-truth代码、配置、package、runtime、public API 或消费者入口真相
validation测试、validate、build、pack、install、Browser、API、benchmark
docs-consumerREADME、website、Profile、prompts、templates、部署副本
release-readinesschangelog、version、staged candidate snapshot、tag、registry、回滚、发布说明
governance-ledgerPI/PF/VL/GAP/ISSUE 台账状态和关闭证据
feature-inventoryProfile feature inventory、capability group、公开面、文档入口和验证路线
batch-evidenceEvidenceLedger、批次矩阵、Progress Card、baseline / executed / partial / failed / not-started 状态
brand-visual-evidenceVisualEvidencePack、母版谱系、主题几何、微尺寸/单色矩阵、人工结论;发现视觉 blocker 后必须追加 reset 并重跑受影响同母版矩阵

与其他 Skill 的关系

  • audit-common:提供审查维度和收敛规则;本 Skill 提供清单文件和逐项证据执行。
  • audit-release:发布前审查触发时叠加本 Skill 管理发布风险清单。
  • test-router:为清单项选择最小但足够的验证路线。
  • report:报告必须引用清单路径、状态新鲜度和未关闭项。
  • document-sync:复审结论影响 README/website/Profile/validate/部署副本时负责同步检查。

禁止

  • 禁止没有清单文件就宣告“复审直至收敛”。
  • 禁止用“审查报告说通过”代替实际验证。
  • 禁止每轮重复同一维度但算作有效零发现。
  • 禁止清单状态未更新、audit-state 仍旧或 SUMMARY 口径漂移时宣告完成。
  • 禁止删除失败项来制造全绿;只能追加修复记录和关闭证据。
  • 禁止发现遗漏后直接修复并跳过 escape record;遗漏原因、防复发策略和重跑证据必须留在清单文件中。

Bundled files

The model reads these on demand while the skill is loaded. They are exposed as readable files and are never executed.

Frequently asked questions

What does the Review Checklist AI skill do?

复审清单整理与审查规范 — 创建、冻结、证据执行、状态更新与收敛关闭

Why use Review Checklist on TypingMind?

Because you install it once and use it with any model. Review Checklist is plain Markdown rather than provider-specific code, so the same skill runs on GPT-5, Claude, Gemini, Grok, or a local model — and you can switch model mid-chat without it breaking. TypingMind runs on your own API keys, so you pay providers directly instead of a per-seat subscription, and your skills and chats stay in your own storage.

How do I install Review Checklist in TypingMind?

Open Plugins → Skills → Install from GitHub in TypingMind and paste https://github.com/devcodex-labs/devcodex/tree/main/content/skills/review-checklist. TypingMind reads its SKILL.md and bundles its files and installs it as a skill you can enable per chat.

Which AI models can use Review Checklist?

Any model you connect in TypingMind. AI skills are plain Markdown instructions rather than provider-specific code, so GPT, Claude, Gemini, Grok, and local models can all load this skill when a request matches it.

How many AI models can I use with Review Checklist?

As many as you like. As long as a model supports skills, you can use Review Checklist with it — GPT, Claude, Gemini, Grok, DeepSeek, Mistral, Llama and more — all on TypingMind with your own API keys.

Is the Review Checklist AI skill free?

Yes. It is published on GitHub by devcodex-labs under the AGPL-3.0 license. You only pay your own AI provider for the tokens you use.

What are AI skills?

An AI skill is a reusable instruction bundle that teaches an AI model how to do one specific task. It follows the open Agent Skills format: a SKILL.md file with a name and description, plus any scripts, templates or reference files the model may need. The model reads the instructions only when your request matches the skill, so an installed skill costs nothing until it is used.

How are AI skills different from plugins or MCP servers?

A plugin or MCP server gives a model new tools to call — code that runs somewhere and returns a result. An AI skill gives the model knowledge and process instead: how to approach a task, which steps to follow, what good output looks like. Skills are plain Markdown, so they need no server, no API key and no runtime, and they work with any model.

View all

Set up your own AI workspace now

Get notified about new features and future giveaways by subscribing to our newsletter 👇