Back to Details

review-work

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Review Work

评审工作

Independently verifies implementation results using only the plan document and the codebase. Receives no information from the execution process.
仅使用计划文档和代码库独立验证实现结果，不接收任何执行过程中的信息。

Core Principle

核心原则

The reviewer shares no memory with the executor. The plan's stated goals and the current state of the codebase — these two alone are the basis for judgment.
评审者与执行者无共享记忆。判断仅基于计划中明确的目标和代码库的当前状态这两项依据。

Hard Gates

硬性要求

  1. Do not receive execution context. No logs from run-plan, no worker output, no diffs, no task completion summaries, no conversation history. The only input is the plan file path.
  2. Read the plan document directly. Read the plan file from disk — not a summary or a passed-along description.
  3. Run all tests yourself. Do not trust previous execution results. Run the full test suite and every verification command specified in the plan.
  4. Verdict is PASS or FAIL. No conditional passes, no "almost done", no "only minor issues remain". Binary only.
  5. Save the review document to a file. Review results must be saved as a structured document. Never end with a verbal report alone.
  6. Do not modify code. This skill is read-only. If issues are found, report them — do not fix them.
  1. 不接收执行上下文：不获取run-plan的日志、工作输出、差异内容、任务完成摘要、对话历史。唯一输入为计划文件路径。
  2. 直接读取计划文档：从磁盘读取计划文件——而非摘要或转述内容。
  3. 自行运行所有测试：不依赖之前的执行结果。运行计划中指定的完整测试套件和每一条验证命令。
  4. 结论仅为PASS或FAIL：无条件通过、“接近完成”、“仅存小问题”这类结论均不允许，仅支持二元判定。
  5. 将评审文档保存至文件：评审结果必须保存为结构化文档，不能仅以口头报告结束。
  6. 不得修改代码：本skill为只读模式。若发现问题，仅需报告，无需修复。

When To Use

适用场景

  • After run-plan execution is complete
  • When the user says "review the work", "verify the implementation", "check if the plan was executed correctly"
  • When implementation is done but independent verification is needed
  • run-plan执行完成后
  • 用户说出“评审工作”、“验证实现”、“检查计划是否正确执行”时
  • 实现完成但需要独立验证时

When NOT To Use

不适用场景

  • While run-plan is still in progress
  • When no plan document exists (use 
    plan-crafting
    first)
  • When the goal is a general code review (this skill verifies "implementation against plan")
  • run-plan仍在执行过程中
  • 无计划文档存在时（需先使用
    plan-crafting
    ）
  • 目标为通用代码评审时（本skill仅验证“实现是否符合计划”）

Input

输入

The only input to this skill is the plan file path.
docs/engineering-discipline/plans/YYYY-MM-DD-<feature-name>.md
The following must never be provided as input:
  • Execution logs or task completion summaries from run-plan
  • Output or diffs from worker subagents
  • Validation results from validator subagents
  • Conversation history from the execution session
本skill的唯一输入为计划文件路径。
docs/engineering-discipline/plans/YYYY-MM-DD-<feature-name>.md
以下内容绝对不能作为输入提供：
  • run-plan的执行日志或任务完成摘要
  • 子worker agent的输出或差异内容
  • 子validator agent的验证结果
  • 执行会话的对话历史

Process

流程

Phase 1: Load and Analyze Plan Document

阶段1：加载并分析计划文档

  1. Receive the plan file path as input
  2. Read the plan document directly from disk
  3. Extract the following:
    • Goal: What this plan implements
    • Work Scope: In scope / Out of scope
    • Task List: Each task's name, acceptance criteria, and target files
    • File Structure Mapping: Complete list of files to be created or modified
    • Commit Structure: Commit messages and scope specified in the plan
    • Test Commands: All test execution commands specified in the plan
Use the extracted results as the foundation for the review document.
  1. 接收计划文件路径作为输入
  2. 直接从磁盘读取计划文档
  3. 提取以下信息：
    • 目标：本计划要实现的内容
    • 工作范围：包含/不包含的内容
    • 任务列表：每个任务的名称、验收标准和目标文件
    • 文件结构映射：需要创建或修改的完整文件列表
    • 提交结构：计划中指定的提交信息和范围
    • 测试命令：计划中指定的所有测试执行命令
将提取结果作为评审文档的基础。

Phase 2: Codebase Inspection

阶段2：代码库检查

Inspect the codebase against the files specified in the plan.
  1. File existence check: Verify that all files specified in the plan actually exist
  2. Content alignment check: Inspect whether each file's content matches the plan's requirements (function signatures, type definitions, logic, etc.)
  3. Residual artifact check:
    • Placeholder code (TODO, FIXME, "implement later", stub functions)
    • Debug code (console.log, print debugging, commented-out code blocks)
    • Unexpected changes outside the plan's scope
  4. Verify acceptance criteria per task. Check each criterion stated in the plan one by one and record whether it is met.
对照计划中指定的文件检查代码库。
  1. 文件存在性检查：验证计划中指定的所有文件是否实际存在
  2. 内容一致性检查：检查每个文件的内容是否符合计划要求（函数签名、类型定义、逻辑等）
  3. 残留工件检查：
    • 占位符代码（TODO、FIXME、“后续实现”、桩函数）
    • 调试代码（console.log、打印调试、注释掉的代码块）
    • 计划范围外的意外变更
  4. 验证每个任务的验收标准：逐一检查计划中列出的每个标准，并记录是否满足。

Phase 3: Test Execution

阶段3：测试执行

  1. Run all individual test commands specified in the plan
  2. Run the full test suite to check for regressions
  3. Record each test's result (PASS/FAIL)
  4. If any test fails, record the error message
  1. 运行计划中指定的所有单独测试命令
  2. 运行完整测试套件以检查是否存在回归问题
  3. 记录每个测试的结果（PASS/FAIL）
  4. 若任何测试失败，记录错误信息

Phase 4: Git History Verification

阶段4：Git历史验证

  1. Compare the commit structure specified in the plan against the actual 
    git log
  2. Verify that commit messages match the plan
  3. Verify that each commit's change scope is appropriate (no unrelated changes mixed into a single commit)
  1. 将计划中指定的提交结构与实际
    git log
    进行对比
  2. 验证提交信息是否与计划匹配
  3. 验证每个提交的变更范围是否恰当（单个提交中未混入无关变更）

Phase 5: Verdict and Review Document

阶段5：结论与评审文档

Combine results from Phases 2–4 to reach a verdict.
PASS conditions (all must be met):
  • All files specified in the plan exist
  • Each task's acceptance criteria are met
  • All tests pass
  • No regressions
  • No placeholder or debug code remains
FAIL (if any of the following apply):
  • A file specified in the plan is missing
  • A test fails
  • A regression is found
  • Placeholder code remains
  • The plan's goal is not achieved
After reaching a verdict, write and save the review document.
结合阶段2-4的结果得出结论。
PASS条件（必须全部满足）：
  • 计划中指定的所有文件均存在
  • 每个任务的验收标准均已满足
  • 所有测试通过
  • 无回归问题
  • 无占位符或调试代码残留
FAIL条件（满足任意一项即可）：
  • 计划中指定的某个文件缺失
  • 某项测试失败
  • 发现回归问题
  • 存在占位符代码残留
  • 计划目标未达成
得出结论后，编写并保存评审文档。

Review Document

评审文档

Save Location

保存位置

docs/engineering-discipline/reviews/YYYY-MM-DD-<feature-name>-review.md
(User preferences for review location override this default.)
docs/engineering-discipline/reviews/YYYY-MM-DD-<feature-name>-review.md
（用户指定的评审位置优先级高于此默认路径。）

Document Structure

文档结构

markdown
undefined
markdown
undefined

[Feature Name] Review

[功能名称] 评审

Date: YYYY-MM-DD HH:MM Plan Document: 
docs/engineering-discipline/plans/YYYY-MM-DD-<feature-name>.md
Verdict: PASS / FAIL
日期： YYYY-MM-DD HH:MM 计划文档： 
docs/engineering-discipline/plans/YYYY-MM-DD-<feature-name>.md
结论： PASS / FAIL

1. File Inspection Against Plan

1. 计划文件匹配检查

Planned FileStatusNotes
path/to/file
OK / Missing / MismatchDetails
计划文件状态备注
path/to/file
正常 / 缺失 / 不匹配详情

2. Test Results

2. 测试结果

Test CommandResultNotes
pytest tests/...
PASS / FAILError details if failed
Full Test Suite: PASS / FAIL (N passed, M failed)
测试命令结果备注
pytest tests/...
PASS / FAIL失败时的错误详情
完整测试套件： PASS / FAIL（通过N项，失败M项）

3. Code Quality

3. 代码质量

  • No placeholders
  • No debug code
  • No commented-out code blocks
  • No changes outside plan scope
Findings:
  • (Describe with file path and line number)
  • 无占位符代码
  • 无调试代码
  • 无注释掉的代码块
  • 无计划范围外的变更
发现问题：
  • （描述问题并标注文件路径和行号）

4. Git History

4. Git历史

Planned CommitActual CommitMatch
feat: add X
abc1234 feat: add X
OK / Mismatch
计划提交信息实际提交信息是否匹配
feat: add X
abc1234 feat: add X
正常 / 不匹配

5. Overall Assessment

5. 整体评估

(Summary of the overall judgment. If FAIL, describe specifically which items failed and why.)
（整体判断总结。若为FAIL，具体描述哪些项未通过及原因。）

6. Follow-up Actions

6. 后续行动

  • (If FAIL: list of items that need to be fixed)
  • (If PASS: record improvement suggestions if any)
undefined
  • （若为FAIL：列出需要修复的项）
  • （若为PASS：记录改进建议，如有）
undefined

When To Stop

终止时机

Stop immediately and notify the user in the following situations:
  • The plan file does not exist or cannot be read
  • The test execution environment is not ready (e.g., dependencies not installed)
  • The plan document format cannot be parsed
When in doubt, do not guess — ask the user.
出现以下情况时立即停止并通知用户：
  • 计划文件不存在或无法读取
  • 测试执行环境未就绪（如依赖未安装）
  • 计划文档格式无法解析
如有疑问，请勿猜测——询问用户。

Anti-Patterns

反模式

Anti-PatternWhy It Fails
Reading run-plan execution logs to verifyInformation isolation violation. Anchors on the executor's framing
Trusting previous test results instead of running testsEnvironment may have changed after execution. Not independent verification
Finding issues and fixing them directlyViolates separation of reviewer and implementer roles
Giving a "close enough, PASS" verdictNo conditional passes. If criteria are not met, it is FAIL
Delivering review results verbally without saving a documentNo verification record remains. Untraceable
Judging by criteria not in the planThe reviewer judges only by the plan's criteria. Adding arbitrary standards is prohibited
Receiving a plan summary and verifying from thatInformation is lost during summarization. The original must be read directly
反模式失败原因
读取run-plan执行日志进行验证违反信息隔离原则，受执行者的框架限制
依赖之前的测试结果而非自行运行测试执行后环境可能已变更，不属于独立验证
发现问题后直接修复违反评审者与执行者的角色分离原则
给出“差不多，可以通过”的结论不允许条件通过，若未满足标准则判定为FAIL
仅口头传达评审结果而不保存文档无验证记录，无法追溯
依据计划外的标准进行判断评审者仅能依据计划中的标准判断，禁止添加任意标准
接收计划摘要并基于摘要进行验证摘要过程中会丢失信息，必须直接读取原始文档

Minimal Checklist

最小检查清单

Self-check when review is complete:
  • Read the plan document directly from disk
  • Did not receive run-plan execution results as input
  • Ran all tests myself
  • Inspected all tasks in the plan
  • Verdict is either PASS or FAIL
  • Saved the review document to a file
评审完成后进行自我检查：
  • 直接从磁盘读取了计划文档
  • 未接收run-plan执行结果作为输入
  • 自行运行了所有测试
  • 检查了计划中的所有任务
  • 结论为PASS或FAIL之一
  • 将评审文档保存至文件

Transition

后续过渡

After review is complete:
  • PASS → Report results to the user and suggest next steps (PR creation, deployment, etc.)
  • FAIL → Report failure items to the user. If fixes are needed, suggest transitioning to the 
    run-plan
    or 
    systematic-debugging
    skill
  • If the plan itself has issues → suggest returning to the 
    plan-crafting
    skill to revise the plan
This skill itself does not invoke the next skill. It saves the review document, reports results, and lets the user decide the next step.
评审完成后：
  • PASS → 向用户报告结果并建议下一步操作（创建PR、部署等）
  • FAIL → 向用户报告失败项。若需要修复，建议切换至
    run-plan
    或
    systematic-debugging
    skill
  • 若计划本身存在问题 → 建议返回
    plan-crafting
    skill修订计划
本skill本身不会调用下一个skill。它会保存评审文档、报告结果，由用户决定下一步操作。