Back to Details

skill-system-auditor

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese

Skill System Auditor

技能系统审计器

Audit a skill collection as a system rather than as isolated skill files.
Use this skill when:
  • the user asks for a global consistency audit of a skill repository
  • newly added skills may have made README, AGENTS, CLAUDE, lifecycle categories, or role categories stale
  • cross-skill routing, pair-with guidance, or memory writeback expectations need alignment
  • future-skill references, obsolete TODOs, or implemented-gap mentions need cleanup
  • helper paths, templates, frontmatter, or validation scripts need a maintenance pass
  • the user wants to decide what skills are missing next
Do not use this skill to design a single new skill from scratch. Use 
skill-creator
for the skill design mechanics, then use this skill to check whether the resulting collection remains coherent.
Pair this skill with:
  • skill-creator
    when audit findings lead to new or revised skill instructions
  • research-project-memory
    when the skill collection's roadmap and decisions should persist
  • update-docs
    when documentation drift is broad but not skill-specific
  • safe-git-ops
    before committing or recovering from Git state issues
将技能集合作为一个系统进行审计，而非单独审计各个技能文件。
在以下场景中使用此技能：
  • 用户要求对技能仓库进行全局一致性审计
  • 新增技能可能导致README、AGENTS、CLAUDE、生命周期类别或角色类别过时
  • 需要统一跨技能路由、配对指引或内存回写预期
  • 需要清理未来技能引用、过时的TODO标记或已实现的缺口提及内容
  • 需要对助手路径、模板、前置内容或验证脚本进行维护
  • 用户想要确定下一步需要补充哪些技能
请勿使用此技能从零开始设计单个新技能。使用
skill-creator
进行技能设计，之后再使用此技能检查生成的技能集合是否保持一致性。
可搭配以下技能使用：
  • 当审计结果需要创建新技能或修订技能说明时，搭配
    skill-creator
  • 当需要保留技能集合的路线图和决策时，搭配
    research-project-memory
  • 当文档偏差范围较广但不针对特定技能时，搭配
    update-docs
  • 在提交代码或从Git状态问题中恢复之前，搭配
    safe-git-ops

Skill Directory Layout

技能目录结构

text
<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── audit-rubric.md
    ├── doc-sync-map.md
    └── report-template.md
text
<installed-skill-dir>/
├── SKILL.md
└── references/
    ├── audit-rubric.md
    ├── doc-sync-map.md
    └── report-template.md

Progressive Loading

渐进式加载

  • Always read 
    references/audit-rubric.md
    and 
    references/doc-sync-map.md
    .
  • Read 
    references/report-template.md
    before writing an audit report.
  • If the repository has its own validation script or AGENTS instructions, read those before making edits.
  • 务必阅读
    references/audit-rubric.md
    和
    references/doc-sync-map.md
    。
  • 在撰写审计报告前，阅读
    references/report-template.md
    。
  • 如果仓库有自己的验证脚本或AGENTS说明，请在编辑前阅读这些内容。

Core Principles

核心原则

  • Audit lifecycle behavior, not only file presence.
  • A skill collection should have clear phase coverage, role coverage, routing, and handoff points.
  • Top-level docs must match the actual inventory.
  • "Future skill" references should not name skills that already exist.
  • Memory writeback expectations should cover every skill that changes durable project state.
  • Keep fixes surgical. Do not rewrite mature skills just to make wording uniform.
  • Validate locally before commit, push, or reinstall.
  • 审计生命周期行为，而非仅检查文件是否存在。
  • 技能集合应具备清晰的阶段覆盖、角色覆盖、路由和交接点。
  • 顶层文档必须与实际技能清单一致。
  • “未来技能”引用不应指向已存在的技能。
  • 内存回写预期应覆盖所有会改变持久化项目状态的技能。
  • 修复应精准到位。不要仅为了统一措辞而重写成熟技能。
  • 在提交、推送或重新安装前进行本地验证。

Step 1 - Recover Repository Rules

步骤1 - 还原仓库规则

Read:
  • AGENTS.md
    , 
    CLAUDE.md
    , README, or equivalent repo guidance
  • validation scripts
  • skill directory layout
  • recent audit reports or roadmap memory, if available
Record the expected install command, validation command, and documentation files that must stay synchronized.
阅读以下内容：
  • AGENTS.md
    、
    CLAUDE.md
    、README或等效的仓库指引
  • 验证脚本
  • 技能目录结构
  • 近期的审计报告或路线图记录（如有）
记录预期的安装命令、验证命令以及必须保持同步的文档文件。

Step 2 - Inventory Skills

步骤2 - 技能清单梳理

Build the actual inventory from 
skills/*/SKILL.md
.
For each skill, capture:
  • name
  • description trigger
  • lifecycle phase
  • role category
  • pair-with routing
  • helper references
  • memory writeback behavior
  • whether it is planned, implemented, deprecated, or duplicated
Compare this with top-level tables in README, AGENTS, CLAUDE, manifests, and audit reports.
从
skills/*/SKILL.md
构建实际的技能清单。
为每个技能记录以下信息：
  • 名称
  • 描述触发条件
  • 生命周期阶段
  • 角色类别
  • 配对路由
  • 助手引用
  • 内存回写行为
  • 状态（计划中、已实现、已弃用或重复）
将此清单与README、AGENTS、CLAUDE、清单文件和审计报告中的顶层表格进行对比。

Step 3 - Audit Lifecycle Coverage

步骤3 - 生命周期覆盖审计

Read 
references/audit-rubric.md
.
Check:
  • idea validation
  • literature and positioning
  • algorithm design
  • project setup
  • experiment design
  • baseline choice
  • experiment execution
  • result diagnosis
  • evidence capture
  • writing and paper evidence
  • reviewer simulation
  • citation coverage and correctness
  • submission
  • rebuttal
  • camera-ready
  • artifact evaluation
  • release and maintenance
  • advisor or collaborator communication
  • skill-system maintenance
Classify gaps as:
  • real-gap
    : useful capability is missing
  • covered-by-existing
    : existing skill covers it
  • not-in-scope
    : outside repository purpose
  • hardening
    : tests, examples, or docs needed rather than a new skill
阅读
references/audit-rubric.md
。
检查以下环节：
  • 想法验证
  • 文献调研与定位
  • 算法设计
  • 项目搭建
  • 实验设计
  • 基准选择
  • 实验执行
  • 结果诊断
  • 证据收集
  • 写作与论文证据整理
  • 审稿人模拟
  • 引用覆盖范围与正确性
  • 投稿
  • 回复审稿意见
  • 终稿提交
  • 成果物评估
  • 发布与维护
  • 与导师或合作者沟通
  • 技能系统维护
将缺口分类为：
  • real-gap
    ：缺少实用功能
  • covered-by-existing
    ：已有技能可覆盖
  • not-in-scope
    ：超出仓库目标范围
  • hardening
    ：需要测试、示例或文档，而非新技能

Step 4 - Audit Cross-Skill Routing

步骤4 - 跨技能路由审计

Check whether each skill routes to adjacent skills when its output naturally feeds another phase.
Look for missing or stale references around:
  • literature -> baseline -> experiment design
  • experiment result -> diagnosis -> paper evidence -> writing
  • reviewer simulation -> evidence board -> experiments/writing
  • rebuttal -> camera-ready -> artifact/release
  • advisor feedback -> decisions/actions/memory
  • audit findings -> skill creation/docs update
Do not add every possible cross-reference. Add only handoffs that change the user's next action.
检查每个技能的输出是否能自然流转到下一阶段的相邻技能。
查找以下环节中缺失或过时的引用：
  • 文献调研 -> 基准选择 -> 实验设计
  • 实验结果 -> 结果诊断 -> 论文证据整理 -> 写作
  • 审稿人模拟 -> 证据看板 -> 实验/写作
  • 回复审稿意见 -> 终稿提交 -> 成果物/发布
  • 导师反馈 -> 决策/行动/记录
  • 审计结果 -> 技能创建/文档更新
无需添加所有可能的交叉引用，仅添加会改变用户下一步操作的交接引用。

Step 5 - Audit Memory Writeback

步骤5 - 内存回写审计

Check that skills which create durable state update or route to memory:
  • decisions
  • claims
  • evidence
  • risks
  • actions
  • paper status
  • code/worktree state
  • reviewer and rebuttal state
  • artifact and release state
  • advisor feedback
  • skill-system roadmap decisions
If a memory protocol exists, update it instead of duplicating memory rules across all skills.
检查所有创建持久化状态的技能是否会更新或路由到内存：
  • 决策
  • 声明
  • 证据
  • 风险
  • 行动
  • 论文状态
  • 代码/工作区状态
  • 审稿人与回复意见状态
  • 成果物与发布状态
  • 导师反馈
  • 技能系统路线图决策
如果存在内存协议，请更新该协议，而非在所有技能中重复内存规则。

Step 6 - Audit Documentation and Stale References

步骤6 - 文档与过时引用审计

Read 
references/doc-sync-map.md
.
Search for:
  • implemented skills still listed as future
  • missing skills in top-level tables
  • old lifecycle counts
  • stale installation examples
  • obsolete helper paths
  • broken reference links
  • duplicate table entries
  • inconsistent skill names
Use repository validation scripts when available.
阅读
references/doc-sync-map.md
。
查找以下问题：
  • 已实现的技能仍被列为未来技能
  • 顶层表格中缺失的技能
  • 过时的生命周期统计
  • 过时的安装示例
  • 过时的助手路径
  • 失效的引用链接
  • 重复的表格条目
  • 不一致的技能名称
如有可用的仓库验证脚本，请使用它们。

Step 7 - Write the Audit Report

步骤7 - 撰写审计报告

Read 
references/report-template.md
.
If saving and no path is given, use:
text
docs/audits/global-consistency-audit_YYYY-MM-DD.md
The report must include:
  • scope
  • inventory count
  • lifecycle decision
  • findings fixed
  • findings left open
  • real remaining gaps
  • validation result
  • recommended next skill or hardening step
阅读
references/report-template.md
。
如果需要保存且未指定路径，请使用：
text
docs/audits/global-consistency-audit_YYYY-MM-DD.md
报告必须包含以下内容：
  • 审计范围
  • 技能清单数量
  • 生命周期决策
  • 已修复的问题
  • 未解决的问题
  • 实际存在的剩余缺口
  • 验证结果
  • 建议的下一步技能补充或强化步骤

Step 8 - Fix, Validate, and Handoff

步骤8 - 修复、验证与交接

When the user asks to implement fixes:
  1. Make the smallest edits that restore consistency.
  2. Run the repo validation command.
  3. Re-run targeted searches for stale skill names.
  4. Summarize changed files and remaining risk.
  5. Commit, push, and reinstall only if the user asks or the local workflow requires it.
当用户要求实施修复时：
  1. 做出最小限度的编辑以恢复一致性。
  2. 运行仓库验证命令。
  3. 重新针对过时技能名称进行搜索。
  4. 总结已修改的文件和剩余风险。
  5. 仅在用户要求或本地工作流需要时，进行提交、推送和重新安装。

Final Sanity Check

最终合理性检查

Before finishing:
  • actual skill inventory matches top-level docs
  • lifecycle and role categories include all implemented skills
  • cross-skill routing covers important feedback loops
  • memory writeback covers state-changing skills
  • future-skill references are accurate
  • validation passes
  • audit report separates fixed issues from remaining gaps
完成前需确认：
  • 实际技能清单与顶层文档一致
  • 生命周期和角色类别包含所有已实现的技能
  • 跨技能路由覆盖重要的反馈循环
  • 内存回写覆盖所有改变状态的技能
  • 未来技能引用准确无误
  • 验证通过
  • 审计报告区分已修复问题和剩余缺口