ln-162-skill-reviewer

Compare original and translation side by side

🇺🇸

Original

English
🇨🇳

Translation

Chinese
Paths: File paths (
shared/
,
references/
,
../ln-*
) are relative to skills repo root. If not found at CWD, locate this SKILL.md directory and go up one level for repo root.
路径说明: 文件路径(
shared/
references/
../ln-*
)均相对于技能仓库根目录。如果在当前工作目录(CWD)中未找到,请定位到该SKILL.md所在目录,再向上一级即可到达仓库根目录。

ln-162-skill-reviewer

ln-162-skill-reviewer

Type: L3 Worker (standalone-capable) Category: 1XX Documentation Pipeline Coordinator: ln-160-docs-skill-extractor (optional)
Universal skill reviewer with two auto-detected modes. Invocable standalone or by ln-160 coordinator.

类型: L3 Worker(可独立运行) 分类: 1XX 文档流水线 协调器: ln-160-docs-skill-extractor(可选)
这是一款具备两种自动检测模式的通用Skill审核工具,可独立调用,也可由ln-160协调器触发调用。

Mode Detection

模式检测

ConditionModeReview Profile
ln-*/SKILL.md
files exist in CWD
SKILLFull D1-D9 + M1-M5
.claude/commands/*.md
files exist
COMMANDStructural + actionability
Both existSKILL (default)Override:
$ARGUMENTS = commands

触发条件模式审核配置文件
当前工作目录中存在
ln-*/SKILL.md
文件
SKILL完整D1-D9 + M1-M5配置
存在
.claude/commands/*.md
文件
COMMAND结构合规性 + 可操作性检查
两种文件均存在SKILL(默认模式)可通过
$ARGUMENTS = commands
覆盖为COMMAND模式

Input

输入参数

$ARGUMENTS
options:
  • Empty -> auto-detect mode + auto-detect scope
  • ln-400 ln-500
    -> SKILL mode, specific skills
  • commands
    -> COMMAND mode, all
    .claude/commands/*.md
  • deploy.md run-tests.md
    -> COMMAND mode, specific files
When invoked by ln-160 coordinator: receives list of file paths to review in COMMAND mode.

$ARGUMENTS
可选配置:
  • 空值 -> 自动检测模式 + 自动检测范围
  • ln-400 ln-500
    -> SKILL模式,指定特定技能进行审核
  • commands
    -> COMMAND模式,审核所有
    .claude/commands/*.md
    文件
  • deploy.md run-tests.md
    -> COMMAND模式,审核指定文件
当由ln-160协调器触发调用时:将接收COMMAND模式下待审核的文件路径列表。

SKILL Mode

SKILL模式

Review SKILL.md files against 9 structural dimensions + 5 intent checks. Fix in-place. Report with PASS/FAIL verdict.
针对SKILL.md文件开展9项结构维度检查 + 5项意图检查,可就地修复问题,并生成包含PASS/FAIL结论的审核报告。

Phase 1: Scope Detection

阶段1:范围检测

If
$ARGUMENTS
provided:
treat each token as a skill directory prefix (e.g.,
ln-400
matches
ln-400-story-executor/
). Glob
{prefix}*/SKILL.md
per token.
If
$ARGUMENTS
empty:
auto-detect from git:
bash
git diff --name-only HEAD
git diff --name-only --cached
git ls-files --others --exclude-standard
Extract unique skill dirs (pattern
ln-\d+-[^/]+
) and shared paths (
shared/
). If shared files changed, Grep all
ln-*/SKILL.md
for references to changed filenames.
Build scope:
  1. Primary -- skills with directly changed files
  2. Affected -- skills referencing changed shared files
  3. Dependencies -- for each primary skill, extract
    ln-\d{3,4}
    references from its SKILL.md (callers, callees, worker tables)
Deduplicate. Report:
Scope: N primary, M affected, K dependency skills.
若提供
$ARGUMENTS
参数:
将每个参数视为技能目录前缀(例如
ln-400
匹配
ln-400-story-executor/
),为每个参数匹配
{prefix}*/SKILL.md
文件。
$ARGUMENTS
参数为空:
通过Git自动检测范围:
bash
git diff --name-only HEAD
git diff --name-only --cached
git ls-files --others --exclude-standard
提取唯一的技能目录(匹配
ln-\d+-[^/]+
模式)和共享路径(
shared/
)。若共享文件发生变更,需在所有
ln-*/SKILL.md
文件中搜索对变更文件名的引用。
构建审核范围:
  1. 核心范围 -- 包含直接变更文件的技能
  2. 关联范围 -- 引用了变更共享文件的技能
  3. 依赖范围 -- 针对每个核心技能,从其SKILL.md文件中提取
    ln-\d{3,4}
    格式的引用(调用方、被调用方、Worker表格)
去重后,输出:
审核范围:N个核心技能,M个关联技能,K个依赖技能。

Phase 2: Automated Verification

阶段2:自动化验证

MANDATORY READ: Load
references/automated_checks_skill_mode.md
Run all bash checks for every SKILL.md in scope. Record failures -- they feed D7/D8 as pre-verified violations. Every FAIL is confirmed -- no judgment needed, no skipping.
必读要求: 加载
references/automated_checks_skill_mode.md
文件
对范围内的每个SKILL.md文件运行所有bash检查,记录检查失败项 -- 这些失败项将作为D7/D8维度的预验证违规内容。所有FAIL结果均需确认 -- 无需主观判断,不得跳过。

Phase 3: Nine-Dimension Review

阶段3:九维度审核

MANDATORY READ: Load
references/dimensions_d1_d9.md
and
docs/SKILL_ARCHITECTURE_GUIDE.md
Read every SKILL.md in scope. Check ALL dimensions across ALL skills. Phase 2 failures are pre-verified -- include directly, do not re-check.
必读要求: 加载
references/dimensions_d1_d9.md
docs/SKILL_ARCHITECTURE_GUIDE.md
文件
读取范围内的每个SKILL.md文件,检查所有技能的所有维度。阶段2的失败项为预验证内容 -- 直接纳入报告,无需重复检查。

Phase 4: Intent Review

阶段4:意图审核

MANDATORY READ: Load
references/intent_m1_m5.md
Evaluate DESIGN INTENT of changes. Applies to primary skills only. For each primary skill, read the git diff (
git diff HEAD -- {skill_dir}/
).
必读要求: 加载
references/intent_m1_m5.md
文件
评估变更的设计意图,仅适用于核心技能。针对每个核心技能,读取其Git差异内容(
git diff HEAD -- {skill_dir}/
)。

Phase 5: Fix

阶段5:修复

For each finding:
  • Fixable (wrong path, stale ref, missing bidirectional ref, duplicated content) -- fix immediately via Edit
  • Ambiguous (conflicting thresholds where correct value unclear) -- list in report, do NOT guess
  • SIMPLIFY (from Phase 4) with unambiguous action -- fix immediately
  • REVERT (from Phase 4) -- roll back the change via Edit
  • RETHINK (from Phase 4) -- do NOT fix, pass to Phase 6 report
针对每个问题项:
  • 可修复(路径错误、过期引用、缺失双向引用、重复内容)-- 立即通过编辑操作修复
  • 不明确(阈值冲突,正确值不清晰)-- 列入报告,不得主观猜测
  • 简化优化(来自阶段4)且操作明确 -- 立即修复
  • 回滚(来自阶段4)-- 通过编辑操作回滚变更
  • 重新考量(来自阶段4)-- 不进行修复,移交至阶段6的报告中

Phase 6: Report

阶段6:生成报告

Verdict rules:
  • Any D1-D9 violation NOT auto-fixed -> FAIL
  • Only RETHINK findings (no unfixed violations) -> PASS with CONCERNS
  • Zero findings -> PASS
undefined
结论规则:
  • 存在任何未自动修复的D1-D9维度违规项 -> FAIL
  • 仅存在“重新考量”类问题(无未修复违规项)-> PASS with CONCERNS
  • 无任何问题项 -> PASS
undefined

Skill Coherence Review -- {PASS|PASS with CONCERNS|FAIL}

Skill一致性审核 -- {PASS|PASS with CONCERNS|FAIL}

Scope: {list of reviewed skills} Verdict: {verdict}
审核范围: {已审核技能列表} 结论: {审核结论}

Automated Checks (Phase 2)

自动化检查结果(阶段2)

CheckResultFailures
Frontmatter (D7){PASS/FAIL}{list or --}
Version/Date (D7){PASS/FAIL}{list or --}
Size <=800 (D8){PASS/FAIL}{list or --}
Description <=200 (D8){PASS/FAIL}{list or --}
MANDATORY READ paths (D2){PASS/FAIL}{list or --}
Orphan references (D7){PASS/FAIL}{list or --}
Passive file refs (D2){PASS/FAIL}{list or --}
Definition of Done (D7){PASS/FAIL}{list or --}
Meta-Analysis L1/L2 (D7){PASS/FAIL}{list or --}
Marketplace paths (D8, optional){PASS/FAIL/SKIP}{list or --}
Root docs stale names (D6){PASS/FAIL}{list or --}
Skill count accuracy (D8){PASS/FAIL}{list or --}
检查项结果失败项
前置元数据(D7){PASS/FAIL}{列表或--}
版本/日期(D7){PASS/FAIL}{列表或--}
文件大小≤800(D8){PASS/FAIL}{列表或--}
描述长度≤200(D8){PASS/FAIL}{列表或--}
必读文件路径(D2){PASS/FAIL}{列表或--}
孤立引用(D7){PASS/FAIL}{列表或--}
被动文件引用(D2){PASS/FAIL}{列表或--}
完成定义(D7){PASS/FAIL}{列表或--}
元分析L1/L2(D7){PASS/FAIL}{列表或--}
市场路径(D8,可选){PASS/FAIL/SKIP}{列表或--}
根目录文档过期名称(D6){PASS/FAIL}{列表或--}
技能数量准确性(D8){PASS/FAIL}{列表或--}

Fixed ({count})

已修复问题({数量})

#SkillDimIssueFix Applied
序号技能维度问题描述修复措施

Intent Findings ({count})

意图检查结果({数量})

M5 Classification: {R} requested, {D} derived, {S} speculative ({K} kept, {V} reverted)
#SkillDimFindingCategory
M5分类: {R}个需求类,{D}个推导类,{S}个推测类({K}个保留,{V}个回滚)
序号技能维度检查结果类别

Remaining Concerns ({count})

剩余待处理问题({数量})

#SkillDimIssueWhy Not Auto-Fixed
序号技能维度问题描述未自动修复原因

Clean

无问题维度

Dimensions with no findings: {list}

If zero findings: `All 9 structural dimensions + 5 intent checks clean. PASS.`
未发现问题的维度:{列表}

若无任何问题项:`所有9项结构维度 + 5项意图检查均无问题。结论:PASS。`

Phase 7: Volatile Numbers Cleanup

阶段7:动态数值清理

Skill/plugin/category counts go stale after every add/remove. One rule: counts ONLY in README.md badge (
skills-NNN
). Everywhere else -- no hardcoded counts.
Remove from any file (including marketplace.json descriptions, CLAUDE.md, AGENTS.md, CHANGELOG.md, SKILL.md):
  • Total skill counts, per-plugin counts, per-category counts
  • Worker/coordinator counts referencing OTHER skills (a skill's OWN internals are fine)
Phase 2 automated check verifies README badge matches actual skill count on disk -- fix if FAIL.

技能/插件/分类数量会在每次添加/删除后过期,遵循一条规则:仅在README.md徽章中保留数量统计(格式为
skills-NNN
)。除此之外,任何位置均不得出现硬编码的数量值。
从所有文件中移除以下内容(包括marketplace.json描述、CLAUDE.md、AGENTS.md、CHANGELOG.md、SKILL.md):
  • 总技能数、插件分类数、技能分类数
  • 引用其他技能的Worker/协调器数量(技能内部的数量统计不受限制)
阶段2的自动化检查会验证README徽章中的数量是否与磁盘上的实际技能数量一致 -- 若不一致则进行修复。

COMMAND Mode

COMMAND模式

Review
.claude/commands/*.md
files against structural + actionability criteria.
MANDATORY READ: Load
references/command_review_criteria.md
针对
.claude/commands/*.md
文件开展结构合规性 + 可操作性检查。
必读要求: 加载
references/command_review_criteria.md
文件

Phase 1: Scope Detection

阶段1:范围检测

If file paths provided: review those files. If
commands
keyword:
Glob
.claude/commands/*.md
. If invoked by ln-160: use file list from coordinator.
若提供文件路径: 审核指定文件 若提供
commands
关键字:
匹配所有
.claude/commands/*.md
文件 若由ln-160触发调用: 使用协调器提供的文件列表

Phase 2: Review

阶段2:审核

For each command file, apply all criteria from
references/command_review_criteria.md
.
针对每个命令文件,应用
references/command_review_criteria.md
中的所有审核标准。

Phase 3: Fix

阶段3:修复

Auto-fix where possible (add missing frontmatter, truncate description). Flag unfixable issues.
自动修复可处理的问题(添加缺失的前置元数据、截断过长描述),标记无法自动修复的问题。

Phase 4: Report

阶段4:生成报告

undefined
undefined

Command Review -- {N} files

Command审核报告 -- {N}个文件

FileVerdictIssues
Verdicts: PASS / FIXED / WARN Pass rate: {X}%

---
文件结论问题项
结论:PASS / FIXED / WARN 通过率:{X}%

---

Rules

规则

  • Automated checks (Phase 2) are NON-NEGOTIABLE -- every FAIL must appear in report
  • Do NOT skip any dimension for any skill in scope
  • If unsure whether something violates a rule -- it violates the rule (strict interpretation)
  • Read ALL skills in scope before reporting
  • Fix errors immediately, do not defer
  • Do NOT update versions or dates unless user explicitly requests it
  • shared/
    changes affect every skill that references them -- check reverse dependencies
  • Intent review (M1-M5) evaluates DESIGN, not correctness -- findings are judgment-based
  • RETHINK findings are advisory -- explain WHY, author decides WHETHER to act
  • REVERT findings are executed immediately -- changes without concrete defect are rolled back
  • SPECULATIVE items (M5) without user response default to REVERT -- no silent acceptance of model-generated additions

  • 自动化检查(阶段2)为强制要求 -- 所有FAIL结果必须纳入报告
  • 不得跳过范围内任何技能的任何维度检查
  • 若不确定某项内容是否违反规则 -- 判定为违规(严格解释规则)
  • 在生成报告前,需读取范围内的所有技能文件
  • 立即修复错误,不得延迟
  • 除非用户明确要求,否则不得更新版本或日期
  • shared/
    目录的变更会影响所有引用该目录的技能 -- 需检查反向依赖
  • 意图审核(M1-M5)评估设计合理性,而非内容正确性 -- 检查结果为主观判断结论
  • “重新考量”类问题为建议性内容 -- 需说明原因,由作者决定是否采取行动
  • “回滚”类问题需立即执行 -- 无明确缺陷的变更需回滚
  • 无用户反馈的推测性内容(M5)默认回滚 -- 不得静默接受模型生成的新增内容

版本: 1.0.0 最后更新日期: 2026-03-13

Definition of Done

  • Scope detected (primary + affected + dependency skills)
  • Phase 2 automated checks executed for all skills in scope
  • D1-D9 dimensions reviewed across all skills
  • M1-M5 intent evaluated for primary skills
  • Fixable findings auto-fixed via Edit
  • Report generated with PASS/PASS with CONCERNS/FAIL verdict

Version: 1.0.0 Last Updated: 2026-03-13