返回技能列表

review-skill-format

pjt222
更新于 2 days ago
7 次查看
17
2
17
在 GitHub 上查看
aidesign

关于

This skill validates SKILL.md files for compliance with the agentskills.io standard. It checks YAML frontmatter, required sections, line count limits, procedure step formatting, and registry synchronization. Use it to review new skills, re-validate modified ones, audit domains, or check contributor PRs.

快速安装

Claude Code

推荐
主要方式
npx skills add pjt222/agent-almanac -a claude-code
插件命令备选方式
/plugin add https://github.com/pjt222/agent-almanac
Git 克隆备选方式
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/review-skill-format

在 Claude Code 中复制并粘贴此命令以安装该技能

技能文档

Review Skill Format

Validate SKILL.md vs agentskills.io open std. Checks YAML frontmatter completeness, required section presence, proc step format (Expected/On failure blocks), line count limits, registry sync. Use before merging any new or modified skill.

Use When

  • New skill authored + needs format validation before merge
  • Existing skill modified + needs re-validation
  • Batch audit of all skills in domain
  • Verify skill created by create-skill meta-skill
  • Review contributor skill submission in PR

In

  • Required: Path to SKILL.md (e.g. skills/setup-vault/SKILL.md)
  • Optional: Strictness level (lenient or strict, default strict)
  • Optional: Check registry sync (default yes)

Do

Step 1: Verify File + Read

Confirm SKILL.md exists at expected path + read full content.

# Verify file exists
test -f skills/<skill-name>/SKILL.md && echo "EXISTS" || echo "MISSING"

# Count lines
wc -l < skills/<skill-name>/SKILL.md

→ File exists + content readable. Line count displayed.

If err: file doesn't exist → check path typos. Verify dir exists ls skills/<skill-name>/. Dir missing → skill not created → use create-skill first.

Step 2: Check Frontmatter Fields

Parse YAML frontmatter (between --- delimiters) + verify all required + recommended fields present.

Required:

  • name — matches dir name (kebab-case)
  • description — < 1024 chars, starts w/ verb
  • license — typically MIT
  • allowed-tools — comma or space-separated tool list

Recommended metadata:

  • metadata.author — author name
  • metadata.version — semantic ver string
  • metadata.domain — one of domains in skills/_registry.yml
  • metadata.complexitybasic, intermediate, advanced
  • metadata.language — primary lang or multi
  • metadata.tags — comma-separated, 3-6 tags, includes domain
# Check required frontmatter fields exist
head -30 skills/<skill-name>/SKILL.md | grep -q '^name:' && echo "name: OK" || echo "name: MISSING"
head -30 skills/<skill-name>/SKILL.md | grep -q '^description:' && echo "description: OK" || echo "description: MISSING"
head -30 skills/<skill-name>/SKILL.md | grep -q '^license:' && echo "license: OK" || echo "license: MISSING"
head -30 skills/<skill-name>/SKILL.md | grep -q '^allowed-tools:' && echo "allowed-tools: OK" || echo "allowed-tools: MISSING"

→ All 4 required fields present. All 6 metadata present. name matches dir. description < 1024 chars.

If err: report each missing as BLOCKING. name doesn't match dir → BLOCKING w/ expected value. description > 1024 chars → SUGGEST w/ current length.

Step 3: Locale-Specific Validation (Translations Only)

Frontmatter has locale field → file is translated SKILL.md. Perform additional checks. No locale → skip.

  1. Translation frontmatter fields — Verify these 5 present:

    • locale — target locale code (e.g. de, ja, zh-CN, es)
    • source_locale — origin (typically en)
    • source_commit — commit hash of English source used
    • translator — who/what produced
    • translation_date — ISO 8601 date
  2. Prose lang scan — Sample 3-5 body paragraphs (outside code blocks, frontmatter, headings). Verify prose written in target locale not English. Ignore: code blocks, inline code, tool names, field names, file paths, English terms w/ no std translation.

  3. Code block identity check — Compare code blocks in translated vs English source at skills/<skill-name>/SKILL.md. Code blocks must be identical (code never translated). Flag any code block content differing from English.

# Check translation frontmatter fields
for field in "locale:" "source_locale:" "source_commit:" "translator:" "translation_date:"; do
  grep -q "^${field}\|^  ${field}" i18n/<locale>/skills/<skill-name>/SKILL.md \
    && echo "$field OK" || echo "$field MISSING"
done

→ All 5 translation fields present. Body paragraphs in target locale. Code blocks match English source exactly.

If err: report missing translation fields as BLOCKING. Body paragraphs in English despite non-English locale → BLOCKING — file has untranslated prose. Code blocks differ from English source → BLOCKING — code must not be translated/modified.

Step 4: Check Required Sections

Verify all 6 required sections in skill body (after frontmatter).

Required:

  1. ## When to Use
  2. ## Inputs
  3. ## Procedure (w/ ### Step N: sub-sections)
  4. ## Validation (may also ## Validation Checklist)
  5. ## Common Pitfalls
  6. ## Related Skills
# Check each required section
for section in "## When to Use" "## Inputs" "## Procedure" "## Common Pitfalls" "## Related Skills"; do
  grep -q "$section" skills/<skill-name>/SKILL.md && echo "$section: OK" || echo "$section: MISSING"
done

# Validation section may use either heading
grep -qE "## Validation( Checklist)?" skills/<skill-name>/SKILL.md && echo "Validation: OK" || echo "Validation: MISSING"

→ All 6 sections present. Procedure section has ≥1 ### Step sub-heading.

If err: report each missing as BLOCKING. Skill w/o all 6 = non-compliant w/ agentskills.io. Provide section template from create-skill.

Step 5: Check Procedure Step Format

Verify each proc step follows pattern: numbered title, ctx, code block(s), Expected/On failure blocks.

For each ### Step N: sub-section, check:

  1. Step has descriptive title (not just "Step N")
  2. ≥1 code block or concrete instr exists
  3. **Expected:** block present
  4. **On failure:** block present

→ Every proc step has both Expected + On failure. Steps have concrete code or instr, not vague descriptions.

If err: report each step missing Expected/On failure as BLOCKING. Steps w/ only vague instrs ("configure system appropriately") → SUGGEST w/ note to add concrete cmds.

Step 6: Verify Line Count

Check SKILL.md ≤ 500-line limit.

lines=$(wc -l < skills/<skill-name>/SKILL.md)
[ "$lines" -le 500 ] && echo "OK ($lines lines)" || echo "OVER LIMIT ($lines lines > 500)"

→ Line count ≤ 500.

If err: > 500 → BLOCKING. Recommend refactor-skill-structure to extract code blocks > 15 lines → references/EXAMPLES.md. Typical reduction: 20-40% by extracting extended examples.

Step 7: Check Registry Sync

Verify skill listed in skills/_registry.yml under correct domain w/ matching metadata.

Check:

  1. Skill id exists under correct domain section
  2. path matches <skill-name>/SKILL.md
  3. complexity matches frontmatter
  4. description present (may be abbreviated)
  5. total_skills count at top matches actual skill count
# Check if skill is in registry
grep -q "id: <skill-name>" skills/_registry.yml && echo "Registry: FOUND" || echo "Registry: NOT FOUND"

# Check path
grep -A1 "id: <skill-name>" skills/_registry.yml | grep -q "path: <skill-name>/SKILL.md" && echo "Path: OK" || echo "Path: MISMATCH"

→ Skill listed under correct domain w/ matching path + metadata. Total count accurate.

If err: not found in registry → BLOCKING. Provide entry template:

- id: skill-name
  path: skill-name/SKILL.md
  complexity: intermediate
  language: multi
  description: One-line description

Check

  • SKILL.md file exists at expected path
  • YAML frontmatter parses w/o errors
  • All 4 required frontmatter present (name, description, license, allowed-tools)
  • All 6 metadata present (author, version, domain, complexity, language, tags)
  • name field matches dir name
  • description < 1024 chars
  • All 6 required sections (When to Use, Inputs, Procedure, Validation, Common Pitfalls, Related Skills)
  • Every proc step has Expected: + On failure:
  • Line count ≤ 500
  • Skill listed in _registry.yml w/ correct domain, path, metadata
  • total_skills count in registry accurate
  • (Translations only) All 5 translation fields present (locale, source_locale, source_commit, translator, translation_date)
  • (Translations only) Body paragraphs in target locale not English
  • (Translations only) Code blocks identical to English source

Traps

  • Check frontmatter w/ regex only: YAML parsing subtle. description: > multiline diff from description: "inline". Check both patterns when searching.
  • Miss Validation section variant: Some skills use ## Validation Checklist not ## Validation. Both acceptable; check for either.
  • Forget registry total count: After adding skill to registry, total_skills must increment. Common miss in PRs.
  • Name vs title confusion: name field = kebab-case matching dir. # Title heading = human-readable + can differ (e.g. name: review-skill-format, title: # Review Skill Format).
  • Lenient mode skip blockers: Even lenient, missing required sections + frontmatter still flagged. Lenient only relaxes style + metadata.
  • Translated skills w/ English prose: File w/ non-English frontmatter, headings, English body passes all structural checks. Always verify body lang for translated — locale field signals prose must be in target lang not English.

  • create-skill — canonical format spec; authoritative ref for valid SKILL.md
  • update-skill-content — after format validation passes, improve content quality
  • refactor-skill-structure — skill fails line count → extract + reorganize
  • review-pull-request — reviewing PR adding/modifying skills → combine w/ format validation

GitHub 仓库

pjt222/agent-almanac
路径: i18n/caveman-ultra/skills/review-skill-format
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

相关推荐技能

content-collections

Content Collections 是一个 TypeScript 优先的构建工具,可将本地 Markdown/MDX 文件转换为类型安全的数据集合。它专为构建博客、文档站和内容密集型 Vite+React 应用而设计,提供基于 Zod 的自动模式验证。该工具涵盖从 Vite 插件配置、MDX 编译到生产环境部署的完整工作流。

查看技能

polymarket

这个Claude Skill为开发者提供完整的Polymarket预测市场开发支持,涵盖API调用、交易执行和市场数据分析。关键特性包括实时WebSocket数据流,可监控实时交易、订单和市场动态。开发者可用它构建预测市场应用、实施交易策略并集成实时市场预测功能。

查看技能

creating-opencode-plugins

该Skill帮助开发者创建OpenCode插件,用于接入命令、文件、LSP等25+种事件。它提供了插件结构、事件API规范和JavaScript/TypeScript实现模式,适合需要拦截操作、扩展功能或自定义事件处理的场景。开发者可通过它快速构建响应式模块来增强OpenCode AI助手的能力。

查看技能

sglang

SGLang是一个专为LLM设计的高性能推理框架,特别适用于需要结构化输出的场景。它通过RadixAttention前缀缓存技术,在处理JSON、正则表达式、工具调用等具有重复前缀的复杂工作流时,能实现极速生成。如果你正在构建智能体或多轮对话系统,并追求远超vLLM的推理性能,SGLang是理想选择。

查看技能