スキル一覧に戻る

update-skill-content

pjt222
更新日 2 days ago
6 閲覧
17
2
17
GitHubで表示
メタapi

について

このスキルは、既存のSKILL.mdドキュメントを更新し、手順の洗練、陥りやすい点の拡充、関連スキルの同期を通じて、正確性と明確性を向上させます。このスキルは、ドキュメントが古いツールを参照していたり、トラブルシューティングのセクションが不十分であったり、壊れた相互参照を含んでいる場合に使用されます。ユーザーフィードバックに基づいて、バージョン番号の更新とコンテンツの改良を行います。

クイックインストール

Claude Code

推奨
メイン
npx skills add pjt222/agent-almanac -a claude-code
プラグインコマンド代替
/plugin add https://github.com/pjt222/agent-almanac
Git クローン代替
git clone https://github.com/pjt222/agent-almanac.git ~/.claude/skills/update-skill-content

このコマンドをClaude Codeにコピー&ペーストしてスキルをインストールします

ドキュメント

Update Skill Content

Improve existing SKILL.md by refining procedure steps, expanding Common Pitfalls with real failure modes, syncing Related Skills section, bumping version. Use this after skill passes format validation but has content gaps, stale references, or incomplete procedures.

When Use

  • Skill procedure steps reference outdated tools, APIs, version numbers
  • Common Pitfalls section thin (fewer than 3 pitfalls) or missing real failure modes
  • Related Skills section has broken cross-references or missing relevant links
  • Procedure steps lack concrete code examples or have vague instructions
  • New skill added to library that should be cross-referenced from existing skills
  • After feedback that skill procedures unclear or incomplete

Inputs

  • Required: Path to SKILL.md file to update
  • Optional: Specific section(s) to focus on (e.g., "procedure", "pitfalls", "related-skills")
  • Optional: Source of updates (changelog, issue report, user feedback)
  • Optional: Whether to bump version (default: yes, minor bump)

Steps

Step 1: Read Current Skill, Assess Content Quality

Read entire SKILL.md. Evaluate each section for completeness, accuracy.

Assessment criteria per section:

  • When to Use: Triggers concrete, actionable? (3-5 items expected)
  • Inputs: Types, defaults, required/optional clearly separated?
  • Procedure: Each step has concrete code, Expected, On failure?
  • Validation: Checklist items objectively testable? (5+ items expected)
  • Common Pitfalls: Pitfalls specific with symptoms, fixes? (3-6 expected)
  • Related Skills: Referenced skills exist? Obvious related skills missing?

Got: Clear picture of which sections need improvement. Specific gaps identified.

If fail: Skill cannot be read (path error)? Verify path. SKILL.md has broken YAML frontmatter? Fix frontmatter first using review-skill-format before attempting content updates.

Step 2: Check for Stale References

Scan procedure steps for version-specific references, tool names, URLs, API patterns that may have changed.

Common staleness indicators:

  • Specific version numbers (e.g., v1.24, R 4.3.0, Node 18)
  • URLs that may have moved or expired
  • CLI flags or command syntax changed
  • Package names renamed or deprecated
  • Configuration file formats evolved
# Check for version-specific references
grep -nE '[vV][0-9]+\.[0-9]+' skills/<skill-name>/SKILL.md

# Check for URLs
grep -nE 'https?://' skills/<skill-name>/SKILL.md

Got: List of potentially stale references with line numbers. Each reference verified as current or flagged for update.

If fail: Too many references to check manual? Prioritize: procedure code blocks first (most likely to cause runtime failures), then Common Pitfalls (may reference old workarounds), then informational text.

Step 3: Update Procedure Steps for Accuracy

For each procedure step identified as needing improvement:

  1. Verify code blocks still execute correct or reflect current best practices
  2. Add missing context sentences explaining why step needed
  3. Ensure concrete commands use real paths, real flags, real output
  4. Update Expected blocks to match current tool behavior
  5. Update On failure blocks with current error messages, fixes

When updating code blocks, preserve original structure:

  • Keep step numbering consistent
  • Maintain ### Step N: Title format
  • No reorder steps unless original order was incorrect

Got: All procedure steps contain current, executable code. Expected/On failure blocks reflect actual current behavior.

If fail: Unsure whether code block still correct? Add note: <!-- TODO: Verify this command against current version -->. No remove working code blocks to replace with untested alternatives.

Step 4: Expand Common Pitfalls

Review Common Pitfalls section. Expand if gaps exist.

Quality criteria for pitfalls:

  • Each pitfall has bold name followed by specific description
  • Description includes symptom (what goes wrong) and fix (how to avoid or recover)
  • Pitfalls drawn from real failure modes, not hypothetical concerns
  • 3-6 pitfalls = target range

Sources for new pitfalls:

  • Procedure steps with complex On failure blocks (likely pitfalls)
  • Related skills that warn about same tools or patterns
  • Common issues reported by users of procedure

Got: 3-6 pitfalls, each with specific symptom and fix. No generic pitfalls like "be careful" or "test thoroughly".

If fail: Only 1-2 pitfalls can be identified? Acceptable for basic-complexity skills. For intermediate, advanced skills, fewer than 3 pitfalls suggests author has not fully explored failure modes — flag for future expansion.

Step 5: Sync Related Skills Section

Verify all cross-references in Related Skills section are valid. Add any missing links.

  1. For each referenced skill, verify exists:
    # Check if referenced skill exists
    test -d skills/referenced-skill-name && echo "EXISTS" || echo "NOT FOUND"
    
  2. Search for skills that reference this skill (should be cross-linked):
    # Find skills that reference this skill
    grep -rl "skill-name" skills/*/SKILL.md
    
  3. Check obvious related skills based on domain, tags
  4. Use format: - \skill-id` — one-line description of relationship`

Got: All referenced skills exist on disk. Bidirectional cross-references in place. No orphaned links.

If fail: Referenced skill doesn't exist? Either remove reference or note as planned future skill with comment. Many skills reference this one but not listed in Related Skills? Add most relevant 2-3.

Step 6: Bump Version in Frontmatter

Update metadata.version field following semantic versioning:

  • Patch bump (1.0 to 1.1): Typo fixes, minor clarifications, URL updates
  • Minor bump (1.0 to 2.0): New procedure steps, significant content additions, structural changes
  • Note: Skills use simplified two-part versioning (major.minor)

Also update any date fields if present in frontmatter.

Got: Version bumped appropriately. Change magnitude matches update scope.

If fail: Current version cannot be parsed? Set to "1.1". Add comment noting version history gap.

Checks

  • All procedure steps contain current, executable code or concrete instructions
  • No stale version references, URLs, or deprecated tool names remain
  • Every procedure step has Expected: and On failure: blocks
  • Common Pitfalls section has 3-6 specific pitfalls with symptoms, fixes
  • All Related Skills cross-references point to existing skills
  • Bidirectional cross-references in place for closely related skills
  • Version in frontmatter bumped appropriately
  • Line count remains under 500 after updates
  • SKILL.md still passes review-skill-format validation after changes

Pitfalls

  • Update code without testing: Changing command in procedure step without verifying it works = worse than leaving old command. Uncertain? Add verification comment rather than untested replacement.
  • Over-expand pitfalls: Adding 10+ pitfalls dilutes section. Keep 3-6 most impactful pitfalls. Move edge cases to references/ file if needed.
  • Break cross-references during updates: When renaming skill or changing domain, grep entire skills library for references to old name. Use grep -rl "old-name" skills/ to find all occurrences.
  • Forget to bump version: Every content update, no matter how small, should bump version. Allows consumers to detect when skill changed.
  • Scope creep into refactoring: Content updates improve what skill says. Find yourself restructuring sections or extracting to references/? Switch to refactor-skill-structure skill instead.

See Also

  • review-skill-format — Run format validation before content updates. Ensure base structure sound
  • refactor-skill-structure — When content updates push skill over 500 lines, refactor structure to make room
  • evolve-skill — For deeper changes beyond content updates (e.g., creating advanced variant)
  • create-skill — Reference canonical format spec when adding new sections or procedure steps
  • repair-broken-references — Use for bulk cross-reference repair across entire skills library

GitHub リポジトリ

pjt222/agent-almanac
パス: i18n/caveman/skills/update-skill-content
0
agentsagentskillsai-assisted-developmentclaude-codeskillsteams

関連スキル

content-collections

メタ

このスキルは、Content Collections(Markdown/MDXファイルを型安全なデータコレクションに変換するTypeScriptファーストのツール)の本番環境でテストされた設定を提供します。Zodバリデーションによる型安全性を実現し、ブログ、ドキュメントサイト、コンテンツ重視のVite + Reactアプリケーション構築時にご利用ください。Viteプラグインの設定、MDXコンパイルから、デプロイ最適化、スキーマバリデーションまで、すべてを網羅しています。

スキルを見る

polymarket

メタ

このスキルは、開発者がPolymarket予測市場プラットフォームを活用したアプリケーション構築を可能にします。API統合による取引や市場データの取得に加え、WebSocketを介したリアルタイムデータストリーミングにより、ライブ取引や市場活動を監視できます。取引戦略の実装や、ライブ市場更新を処理するツールの作成にご利用ください。

スキルを見る

creating-opencode-plugins

メタ

このスキルは、開発者がコマンド、ファイル、LSP操作など25種類以上のイベントタイプにフックするOpenCodeプラグインを作成することを支援します。JavaScript/TypeScriptモジュール向けに、プラグイン構造、イベントAPI仕様、および実装パターンを提供します。カスタムイベント駆動ロジックでOpenCode AIアシスタントのライフサイクルをインターセプト、監視、または拡張する必要がある場合にご利用ください。

スキルを見る

sglang

メタ

SGLangは、高性能なLLMサービングフレームワークであり、RadixAttentionプレフィックスキャッシュを活用したJSON、正規表現、エージェントワークフロー向けの高速で構造化された生成を特長とします。特にプレフィックスが繰り返されるタスクにおいて、大幅に高速な推論を実現し、複雑な構造化出力やマルチターン対話に最適です。制約付きデコードが必要な場合や、広範なプレフィックス共有を伴うアプリケーションを構築する場合は、vLLMなどの代替案ではなくSGLangを選択してください。

スキルを見る