update-skill-content
について
このスキルは、既存のSKILL.mdファイルを更新し、そのドキュメントの品質と正確性を向上させます。手順の洗練、一般的な落とし穴の拡充、関連スキルの同期に重点を置き、特にコンテンツが古くなっていたり不明確な場合に対応します。API変更後やユーザーフィードバックに基づいて、スキルドキュメントを維持管理するためにご利用ください。
クイックインストール
Claude Code
推奨npx skills add pjt222/agent-almanac -a claude-code/plugin add https://github.com/pjt222/agent-almanacgit 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 an existing SKILL.md by refining procedure steps, expanding Common Pitfalls with real failure modes, synchronizing the Related Skills section, and bumping the version number. Use this after a skill passes format validation but has content gaps, stale references, or incomplete procedures.
When to Use
- A skill's procedure steps reference outdated tools, APIs, or version numbers
- Common Pitfalls section is thin (fewer than 3 pitfalls) or missing real failure modes
- Related Skills section has broken cross-references or is missing relevant links
- Procedure steps lack concrete code examples or have vague instructions
- A new skill has been added to the library that should be cross-referenced from existing skills
- After receiving feedback that a skill's procedures are unclear or incomplete
Inputs
- Required: Path to the 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)
Procedure
Step 1: Read Current Skill and Assess Content Quality
Read the entire SKILL.md and evaluate each section for completeness and accuracy.
Assessment criteria per section:
- When to Use: Are triggers concrete and actionable? (3-5 items expected)
- Inputs: Are types, defaults, and required/optional clearly separated?
- Procedure: Does each step have concrete code, Expected, and On failure?
- Validation: Are checklist items objectively testable? (5+ items expected)
- Common Pitfalls: Are pitfalls specific with symptoms and fixes? (3-6 expected)
- Related Skills: Do referenced skills exist? Are obvious related skills missing?
Got: A clear picture of which sections need improvement, with specific gaps identified.
If fail: If the skill cannot be read (path error), verify the path. If the 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, and 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 that has changed
- Package names that have been renamed or deprecated
- Configuration file formats that have 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: A list of potentially stale references with line numbers. Each reference is verified as current or flagged for update.
If fail: If too many references to check manually, 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:
- Verify code blocks still execute correctly or reflect current best practices
- Add missing context sentences that explain why the step is needed
- Ensure concrete commands use real paths, real flags, and real output
- Update Expected blocks to match current tool behavior
- Update On failure blocks with current error messages and fixes
When updating code blocks, preserve the original structure:
- Keep step numbering consistent
- Maintain the
### Step N: Titleformat - Do not reorder steps unless the original order was incorrect
Got: All procedure steps contain current, executable code. Expected/On failure blocks reflect actual current behavior.
If fail: If unsure whether a code block is still correct, add a note: <!-- TODO: Verify this command against current version -->. Do not remove working code blocks to replace with untested alternatives.
Step 4: Expand Common Pitfalls
Review the Common Pitfalls section and expand if gaps exist.
Quality criteria for pitfalls:
- Each pitfall has a bold name followed by a specific description
- The description includes the symptom (what goes wrong) and the fix (how to avoid or recover)
- Pitfalls are drawn from real failure modes, not hypothetical concerns
- 3-6 pitfalls is the target range
Sources for new pitfalls:
- Procedure steps with complex On failure blocks (these are likely pitfalls)
- Related skills that warn about the same tools or patterns
- Common issues reported by users of the procedure
Got: 3-6 pitfalls, each with a specific symptom and fix. No generic pitfalls like "be careful" or "test thoroughly".
If fail: If only 1-2 pitfalls can be identified, that is acceptable for basic-complexity skills. For intermediate and advanced skills, fewer than 3 pitfalls suggests the author has not fully explored failure modes — flag this for future expansion.
Step 5: Synchronize Related Skills Section
Verify all cross-references in the Related Skills section are valid and add any missing links.
- For each referenced skill, verify it exists:
# Check if referenced skill exists test -d skills/referenced-skill-name && echo "EXISTS" || echo "NOT FOUND" - Search for skills that reference this skill (they should be cross-linked):
# Find skills that reference this skill grep -rl "skill-name" skills/*/SKILL.md - Check for obvious related skills based on domain and tags
- Use the format:
- \skill-id` — one-line description of the relationship`
Got: All referenced skills exist on disk. Bidirectional cross-references are in place. No orphaned links.
If fail: If a referenced skill does not exist, either remove the reference or note it as a planned future skill with a comment. If many skills reference this one but are not listed in Related Skills, add the most relevant 2-3.
Step 6: Bump Version in Frontmatter
Update the 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 the frontmatter.
Got: Version is bumped appropriately. The change magnitude matches the update scope.
If fail: If the current version cannot be parsed, set it to "1.1" and add a comment noting the version history gap.
Validation
- 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 and fixes
- All Related Skills cross-references point to existing skills
- Bidirectional cross-references are in place for closely related skills
- Version in frontmatter has been bumped appropriately
- Line count remains under 500 after updates
- SKILL.md still passes
review-skill-formatvalidation after changes
Pitfalls
- Updating code without testing: Changing a command in a procedure step without verifying it works is worse than leaving the old command. When uncertain, add a verification comment rather than an untested replacement.
- Over-expanding pitfalls: Adding 10+ pitfalls dilutes the section. Keep the 3-6 most impactful pitfalls; move edge cases to a
references/file if needed. - Breaking cross-references during updates: When renaming a skill or changing its domain, grep the entire skills library for references to the old name. Use
grep -rl "old-name" skills/to find all occurrences. - Forgetting to bump version: Every content update, no matter how small, should bump the version. This allows consumers to detect when a skill has changed.
- Scope creep into refactoring: Content updates improve what the skill says. If you find yourself restructuring sections or extracting to
references/, switch to therefactor-skill-structureskill instead.
Related Skills
review-skill-format— Run format validation before content updates to ensure the base structure is soundrefactor-skill-structure— When content updates push the skill over 500 lines, refactor structure to make roomevolve-skill— For deeper changes that go beyond content updates (e.g., creating an advanced variant)create-skill— Reference the canonical format spec when adding new sections or procedure stepsrepair-broken-references— Use for bulk cross-reference repair across the entire skills library
GitHub リポジトリ
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。
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。
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。
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。
