evolve-agent
について
`evolve-agent`スキルは、既存のエージェントのコア定義を洗練させたり、高度なバリアントを作成したりすることで更新します。このスキルは、エージェントの評価、そのスキルやツールへの変更の適用、メタデータやレジストリ参照の更新を扱います。エージェントの能力が陳腐化した場合、フィードバックからギャップが明らかになった場合、あるいは実世界での使用後にその適用範囲を調整する必要がある場合に、このスキルを使用してください。
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/evolve-agent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
Evolve an Existing Agent
Fix, grow, or make advanced variant of agent first made with create-agent. This proc covers upkeep side of agent life: check gaps vs best practices, apply tight fixes to persona def, bump versions, keep registry and cross-refs in sync.
When Use
- Agent skills list stale after new skills added to library
- User feedback shows missing capabilities, unclear purpose, or weak examples
- Tool rules shifted (new MCP server, tool gone, privilege drop needed)
- Agent scope needs sharpening — overlaps with another agent or too broad
- Advanced variant needed next to original (e.g.,
r-developerandr-developer-advanced) - Related agents or teams added, cross-refs in See Also are stale
Inputs
- Required: Path to existing agent file to evolve (e.g.,
agents/r-developer.md) - Required: Evolve trigger (feedback, new skills, tool change, scope overlap, team integration, spotted limits)
- Optional: Target version bump size (patch, minor, major)
- Optional: Make advanced variant vs refine in-place (default: refine in-place)
Steps
Step 1: Assess the Current Agent
Read existing agent file and check each section vs quality list from guides/agent-best-practices.md:
| Section | What to Check | Common Issues |
|---|---|---|
| Frontmatter | All required fields present (name, description, tools, model, version, author) | Missing tags, stale version, wrong priority |
| Purpose | Specific problem statement, not generic "helps with X" | Vague or overlapping with another agent |
| Capabilities | Concrete, verifiable capabilities with bold lead-ins | Generic ("handles development"), no grouping |
| Available Skills | Matches frontmatter skills list, all IDs exist in registry | Stale IDs, missing new skills, lists default skills unnecessarily |
| Usage Scenarios | 2-3 realistic scenarios with invocation patterns | Placeholder text, unrealistic examples |
| Examples | Shows user request and agent behavior | Missing or trivial examples |
| Limitations | 3-5 honest constraints | Too few, too vague, or missing entirely |
| See Also | Valid cross-references to agents, guides, teams | Stale links to renamed or removed files |
# Read the agent file
cat agents/<agent-name>.md
# Check frontmatter parses
head -20 agents/<agent-name>.md
# Verify skills in frontmatter exist in registry
grep "skills:" -A 20 agents/<agent-name>.md
# Check if agent is referenced by any team
grep -r "<agent-name>" teams/*.md
Got: List of specific gaps, weak spots, or fix chances sorted by section.
If fail: Agent file not exist or no frontmatter? This skill not apply — use create-agent instead to make from scratch.
Step 2: Gather Evolution Requirements
Spot and sort what fired the evolution:
| Trigger | Example | Typical Scope |
|---|---|---|
| User feedback | "Agent missed XSS in review" | Add skill or capability |
| New skills available | Library gained analyze-api-security | Update skills list |
| Tool change | New MCP server available | Add to tools/mcp_servers |
| Scope overlap | Two agents both claim "code review" | Sharpen purpose and limitations |
| Team integration | Agent added to a new team | Update See Also, verify capabilities |
| Model upgrade | Task requires deeper reasoning | Change model field |
| Privilege reduction | Agent has Bash but only reads files | Remove unnecessary tools |
Log specific changes needed before edit. List each change with target section:
- Frontmatter: add `new-skill-id` to skills list
- Capabilities: add "API Security Analysis" capability
- Available Skills: add `new-skill-id` with description
- Limitations: remove outdated limitation about missing skill
- See Also: add link to new team that includes this agent
Got: Concrete list of changes, each mapped to specific section of agent file.
If fail: Changes unclear? Ask user for clarity before go on. Vague evolution goals give vague fixes.
Step 3: Choose Evolution Scope
Use this pick matrix to decide refine in-place or make variant:
| Criteria | Refinement (in-place) | Advanced Variant (new agent) |
|---|---|---|
| Agent ID | Unchanged | New ID: <agent>-advanced or <agent>-<specialty> |
| File path | Same .md file | New file in agents/ |
| Version bump | Patch or minor | Starts at 1.0.0 |
| Model | May change | Often higher (e.g., sonnet → opus) |
| Registry | Update existing entry | New entry added |
| Original agent | Modified directly | Left intact, gains See Also cross-reference |
Refinement: Pick when update skills, fix docs, sharpen scope, or tune tools. Agent keeps its identity.
Variant: Pick when evolved version would serve a different audience, need different model, or add capabilities that would make original too broad. Original stays as-is for simpler use cases.
Got: Clear pick — refine or variant — with reason.
If fail: Unsure? Default to refine. Can always pull variant later; harder to merge one back.
Step 4: Apply Changes to the Agent File
For Refinements
Edit existing agent file direct:
- Frontmatter: Update
skills,tools,tags,model,priority,mcp_serversas need - Purpose/Capabilities: Revise to show new scope or added function
- Available Skills: Add new skills with desc, remove deprecated ones
- Usage Scenarios: Add or revise scenarios to show new capabilities
- Limitations: Remove limits no longer apply, add new honest ones
- See Also: Update cross-refs to show current agent/team/guide landscape
Follow these edit rules:
- Keep all existing sections — add content, do not remove sections
- Keep Available Skills section in sync with frontmatter
skillslist - Do not add default skills (
meditate,heal) to frontmatter unless core to agent methodology - Check each skill ID exists:
grep "id: skill-name" skills/_registry.yml
For Variants
# Copy the original as a starting point
cp agents/<agent-name>.md agents/<agent-name>-advanced.md
# Edit the variant:
# - Change `name` to `<agent-name>-advanced`
# - Update `description` to reflect the advanced scope
# - Raise `model` if needed (e.g., sonnet → opus)
# - Reset `version` to "1.0.0"
# - Expand skills, capabilities, and examples for the advanced use case
# - Reference the original in See Also as a simpler alternative
Got: Agent file (refined or new variant) passes check list from Step 1.
If fail: Edit breaks doc shape? Use git diff to review changes and revert partial edits with git checkout -- <file>.
Step 4.5: Sync Translated Variants
Required when translations exist. This step applies to both human authors and AI agents following this procedure. Do not skip — stale
source_commitvalues causenpm run validate:translationsto report false staleness warnings across all locales.
Check whether translations exist for evolved agent and update to match new source state:
# Check for existing translations
ls i18n/*/agents/<agent-name>.md 2>/dev/null
If translations exist
- Get current source commit hash:
SOURCE_COMMIT=$(git rev-parse HEAD)
- Update
source_commitin each translated file frontmatter:
for locale_file in i18n/*/agents/<agent-name>.md; do
sed -i "s/^source_commit: .*/source_commit: $SOURCE_COMMIT/" "$locale_file"
done
- Flag files for re-translation by adding affected locales in commit msg:
evolve(<agent-name>): <description of changes>
Translations flagged for re-sync: de, zh-CN, ja, es
Changed sections: <list sections that changed>
- Regen translation status files:
npm run translation:status
If no translations exist
No action needed. Go to Step 5.
For variants
Wait translation of new variants until variant stabilizes (1-2 versions). Add translations after variant refined at least once.
Got: All translated files have source_commit updated to current commit. npm run translation:status exits 0.
If fail: sed fails to match frontmatter field? Open translated file by hand and check it has source_commit in its YAML frontmatter. Field missing? Re-scaffold with npm run translate:scaffold -- agents <agent-name> <locale>.
Step 5: Update Version and Metadata
Bump version field in frontmatter by semantic versioning:
| Change Type | Version Bump | Example |
|---|---|---|
| Typo fix, wording clarification | Patch: 1.0.0 → 1.0.1 | Fixed unclear limitation |
| New skills added, capability expanded | Minor: 1.0.0 → 1.1.0 | Added 3 new skills from library |
| Restructured purpose, changed model | Major: 1.0.0 → 2.0.0 | Narrowed scope, upgraded to opus |
Also update:
updateddate to current datetagsif agent domain coverage shifteddescriptionif purpose materially differentpriorityif agent importance vs others shifted
Got: Frontmatter version and updated match size and date of changes. New variants start at "1.0.0".
If fail: Forget to bump version? Next evolve will have no way to tell current state from old. Always bump before commit.
Step 6: Update Registry and Cross-References
For Refinements
Update existing entry in agents/_registry.yml to match revised frontmatter:
# Find the agent's registry entry
grep -A 10 "id: <agent-name>" agents/_registry.yml
Update description, tags, tools, and skills fields to match agent file. No count change needed.
Update cross-refs in other files if agent capabilities or name shifted:
# Check if any team references this agent
grep -r "<agent-name>" teams/*.md
# Check if any guide references this agent
grep -r "<agent-name>" guides/*.md
For Variants
Add new agent to agents/_registry.yml in alpha spot:
- id: <agent-name>-advanced
path: agents/<agent-name>-advanced.md
description: One-line description of the advanced variant
tags: [domain, specialty, advanced]
priority: normal
tools: [Read, Write, Edit, Bash, Grep, Glob]
skills:
- skill-id-one
- skill-id-two
Then:
- Bump
total_agentsat top of registry - Add See Also cross-ref in original agent pointing to variant
- Add See Also cross-ref in variant pointing to original
.claude/agents/symlink toagents/means variant auto-found
Got: Registry entry matches agent file frontmatter. For variants, total_agents equals real count of agent entries.
If fail: Count entries with grep -c "^ - id:" agents/_registry.yml and check it matches total_agents.
Step 7: Validate the Evolved Agent
Run full check list:
- Agent file exists at expected path
- YAML frontmatter parses with no errors
-
versionwas bumped (refinement) or set to "1.0.0" (variant) -
updateddate shows today - All required sections present: Purpose, Capabilities, Available Skills, Usage Scenarios, Examples, Limitations, See Also
- Skills in frontmatter match Available Skills section
- All skill IDs exist in
skills/_registry.yml - Default skills (
meditate,heal) not listed unless core to method - Tools list follows least-privilege rule
- Registry entry exists and matches frontmatter
- For variants:
total_agentscount matches real count on disk - Cross-refs two-way (original ↔ variant)
-
git diffshows no slip deletions from original content
# Verify frontmatter
head -20 agents/<agent-name>.md
# Check skills exist
for skill in skill-a skill-b; do
grep "id: $skill" skills/_registry.yml
done
# Count agents on disk vs registry
ls agents/*.md | grep -v template | wc -l
grep total_agents agents/_registry.yml
# Review all changes
git diff
Got: All check items pass. Evolved agent ready to commit.
If fail: Fix each failing item one by one. Most common post-evolve issues are stale skill IDs in Available Skills section and forgot updated date.
Validation
- Agent file exists and has valid YAML frontmatter
-
versionfield shows changes made -
updateddate is current - All sections present and in sync
- Frontmatter
skillsarray matches Available Skills section - All skill IDs exist in
skills/_registry.yml - Default skills not listed needlessly
- Registry entry matches agent file
- For variants: new entry in
agents/_registry.ymlwith right path - For variants:
total_agentscount updated - Cross-refs valid (no broken links in See Also)
- For refinements with translations:
source_commitupdated in all locale files -
git diffconfirms no slip content removal
Pitfalls
- Forget to bump version: No version bump, no way to track what changed or when. Always update
versionandupdatedin frontmatter before commit. - Stale translations after evolve: With 1,288+ translation files in repo, every agent evolve fires staleness in up to 4 locale files. Always check for existing translations with
ls i18n/*/agents/<agent-name>.mdand updatesource_commitin each, or flag them for re-translation in commit msg. - Skills list drift: Frontmatter
skillsarray and## Available Skillssection must stay in sync. Update one without other gives confusion for both humans and tools. - List default skills needlessly: Adding
meditateorhealto frontmatter when already inherited from registry. Only list if core to agent method (e.g.,mystic,alchemist). - Tool over-add during evolve: Adding
BashorWebFetchduring evolve "just in case." Every tool add must be backed by specific new capability. - Stale See Also after variant make: When making variant, both original and variant need to ref each other. One-way refs leave graph broken.
- Registry entry not updated: After changing agent skills, tools, or desc,
agents/_registry.ymlentry must update to match. Stale registry entries cause discovery and tool fails.
See Also
create-agent— base for making new agents; evolve-agent assumes this was followed firstevolve-skill— parallel proc for evolving SKILL.md filescommit-changes— commit evolved agent with clear msg
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。
