skill evolution manager

Skill Evolution Manager

This is the "evolution hub" of the entire AI skill system. It is responsible not only for optimizing individual Skills, but also for cross-Skill experience review and accumulation.

Core Responsibilities

1. Session Review: At the end of a conversation, analyze the performance of all invoked Skills.
2. Experience Extraction: Convert unstructured user feedback into structured JSON data (evolution.json）。
3. Smart Stitching: Automatically write the accumulated experience into SKILL.md, ensuring persistence and preventing it from being overwritten by version updates.

Use Cases

Trigger:

• /evolve
• "Review the conversation just now"
• "I think that tool just now wasn't very easy to use, record it"
• "Save this experience to the Skill"

Workflow (The Evolution Workflow)

1. Experience Review (Review & Extract)

When the user triggers a review, the Agent must execute:

1. Scan Context: Identify points the user was dissatisfied with (errors, wrong style, incorrect parameters) or satisfied with (a specific Prompt worked well).
2. Locate Skill: Determine which Skill needs to evolve (for example yt-dlp or baoyu-comic）。
3. Generate JSON: Build the following JSON structure in memory:

   {
     "preferences": ["用户希望下载默认静音"],
     "fixes": ["Windows 下 ffmpeg 路径需转义"],
     "custom_prompts": "在执行前总是先打印预估耗时"
   }
   

2. Experience Persistence (Persist)

The Agent calls scripts/merge_evolution.py, incrementally writing the above JSON into the target Skill's evolution.json file.

• Command: python scripts/merge_evolution.py <skill_path> <json_string>

3. Document Stitching (Stitch)

The Agent calls scripts/smart_stitch.py, converting the content of evolution.json into Markdown and appending it to the end of SKILL.md .

• Command: python scripts/smart_stitch.py <skill_path>

4. Cross-Version Alignment (Align)

When skill-manager has updated a certain Skill, the Agent should proactively run smart_stitch.pyto "re-stitch" the previously saved experience into the new version of the document.

Core Scripts

• scripts/merge_evolution.py: Incremental merge tool. Responsible for reading the old JSON, deduplicating and merging the new List, and saving it.
• scripts/smart_stitch.py: Document generation tool. Responsible for reading JSON and generating or updating at the end of SKILL.md the ## User-Learned Best Practices & Constraints section.
• scripts/align_all.py: Full alignment tool. One-click traversal of all Skill folders to re-stitch existing evolution.json experience back into the corresponding SKILL.md. Commonly used for experience restoration after skill-manager batch updates.

Best Practices

• Do not directly modify the body of SKILL.md: Unless it is an obvious spelling error. All experience corrections should be made through the evolution.json channel, so that experience is not lost when the Skill is upgraded.
• Multi-Skill Collaboration: If a conversation involves multiple Skills, execute the above process for each Skill in turn.