evolve

This command improves skill .md files based on lessons accumulated in the global skill lesson store.

Argument handling

The user may invoke this command with arguments:

• /autocontext:evolve, default: scan and evolve interactively
• /autocontext:evolve --rollback <skill-name>, restore from backup
• /autocontext:evolve --export, export lessons to JSON
• /autocontext:evolve --import <path>, import lessons from JSON

Parse the arguments from the user's input. If --rollback is present, run the rollback flow. If --export or --import, run the sync flow. Otherwise, run the default evolution flow.

Default evolution flow

1. Run the scan script to find eligible skills:

   bash "${CLAUDE_PLUGIN_ROOT}/scripts/skill-evolution/scan_eligible.sh"
   

2. If no skills are eligible, report that and exit.

3. Present the summary to the user using AskUserQuestion: which skill(s) to evolve? Include lesson count and average confidence per skill. Add an "All" option and a "Skip" option.

4. For each selected skill: a. Read the current skill .md file using the find_skill_path function:

   python3 -c "
   import sys
   sys.path.insert(0, '${CLAUDE_PLUGIN_ROOT}/scripts/skill-evolution')
   from apply_edit import find_skill_path
   path = find_skill_path('SKILL_NAME')
   print(path or 'NOT_FOUND')
   "
   

   b. If NOT_FOUND, warn and skip. c. Read the skill file content. d. Run generate_diff to create an improved version:

   python3 -c "
   import sys, json
   sys.path.insert(0, '${CLAUDE_PLUGIN_ROOT}/scripts/skill-evolution')
   from generate_diff import generate_evolved_skill
   from store import get_eligible_lessons
   
   lessons = get_eligible_lessons('SKILL_NAME')
   with open('SKILL_PATH') as f:
       content = f.read()
   result = generate_evolved_skill(content, lessons)
   if result:
       print(result)
   else:
       print('GENERATION_FAILED')
   "
   

   e. If generation failed, offer the append fallback. f. Show a diff between the original and evolved content. g. Ask the user via AskUserQuestion:

   • Accept, apply the edit
   • Edit, let user make manual changes first
   • Reject, skip, lessons stay
   • Append instead, use the fallback section h. Apply the chosen action using apply_edit functions. i. Mark evolved lessons as folded.