Optimize Prompts

Run Phase 3 of the learning pipeline - route lessons and optimize rule files.

Usage

/rules-learning-pipeline:optimize-prompts <lessons-file> [--dry-run]

What It Does

1. Route lessons by validated scope:

   • user-general → ~/.claude/CLAUDE.md
   • cross-project-domain → ~/.claude/rules/*.md
   • project → project/CLAUDE.md
   • domain → context/rules/*.md

2. For each target file:

   • Merge with existing rules
   • Remove duplicates
   • Apply peripheral bias (critical at start/end)
   • Add emphasis keywords (MUST, NEVER, CRITICAL)
   • Enforce line limits (< 60 ideal, < 300 max)

3. Route expansion decisions:

   • Critical pattern (≥1 rule) → CREATE immediately
   • Standard pattern (≥3 rules) → CREATE domain file
   • Below threshold → Add to pending section

Execution

Task(
  subagent_type: "prompt-optimizer",
  prompt: "Optimize lessons at {lessons_file}:

    1. GROUP lessons by scope:
       - user-general → ~/.claude/CLAUDE.md
       - cross-project-domain → ~/.claude/rules/{framework}.md
       - project → project/CLAUDE.md
       - domain → context/rules/{domain}.md

    2. FOR each target file:
       - Read existing rules
       - Merge new lessons
       - Remove duplicates
       - Apply peripheral bias
       - Add emphasis (MUST/NEVER/CRITICAL)
       - Verify < 60 lines (warn if exceeded)

    3. ROUTE expansion:
       IF critical pattern → CREATE file immediately
       ELSE IF ≥3 rules → CREATE domain file
       ELSE → Add to pending section

    4. UPDATE router index

    Output to:
    - docs/evaluations/{session}-optimized.md
    - Updated rule files (unless --dry-run)"
)

Anthropic Best Practices Applied

Output

• docs/evaluations/{session}-optimized.md - Optimization report
• Updated ~/.claude/CLAUDE.md (if user-general rules)
• Updated project/CLAUDE.md (if project rules)
• Updated/created context/rules/*.md (if domain rules)