/optimize-claude - CLAUDE.md Hierarchy Optimization

Optimize the CLAUDE.md instruction hierarchy to reduce token bloat while preserving all critical rules.

Usage

• •/optimize-claude - Run full optimization (discovery + compression + validation)
• •/optimize-claude check - Measure current char counts and redundancy only

What This Does

Uses the Ralph Wiggum iterated self-correction pattern with 2 serial agents:

1. •Agent 1 (Compression): Deduplicates, compresses prose to tables, moves domain-specific content out of always-loaded files
2. •Agent 2 (Clarity): Reviews compression for LLM instruction effectiveness, restructures for front-loaded critical rules

Each iteration requires human approval before edits are applied. Git checkpoints enable rollback.

Targets

Activation

Before optimizing, run the ACTIVATION-CHECKLIST:

1. •Discover all CLAUDE.md files
2. •Measure baseline char counts
3. •Snapshot critical rules
4. •Git tag baseline

Protocol

Iteration N (max 3):
  Agent 1: Compress & deduplicate → structured proposals
  Agent 2: Review clarity & architecture → revised proposals
  Main: Synthesize, show diff, get human approval
  Execute: Apply edits, git commit
  Validate: Check all 14 critical rules
  Converge: Targets met? Plateau? Continue?

Files

• •ACTIVATION-CHECKLIST.md - Pre-execution discovery
• •protocols/agent-1-compression.md - Compression agent prompt
• •protocols/agent-2-clarity.md - Clarity agent prompt
• •protocols/convergence.md - When to stop
• •protocols/validation.md - Post-edit checks
• •reference/hierarchy-levels.md - Content placement guide
• •reference/critical-rules-checklist.yaml - Must-preserve rules
• •reference/llm-optimization-heuristics.md - Writing effective LLM instructions