rsi-phase3-metasketills

RSI Phase 3: Cross-Domain Metasketill Extraction

Purpose

Extract reusable domain-agnostic metasketills from accumulated Phase 2 learned proposals, promote to permanent core rules (P8, P9, etc.), and build the pattern library for Phase 4 Benchmark.

Phase 3 Trigger ✅

Condition: Phase 2 acceptance rate ≥ 0.5 (≥5 learned/10 cycles)
Status: Infrastructure documented, data accumulation pending

Metasketill Definition

Domain-agnostic patterns vs. domain-specific skills.

Extraction Algorithm (3-Phase)

• 3A: Accumulate from improvements.jsonl, cluster by family
• 3B: Abstract each cluster to domain-agnostic pattern (name, trigger, recipe, anti-patterns)
• 3C: Validate against ≥2 new files, promote to permanent rules (P8, P9+)

Success Criteria

≥3 metasketills + ≥2 permanent rules → transition to Phase 4

Phase 3 → Phase 4

Export patterns to Benchmark schema, build test cases, publicly make Benchmark Suite.

Safety

Never promote failing validations; log all extractions; keep learned-proposal fallback.

Files

• Skill: rsi-phase3-metasketills/SKILL.md
• Library: rsi-pattern-library/ (emerging)
• World Model: phase3 design log
• Logging: improvements.jsonl with "metasketill" tag

Checklist

• Accumulate 50-cycle learned proposals
• Cluster by pattern family (timing/quality/safety)
• Abstract to domain-agnostic templates
• Validate against 2+ new files
• Promote 3+ as permanent rules (P8, P9, P10)
• Measure acceptance rate before/after

Expected Outcome

3-5 reusable metasketills, 2-3 permanent rules, measurable acceptance rate improvement, Phase 4 benchmark foundation.

RSI Phase 3: Extract cross-domain metasketills from learned proposals, promote to permanent core rules, build Pattern Library for Phase 4 Benchmark.