Genie Metadata Optimizer

Analyzes evaluation failures and generates metadata change proposals using GEPA or structured LLM introspection. Produces control lever mappings consumed by the Applier worker.

When to Use This Skill

• Evaluation scores are below target after benchmarking
• Need to generate metadata change proposals from failure patterns
• Running GEPA-powered metadata evolution (L2)
• Running failure clustering and ASI-grounded introspective analysis (L1)
• Optimizing judge prompt accuracy (SIMBA)

Inputs (from Orchestrator)