What problem does it solve? Optimizing an agent skill's SKILL.md file requires coordinating workspace setup, dataset synthesis, judge-mode selection, spec-compliance validation, and safe write-back, which is error-prone when done ad hoc. This Skill provides the orchestration contract for running that full trainer loop against skill-type targets. ## Core Features & Use Cases - Two-Concern Optimization: Separates frontmatter triggering (description field) from body content execution quality, routing observed failure modes like under-triggering or bloated context to the right concern. - Spec-Compliance Gating: Enforces agentskills.io rules before write-back, including required YAML fields, the 500-line progressive disclosure limit, unchanged name field, and isolation of evaluator-only fields. - Judge-Mode Inference: Defaults to llm_judge scoring for open-ended skill quality while honoring explicit row-level scoring declarations. - Use Case: A skill under-triggers because agents never invoke it. Use this Skill to initialize the trainer workspace, prioritize frontmatter description optimization, run an optimization pass, and write back a validated candidate. ## Quick Start Run the trainer loop on my skill at skills/researcher-research/SKILL.md to fix its under-triggering description and validate the result before write-back.
