What problem does it solve? Teams repeatedly encode agent workflows as ad-hoc prompts that are untested, overlapping, and impossible to improve systematically. This Skill turns repeated agent workflows into auditable Agent Skill packages with clear routing boundaries, progressive resources, and evidence-based optimization. ## Core Features & Use Cases - Skill authoring and restructuring: Scaffold new SKILL.md packages, split or merge skills, and route conditional detail into references, scripts, and assets. - Deterministic auditing: Run structural, frontmatter, secret-signature, reference-integrity, and fixed-rubric evaluation checks without model or network calls. - Controlled optimization: Optimize skills through bounded edits, held-out validation, rejected-edit memory, and an untouched final test set. - Use Case: Convert a repeated incident-triage workflow into a portable skill, then prove with paired baseline trajectories and fixed rubrics that the new package repairs failures without regressing existing successes. ## Quick Start Ask the agent to turn your repeated workflow into an Agent Skill package with a routing boundary, fixed-rubric evaluation cases, and a baseline comparison.
