What problem does it solve? Creating high-quality AI agent skills requires consistent structure, security hardening, objective evaluation, and trigger tuning, which is difficult to do manually and consistently across platforms like Claude Code, Cursor, and OpenAI Codex. ## Core Features & Use Cases - Skill Scaffolding & Authoring: Generate standardized skill directories with SKILL.md, scripts, references, assets, and evals using init_skill.py and archetype patterns (workflow, cli-tool, hybrid). - Benchmarking & Evaluation: Run parallel with-skill vs baseline evals, grade outputs with grader subagents, and aggregate pass rates, timing, and token statistics into benchmark reports. - Trigger Optimization & Packaging: Optimize frontmatter descriptions against train/test query splits with run_loop.py, then package skills into portable .skill archives for cross-agent deployment. - Use Case: After completing a complex workflow in a session, ask the agent to distill it into a reusable skill, benchmark it against a baseline, and deploy it to your preferred agent platform. ## Quick Start Turn the workflow we just completed into a reusable skill and benchmark it against a baseline run.
