skills-management

To find and install an AI coding agent skill, use a skill discovery process that searches local and remote repositories. It matches your requested capability against available skills, then installs the selected skill across multiple agents and scopes.

SkillOpt optimization improves agent skills by running bounded-edit loops with held-out validation. It iteratively refines SKILL.md content, buffers rejected edits, and applies epoch-boundary slow updates to measurably enhance skill reliability.

Yes, you can manage skill lifecycles across multiple agents and scopes. This includes listing, showing details, deleting, moving, copying, and transferring skills between agents to maintain consistent multi-agent environments.

To run a SkillOpt-style optimization loop, you need a measurable verifier and a task set formatted in tasks.jsonl. Access to the repository's deterministic scripts for listing, reviewing, and discovery is also required.

The best way to review an installed agent skill is by executing a validation-driven review script. This process analyzes the SKILL.md content and structural files to ensure the skill meets deterministic quality standards.

You can update or remove outdated skills from coding agents by executing targeted lifecycle operations. This process uninstalls or replaces the deprecated skill versions across local and remote scopes to keep your agents current.