distill

Use cases

distill closes the loop between what your agent did and what it knows: past conversations become reusable skills.

Learning by doing. A background subagent reflects on your sessions and creates or updates skills from the patterns it sees. The agent gets better at the tasks you actually do, not generic ones.

Onboarding junior agents. The distilled skills encode your team's working style — commit conventions, review preferences, project idioms — so new sessions (or new people's agents) start from your norms.

Reducing repeat work. Every time you manually explain the same thing to an agent, that's a skill waiting to be distilled. Over weeks, the harness stops asking and starts knowing.

Practical tip: review distilled skills occasionally — delete stale ones, merge overlapping ones, and edit wording so they read like your standards. Distillation is a feedback loop: the better your sessions are, the better the skills it produces.