Agent Skill Engineering

Frame a portable skill before writing it, create or update it under an explicit write boundary, or review it before making a measured optimization. Framing and review are read-only; every write, and every optimization, needs its own explicit authorization, and an optimization needs an observed failure or a measured baseline before it starts.

Sixteen governed topics cover trigger quality, progressive disclosure, and deterministic script and exit contracts. When a compatible knowledge provider is available the workflows call it explicitly; when none is, they keep their full baseline rather than guessing. Candidate files and provider replies are treated as untrusted evidence throughout, and authentication stays outside the model’s context.

agentbundle install --pack agent-skill-engineering --scope user

claude plugin install agent-skill-engineering@agent-ready-repo