Add hooks or middleware

Sometimes it’s useful to have code that runs before or after an agent run. Examples include preprocessing of input like http request, checking auth, or running guardrails on responses.

Maybe defined like tools, and opt in via frontmatter.

Also look at hooking into other parts of the lifecycle.

Start by mapping the agent run lifecycle, including the before-run and after-run points described in the issue, and review how tools and frontmatter are represented. Done should define an agreed hook or middleware scope covering input preprocessing, authorization, response guardrails, and any additional lifecycle points.

Written by the indexing model from the issue text.