Every command an agent was stopped from running, across every session. Approving one lets that agent retry it; it does not change what the agent is allowed to do next time.

An agent is a job description, not a program: what it may read, what it may change, and how it should answer. A new one starts as a draft: anyone can try it, but no unaddressed message is routed to it until you publish it.

A skill is a method — how to do something, reusable by anyone. Write one when several agents need the same procedure, instead of copying it into each prompt. Every agent can reach every skill, so the description matters more than it looks: it is all an agent sees when deciding whether this is the right method for the job.

A schedule runs an agent on a timer. Each firing is an ordinary session, with the same trace and in the same list.