Welcome! Register below to get the join link and a calendar invite by email.
A few weeks ago I showed the set of Claude skills I built for my own job search. This session is the sequel: I stopped running those skills by hand and turned them into agents that run on a schedule, update my tracker, and report back through a live status page.
A skill makes Claude excellent at one task when you ask for it. It does nothing when you forget, and it never hands its result to the next step on its own. So you end up being the glue: remembering, opening a chat, pasting context, copying the result somewhere.
You'll see the real setup, the prompts behind it, what broke, and how to copy it for any repeating process, not only job hunting.
What we will cover
- The skills toolbox: why a skill only becomes agent material once its output lands in a predictable place
- From skills to agents: two scheduled agents, the loop every run follows, and the anatomy of a prompt that runs with no memory of yesterday
- Guardrails: add-only access, no outward actions, and visible judgment, so it is safe to leave alone
- The status page: a live page that shows where everything stands, without opening a chat
- Live demo: a real run, from the trigger to the updated status page
- What went wrong: honest lessons and the rules I follow now
What you will walk away with
A clear model for going from a single skill to an agent that runs on its own, a prompt structure you can copy, a blueprint for a status page, and a short list of mistakes to skip.
Who it is for
Anyone who already uses Claude skills and feels they are still the glue between them. And anyone with a repeating process (hiring, research, reporting, lead follow-up) who wants it to run without being asked. No coding needed.