Long-running agents
Suspend at every tool call, resume from the store, drop forged events, know when to stop.
What you will build
By the end you will have a run that survives restarts: it suspends at every tool call, keeps all state in a store, resumes only from signed results for the exact call it is waiting on, and stops itself on a deadline or a stall.
Why it matters
A long agent run held in memory dies with the process, and a deploy at the wrong minute loses an hour of work or, worse, repeats a refund. Webhooks that resume runs are another door: if anyone can post a result, anyone can tell your agent that order 1042 cost $4,800.
Inside this module
- 1.Suspend at every tool call
Advance a run to its next tool call, save it, and stop. - 2.Resume only from signed events
Apply a result only if it is signed and matches the exact pending call. - 3.Know when to stop
End a run on a deadline or when it stops making progress.
Every step comes with the exact code diff, a run command, the expected output, a common mistake, a production note and a checkpoint. Then 2 graded exercises (wake, wouldApply) checked by npm test, a solution branch, and a 3-question quiz. TypeScript, scripted model, no API key.
This lesson unlocks with the full course
The full course is delivered as private GitHub repos with the lessons, exercises and solutions. After you buy, you get a read-access invite, so you need a GitHub username.
Checkout opens soon. The waitlist gets first access at the $149 founding price (regular $199), for the first 50 buyers or through Nov 8. No spam, only course updates.