Examples over prose

Show one example that matches the current code. When a procedure has several steps, turn it into a script, a CLI command or a package, so there's nothing to follow by hand.

A guide then keeps three things: when it applies, what to run, and what failure looks like.

Example

Minion's built-in guide for failed runs is five lines, each a call you can paste:

1. Is anything broken? await minion.runs.summary()
2. Which runs failed? await minion.runs.list({ status: 'error', limit: 10 })
3. Why? await minion.runs.get({ id }) returns log lines, the error and stack.
4. Fix the package (minion.packages.save) and re-run the job with minion.jobs.runNow.
5. An execute call timed out on your side? Poll minion.runs.get({ id: runId }) until status isn't running.

Agents find it with search, next to the capabilities it mentions. See Search and execute .

The same rule applies inside the project. Agents verify Minion with one checked-in control CLI rather than writing a throwaway curl script each time, and a test fails if that CLI's feature map drifts from the real routes.

In your packages

If you catch yourself writing "first do this, then that, then check this" in a README or a team skill, write the package instead. A team skill should say when to use a package and what its errors mean, not repeat its code in prose.