# Coding agent task checklist

A one-page checklist for handing work to an AI coding agent. Copy it into your repo or ticket template.
By Morteza Bazrafshan, bazrafshan.ir.

## Before you start
- [ ] The task has a visible end state ("the page does X", "the test Y passes").
- [ ] I know how to verify it (test, command, screenshot).
- [ ] Scope is small enough for one reviewable change.
- [ ] The agent knows where to look: relevant files, docs, related tickets.
- [ ] A context file exists in the repo (how to run tests, layout, conventions, off-limits areas).

## Boundaries
- [ ] Allowed without asking: read files, run tests, lint, type-check.
- [ ] Needs my approval: installing dependencies, schema changes, deleting files, network calls to new hosts.
- [ ] Never: pushing to shared branches, touching secrets, production data, sending messages to customers.
- [ ] Secrets are not in the agent's environment.

## While it works
- [ ] It runs the checks and reads the failures.
- [ ] It stays on the task (no unrequested refactors).
- [ ] I stop and re-scope if it loops or guesses.

## Review the diff
- [ ] Does it do what was asked, and nothing else?
- [ ] Error handling and edge cases covered?
- [ ] Tests added or updated for the behaviour?
- [ ] No leftover debug code, commented-out blocks or stray files?
- [ ] Dependencies, permissions and config changes reviewed by hand.

## After merging
- [ ] If the agent repeated a mistake, add a rule to the context file.
- [ ] Note anything surprising for next time.