executing-plans

Executing Plans

Take a written plan and walk it task-by-task. Stop for user review at defined checkpoints.

Before starting

1. Read the entire plan once before doing anything.
2. Confirm you're in the right repo / branch / worktree.
3. Make sure the working tree is clean. If not, ask the user.

Per-task loop

For each task in the plan:

1. Announce which task you're starting (number + name).
2. Execute every step in order. Do not skip steps. Do not batch them silently — work the steps sequentially.
3. For TDD steps: invoke test-driven-development.
4. Verify the task is complete by running the verification command defined in the plan. Quote the actual output.
5. Commit as the plan instructs.
6. Tick the checkbox in the plan file to mark the task done.
7. Pause if the plan flags this task as a checkpoint, OR if you hit anything unexpected.

Checkpoints

A checkpoint is any of:

• Explicit "pause here" markers in the plan
• End of a logical section (e.g., a major feature complete)
• Any step that fails or produces unexpected output
• Anytime you'd be making a non-trivial decision the plan doesn't cover

At a checkpoint, stop and ask the user before proceeding.

When the plan is wrong

Plans get details wrong. If a step doesn't make sense, doesn't apply, or is missing context: STOP. Don't improvise. Tell the user what's wrong and propose an update to the plan.

Done

When the last task is complete and verified:

1. Run any whole-system verification (full test suite, smoke test).
2. Invoke verification-before-completion.
3. Hand off to requesting-code-review if the plan is heading to a PR.