A feature-plan template for agents

When to use this

Use it before an agent writes anything non-trivial. A plan in a fixed shape is easy to read, easy to correct, and easy to turn into tasks. Ask for the plan first, fix what is wrong, then let the agent build it.

# Feature plan: [feature name]

## Goal
One or two sentences: what this feature does and who it is for.

## Acceptance criteria
- [ ] [observable outcome]
- [ ] [observable outcome]
- [ ] Tests cover the new behavior.

## Approach
How you will build it, and the main files you will touch.

## Tasks
1. [task] (P0): [acceptance note]
2. [task] (P1): [acceptance note]
3. [task] (P2): [acceptance note]

## Risks and unknowns
- [anything you are unsure about, or a decision you need from a person]

## Out of scope
- [what this plan does not do]

Why each section is there

• Goal keeps the feature to one clear sentence, so scope stays honest.
• Acceptance criteria are the shared definition of done, and they map to tests.
• Approach and tasks turn the plan into ordered, prioritized work.
• Risks and out of scope surface the unknowns and the decisions a person should make.

An agent can take this plan and work the tasks in order. For what that looks like in practice, see working with agents, and pair it with a CLAUDE.md so the agent knows the house rules too.