Task brief for an AI agent

An AI coding agent does better work with a short, specific brief than with a long conversation. The brief below gives it a goal, the boundaries, a definition of done and the points where it must stop and ask. It also says what to report, so a person can review the result quickly.

Standing rules, such as build commands, folders it must not touch and coding conventions, belong in the repository’s AGENTS.md (or the agent’s own instruction file). The brief covers only this task.

Template

# Task: short title

## Goal

One or two sentences: the outcome wanted, and why.

## Requirement and context

- Requirement IDs: REQ-...
- Related ADRs: ADR-...
- Files or areas in scope:

## Constraints

- Do not touch (as well as the boundaries in AGENTS.md):
- Keep the change as small as it can be. Do not refactor beyond the task.
- Prefer the simple solution. State any assumption rather than guess.

## Definition of done

- Build, tests and lint pass locally, using the commands in AGENTS.md.
- Tests added or updated, and named after the requirement IDs.
- Documentation updated.
- Commits and pull request follow the conventions, with `Refs:` and
  `Assisted-by:` trailers.

## Ask first if

- The change needs a new dependency, an interface change or an ADR.
- A requirement is missing, unclear or contradicts another.
- Anything touches a boundary listed in AGENTS.md.

## Report back

- What you did and why.
- Assumptions you made.
- Risks and anything left undone.
- Evidence: test and analysis output.

Review the agent’s work as you would a colleague’s: against the code review checklist, with the same bar for tests and evidence.