agent-brief
Agent Brief — the plan TDD builds from
An agent brief is the authoritative build plan for a feature or fix: pure text — a structured plan, not code and not a code change. It is the contract the TDD stage executes: read top-to-bottom, its acceptance criteria are already a red→green build order.
Write it to .workflow/brief.md in the repo being worked on, not just into the conversation. Every later stage reads it, the harness parses it, and a subagent can only be pointed at a path; a plan that exists only in scrollback stops being authoritative around a loop's second iteration. (Writing a plan file is not a code change — the rule against emitting code stands.)
This is stage 2 of a seven-stage workflow: architect-deep-dive → agent-brief → acceptance-tests → tdd → refactor-review → devils-advocate → tyr-verdict. What each stage reads and leaves, and the rules they share, are in pipeline.md — read it once per session.
Requires: python3 (runs the bundled brief checker, scripts/brief.py, before the approval gate). git is not needed: this stage writes and moves files under .workflow/ and commits nothing. If python3 is missing, skip the check and say so in your closing lines.
Reference files
- pipeline.md — open it once per session, when you need what each stage reads and leaves or the rules all stages share.
- examples.md — open it when you are unsure what a good criterion or a good brief looks like.
- parallel-tracks.md — open it only when the design's decomposition or the brief's size suggests the work could split.
- delegation.md — open it when you are about to dispatch any subagent.