| name | milestone-task |
| description | Test-driven implementation flow for one TASKS.md item — load, red, green, refactor, verify, archive — driving the six-status vocabulary. Use when picking up any pending task. |
| argument-hint | [task id, e.g. #001] |
Work exactly one TASKS.md item ($ARGUMENTS) through pending → running → completed (or → needs-fix/→ broken if something goes wrong).
Steps
- Load the task. Read its
TASKS.md entry (Status/Files/Notes) and any spec it references.
- Flip status to
running in TASKS.md before starting.
- Red. Write one failing test capturing the smallest slice of the behavior; confirm it fails for the expected reason.
- Green. Minimum code to pass that one test (hardcoding is fine if only one test forces it).
- Refactor only with tests green — one transformation at a time, re-run after each.
- Repeat red→green→refactor per slice until the requirement is covered. If the task turns out bigger than one step, split it and note that in
TASKS.md — don't silently scope-creep.
- Gate. Run build + test + lint + format from
.claude/harness.env; all must pass.
- Verify per
.claude/rules/ship-verification.md — actually run the behavior end-to-end, not just unit tests, if it touches a runtime path.
- Archive. Move the entry to
ARCHIVE.md with a Completed: date and how it was verified; set its status.json entry to completed.
If you hit a wall (spec unclear, design needs revisiting), stop and flag it — flip status to needs-fix with a note, never leave it silently running.
