Writing plans
Save to docs/plans/YYYY-MM-DD-<feature-name>.md unless the user named another path. Named acceptance criteria, or write them. Follow an architecture decision already in the repo. Do not invent docs/adr/ or a triage ceremony to satisfy this skill.
Stay read-only until the plan is written. Independent subsystems get separate plans.
Order. Draw the dependency graph first. Implement bottom-up so later tasks only use things that already exist.
Slice vertically. One task is one working path (schema + API + UI for that path), not "all of the database, then all of the API."
Map files before tasks: create / modify / test, each with a path. Follow the repo's existing file shapes.
Trace the spec.
| Requested behavior | Task | Verification |
|---|---|---|
/login is the real entry |
Tasks 1-2 | Playwright login |
For a user-facing app, list the real routes. No marketing-page stand-ins.
Each task:
- Files: exact paths
- Acceptance criteria: testable bullets
- Dependencies: earlier task numbers, or none
- Steps: the actual code or command, including the test and the expected result
Split a task that touches two subsystems, needs more than one session, or has an "and" in the title.
No placeholders: no TBD, "add validation", "similar to Task N", or "implement the feature." Repeat the code if a later task needs it.
Checkpoints after every 2-3 tasks (tests, build, the user-visible path). End with risks and open questions, or "None."