task-generator

# Task Generator ## Trigger Use when a feature spec needs a PR-sized plan with waves. ## Inputs - Feature description and source docs. - Optional repo context from `repo-context-map`. - Validation rules from wave/task validators. ## Outputs - `docs/agent-plans/<feature>/plan.md` - `docs/agent-plans/<feature>/plan.json` - `docs/agent-plans/<feature>/tasks.csv` - `docs/agent-plans/<feature>/tasks/<TASK-ID>.md` ## Procedure 1. Parse feature and determine epics. 2. Create tasks with required fields. 3. Assign conservative wave numbers and dependency edges. 4. Produce `plan.json` including `feature`, `source_documents`, `assumptions`, `open_questions`, `epics`, `tasks`, `waves`. 5. Emit task markdown files with exact objective and acceptance criteria. 6. Export `tasks.csv`. ## Required task schema Each task includes: `id`, `epic_id`, `wave`, `title`, `branch`, `objective`, `non_goals`, `context_to_load`, `read_set`, `write_set`, `dependencies`, `parallel_conflicts`, `implementation_steps`, `tests_to_write_first`, `verification_commands`, `acceptance_criteria`, `review_focus`, `rollback_notes`. ## Guardrails - Reject unrelated behavior. - Keep tasks PR-sized and independently reviewable. - Enforce same-wave dependency and write-set safety. ## Success criteria - Tasks are concrete and actionable with explicit read/write sets and post-wave verification. ## Prefer Superpowers - Use `writing-plans`, `writing-skills`, and `planning` style workflows from Superpowers for decomposition. ## Avoid scope creep - Include implementation guidance, but do not prescribe exact code unless required. Give enough detail for a PR-sized implementor to proceed without re-planning. - Do not assign changes outside the requested feature.