execute-tasklisted

# Execute Task Autonomous end-to-end implementation using the coordinator-subagent pattern. > Part of the task planning workflow. See [references/workflow-overview.md](references/workflow-overview.md) for the full pipeline. ## Input Task ID (numeric, e.g. `042`). ## Startup 1. Find task folder matching ID under `docs/tasks/` (glob `docs/tasks/$ID-*`) 2. Read ALL files: `requirements.md`, `solution.md`, `context.md`, `plan.md` 3. Verify prerequisites: - All 4 docs exist - No unanswered Open Questions - Every AC is covered by plan steps 4. Create isolated worktree with branch `task/$ID-$NAME` 5. Parse `plan.md` for phases, steps, and the Execution Sequence DAG 6. Find first unchecked phase (supports session resumption) ## Coordinator-Subagent Pattern Dispatch one subagent per phase. No nesting beyond two levels. ### What each subagent receives - Absolute worktree path (critical -- subagents do not inherit coordinator cwd) - Task folder path (absolute) - Phase number and name to execute - Instructions: - Read `plan.md`, `context.md`, `solution.md` before starting - Identify and use relevant skills before coding - Only touch files listed in plan.md -- don't "improve" adjacent code or refactor things that aren't broken - Match existing code style, even if you'd do it differently - State assumptions before coding. If the plan step is ambiguous, surface the ambiguity back to the coordinator rather than guessing - Fix routine failures (test bugs, type e