mde-task-lifecycle

Invoke the lifecycle with a task ID, such as "work on task 17A". The Skill loads the existing prompt, runs research and implementation phases, routes testing through Vitest and Playwright, then updates todo.md, CHANGELOG.md, and the prompt status before committing.

Say "process backlog" or "next P0" and the Skill reads tasks/prompts/INDEX.md, picks the next unstarted P0 task, and runs all five phases in order. Each phase hands off to specialist skills for planning, execution, testing, and shipping.

Do not use it for writing new PRDs (use mde-writing-plans), debugging arbitrary bugs without a task ID (use systematic-debugging), authoring new skills (use mde-prompting), or one-off PRs and commits unrelated to a tracked task.

Planning can be skipped if a prompt already exists, research if the wiring plan is concrete and the change is trivial, and implementation for docs-only tasks. Testing is never skipped on code changes, and shipping is never skipped because it is the only durable proof of completion.

A Phase 4 test failure loops work back to Phase 3 implementation until the build is green; the lifecycle never advances past a failing gate. If research uncovers a contradiction, the flow loops back to Phase 1 for replanning.

No. Commits are always created during the shipping phase, but pushes happen only on explicit user request, per the repository's git protocol. Force-pushes to main are never allowed.