implement

Create a TDD plan first, then invoke this Skill or reply "go" in the same session. It resolves the plan file, confirms a green test baseline, and implements each item in Red-Green-Refactor cycles with recorded command evidence.

Delegate batches that need several knowledge-catalog reads, produce long test output, or start a long plan, since isolation keeps the parent context lean. Small, clear batches in one area run directly in the parent context.

Yes, the Skill is runtime-neutral and works in Claude Code and Codex. It resolves sibling skills relative to its own directory and never depends on runtime-specific named agents.

A failing run triggers exactly one fix pass scoped to making the failing command pass. If it fails again, the Skill stops, leaves items unchecked, and reports the failing state and output so the user decides between reverting and debugging.

Do not use it for work without testable application behavior such as docs, configuration, CI, or housekeeping; use work-plan and work-implement instead. Planning-only requests belong to tdd-plan, and code review belongs to local-review.

No, it never commits, pushes, stashes, or alters Git refs or staging. When all items are done it suggests committing via the git-commit skill and leaves the decision to the user.