What problem does it solve? Teams with an approved implementation plan often struggle to execute it with true test-first discipline: tests get written after the code, phases lose track of progress, and non-testable work (scaffolding, config, infra) gets forced into awkward TDD loops. This Skill drives a plan from context/changes/<change-id>/plan.md to completion one phase at a time, writing a failing test before any production code, while keeping execution state in the plan's canonical Progress section. ## Core Features & Use Cases - Red-Green-Refactor Loop per Phase: For each eligible phase, writes one failing test (RED), implements the minimal code to pass it (GREEN), then refactors while staying green, with a focused budget of 2-5 tests per phase. - TDD Eligibility Gate: Before each phase, checks whether implementation is absent and whether the phase is test-first-able; redirects non-TDD'able phases (scaffolding, CI, styling) to /10x-implement and stops when code already exists. - Shared Progress Tracking: Mutates the same ## Progress section as /10x-implement, flipping checkboxes and writing back commit SHAs at phase end, so both skills can interleave on one plan. - Phase-End Commit Ritual: Runs the full test suite, gates on manual verification, stages files explicitly by path, and proposes Conventional-Commits messages with SHA write-back. - Use Case: Given an approved plan for an OAuth login feature, run the skill to TDD the token-validation phase, hand the Docker config phase off to /10x-implement, and resume test-first on the next phase without losing state. ## Quick Start Ask the AI to drive the approved plan for your change test-first by running /10x-tdd with your change-id, for example "/10x-tdd oauth-login phase 1".