What problem does it solve? Executing an approved implementation plan by hand requires constant human supervision: checking off progress, running tests, committing each phase, and deciding what to do when reality diverges from the plan. This Skill removes the human from the loop entirely, running the full plan under Claude Code's /goal with explicit automatic policies for every decision point. ## Core Features & Use Cases - Autonomous phase execution: Reads context/changes/<change-id>/plan.md, treats the ## Progress section as the single source of truth, and works through every Automated checkbox without asking questions. - Fixed quality-gate stack: Each phase runs plan success criteria, a deliberate-break test check, repo-wide tests/lint/typecheck, and a Conventional-Commits commit only when all gates are green, with a 2-attempt self-fix budget. - Honest failure surfacing: Structural mismatches and repeated gate failures produce a STOP block and a run report listing adaptations, uncertainties, and pending Manual rows as a human checklist. - Use Case: Run headless with claude -p to implement all phases of a planned change overnight, then return to a transcript narrating every gate verdict, commit SHA, and remaining manual verification step. ## Quick Start Run /goal with a completion condition followed by /10x-goal-implement <change-id> to autonomously implement every phase of the plan in context/changes/<change-id>/plan.md.
