What problem does it solve? Executing an approved technical plan by hand requires constant human supervision: running tests, staging files, writing commits, and tracking progress. This Skill removes that bottleneck by implementing every phase of a plan.md file autonomously under a /goal session, with no human interaction at any point. ## Core Features & Use Cases - Autonomous phase execution: Delegates each phase's code changes to a subagent, then runs a fixed gate stack (plan success criteria, deliberate-break check, full test/lint/typecheck suite) before committing with Conventional Commits. - Progress as single source of truth: Flips Automated rows in the plan's ## Progress section, writes commit SHAs back into each row, and resumes from the first pending row after any stop. - Mismatch taxonomy and STOP protocol: Classifies plan-vs-codebase gaps as Minor (adapt) or Structural (halt with a structured STOP block), with a 2-attempt self-fix budget per failing gate. - Use Case: Point it at context/changes/<change-id>/plan.md in a headless run; it implements all phases, commits each on green, and ends with a run report listing pending Manual rows as a human checklist. ## Quick Start Run /goal with a completion condition followed by /10x-goal-implement <change-id> to autonomously implement all phases of the plan.
