What problem does it solve? AI coding sessions often drift from what was agreed, improvise scope, or silently work around a flawed plan, leaving no record of what actually happened. This Skill enforces disciplined implementation: it builds only what an approved plan specifies, ticks off steps as it goes, and records every deviation so later review has an honest baseline. ## Core Features & Use Cases - Plan-gated execution: Reads docs/work/<slug>/plan.md in full and refuses to build unless the plan status is approved, preventing unauthorized work. - Resumable step tracking: Ticks each step's checkbox in plan.md as it completes, so an interrupted session can resume exactly where it stopped. - Deviation and PLAN DEFECT logging: Appends deviations to Build notes and flags genuinely wrong plans under a PLAN DEFECT heading, which downstream review escalates. - Honest test reporting: Runs the repo's test suite and records the result as green, red, or none in the state file, never reporting success over failing tests. - Use Case: After approving a plan for a new API endpoint, run this step to have the change implemented exactly as specified, with each step checked off, two minor deviations documented, and the test suite result recorded for the review stage. ## Quick Start Run /quorum:2-build with the slug of an approved plan in docs/work to implement it step by step.
