What problem does it solve? A plan that a low-judgement executor will follow literally can fail in ways nobody notices until execution breaks: untraced acceptance criteria, tasks without contracts, fake checks, broken dependency order, or a missing full-suite gate. This Skill provides a read-only audit round that catches those defects before sdlc-execute dispatches the plan's tasks. ## Core Features & Use Cases - Prose failure-mode audit: Applies the eleven shared prose failure modes loaded from the sdlc-audit-criteria skill, returning findings in four fixed evidence shapes. - Plan contract audit: Checks eleven plan-specific points including AC traceability, task contracts (Inputs/Outputs/Data mapping/Errors/Behavior), real check files with shell-safe Run commands, heading format, dependency order, and the full-suite gate. - Recorded audit trail: Registers each round via source_add as a SOURCE supporting plan.md, and returns findings as a JSON envelope to the dispatching main agent. - Use Case: After sdlc-plan produces plan.md and before sdlc-execute runs, dispatch this Skill to verify every spec AC is traced to a task and every task is executable, so undispatchable or incomplete plans are sent back for revision. ## Quick Start Dispatch a plan audit round on the current plan.md and return the findings JSON to the main agent.
