What problem does it solve? AI coding sessions often produce plausible code with no written record of what was actually requested, making it impossible to verify the result against the original intent. This Skill forces intent, acceptance criteria, and non-goals to be written down in testable terms before any code is written, creating a fixed target the rest of the delivery pipeline can be measured against. ## Core Features & Use Cases - Branch and work-item resolution: Settles which work item a request belongs to, creates or reuses the appropriate git branch, and derives a kebab-case slug that names all downstream artifacts. - Structured plan authoring: Writes docs/work/<slug>/plan.md with verbatim prompt, intent, falsifiable acceptance criteria, non-goals, open questions, numbered verifiable claims, ordered steps, and optional mermaid diagrams. - Approval gate with state recording: Holds an explicit human approval checkpoint, fingerprints requirements via a requirements hash, and records pipeline state so later steps can prove the target never moved. - Use Case: A developer asks for a webhook retry feature. The Skill creates a feature branch, investigates the repository, writes a plan with observable acceptance criteria, and waits for explicit approval before any build step runs. ## Quick Start Ask the AI to plan a change, for example: run /quorum:1-plan with a description of what to build, then review the written plan and approve it.
