What problem does it solve? AI coding agents frequently read a requirement and never turn it into an active obligation, then claim completion on top of it. This Skill freezes a small set of proof-backed obligations before any code is written and enforces verification by an independent agent, so unproven claims cannot pass as done. ## Core Features & Use Cases - Four-phase workflow: Plan (EARS acceptance criteria, flow, relations, surface, one-way doors), Checks (claims with concrete proofs and a coverage join), Build (tests written from checks, never from implementation), and Verify (a fresh sub-agent reviews the full diff). - Deterministic validation scripts: Python validators gate each phase, checking plan structure, check proofs, coverage completeness, Conventional Commits messages, and verification report integrity. - Profile-based verification depth: light, standard, and ui profiles declare how much verification runs, enforced by the completion gate script. - Use Case: A developer asks the agent to plan a lockfile migration feature. The Skill produces a human-reviewable plan with EARS criteria, derives proof-backed checks, builds against them, and dispatches an independent Verifier whose report must pass validate_verification.py before the work is declared done. ## Quick Start Ask the agent to plan a feature with pbs-spec-lean, for example by saying "pbs-spec-lean — plan the lockfile v2 migration", then review the generated plan before checks and code are written.
