What problem does it solve? Feature work with AI agents often drifts: vague requirements, skipped tests, batched commits, and self-verified code that passes review but fails in production. This Skill enforces a disciplined Specify → Design → Tasks → Execute workflow where requirements are testable, commits are atomic, and an independent Verifier (author ≠ verifier) validates the result with evidence before anything is called done. ## Core Features & Use Cases - Adaptive 4-phase workflow: Auto-sizes depth by complexity (Small/Medium/Large/Complex), writing EARS-notation requirements, atomic tasks, and Conventional Commits only as deep as the feature needs. - Deterministic validation scripts: Python gates (validate_spec.py, validate_tasks.py, check_commit.py, validate_state.py) enforce structural rules by code, not by model memory. - Independent Verifier with discrimination sensor: A fresh sub-agent re-derives test coverage from the spec, injects behavior-level mutations to confirm tests actually catch regressions, and writes an evidence-based validation report. - Project memory and lessons layer: STATE.md decision log and handoff snapshots enable pause/resume, while lessons.py distills verification failures into reusable project-local guidance. - Use Case: Ask the agent to "specify feature: user notification preferences" — it produces a testable spec, breaks it into atomic tasks, implements each with its own commit, then runs an independent verification pass before declaring completion. ## Quick Start Ask the agent to specify and implement a new feature, for example: "specify feature: add password reset flow, then implement it with verification."
