What problem does it solve? Long coding sessions lose discipline: edits happen without reading files, tests run after implementation instead of before, and verification stops at "it ran". This Skill keeps a structured five-gate workflow (spec, evidence, adversarial review, layered verification, diff-shaped report) anchored in on-disk state so the process survives long contexts. ## Core Features & Use Cases - Five-gate task loop: Routes feature work to test-driven development, bugs to diagnosis, refactors to behavior-preserving fixes, and security changes to threat modeling, with gate state persisted in .work/GATES.md. - Red-green inner loop: Enforces fresh reads before every edit, failing tests before implementation, and full-suite runs after changes. - Habit scorecard: The code-score.py script measures real session logs for discipline habits (read-before-edit, red-first ordering, full-suite verification) and compares models or code-mode ON vs OFF sessions. - Use Case: Activate code mode before fixing a production bug; the Skill reproduces the failure first, delegates an adversarial review of the fix, verifies across test suite and running app, then produces a diff-shaped report separating verified from assumed claims. ## Quick Start Say "code mode" before starting a real coding task so the Skill creates its gate-state files and walks the change through spec, evidence, adversarial review, verification, and reporting.
