What problem does it solve?
Codebases are full of footguns: swappable same-type parameters, unbounded deletes, swallowed errors, and non-idempotent retries that let ordinary mistakes become production defects. This Skill audits code for these hazards and redesigns interfaces so the mistakes cannot be expressed at all, rather than relying on documentation or reviewer vigilance.
Core Features & Use Cases
- Hazard Scanning: A bundled Python script (stdlib only) scans Python, TypeScript, Go, Rust, and SQL for 25+ hazard patterns across contact, fixed-value, and motion-step lenses, with diff, staged, and CI-friendly JSON modes.
- Design Guidance: Applies parse-don't-validate, discriminated unions, branded types, typestate, and required idempotency keys so invalid states become compile errors.
- Ranked Audit Output: Every finding is classified by what happens when the mistake occurs (Control, Warning, Detection) and named with the specific device that closes it.
- Use Case: Before shipping a payments endpoint, run the scanner on the diff, then apply the reference guides to require an idempotency key backed by a unique constraint, so a retried charge cannot double-bill a customer.
Quick Start
Ask the AI to poka-yoke your current diff or audit a module for footguns, for example: review this pull request for ways callers could misuse it and rank each finding by severity.