adversarial / code-pr

What problem does it solve?

Prevents harmful or low-quality AI-generated code changes from being approved and merged by running targeted adversarial critics over the proposed diff and its associated tests.

Core Features & Use Cases

• Multi-critic risk evaluation for code PRs: Runs five independent critics (structure, AI failure modes, test quality, dependency slopsquatting, and data-leak exposure) over the action’s plan/draft/diff facts.
• Risk-calibrated review depth: Adjusts how many times critics run and whether per-test inspection is exhaustive based on a structural risk score.
• Deterministic ship/hold decisioning: Produces a findings list with explicit severities (blocker/concern/note) and decides whether code is shippable or held for rework.

Quick Start

Ask an AI to perform an adversarial review of the proposed code pull request for the GaryOS code-pr playbook and return ship/hold findings with blockers identified.