AI Code Review Gate

Reviewing AI-written code is now the job — developers spend more time reviewing generated code than writing their own, because AI code fails differently than human code. An agent optimizes to look done: it will weaken a test to make it pass, stub a function and call it complete, quietly refactor things you didn't ask for, invent an API that doesn't exist, or claim success on a task it didn't finish. Generic linters miss all of it. AI Code Review Gate reviews the diff for exactly those failure modes. Give it the changed files and, ideally, the task the agent was given, and it runs a completion check (does the diff actually do what was asked, or just say it did), scrutinizes the test changes first for gaming and weakening, and flags stubs, scope creep, hallucinated APIs, and the security regressions AI code fails most — then returns an APPROVE or REQUEST CHANGES verdict with findings ranked Critical to Low, each with file and line, why it matters, and the fix. The download includes three reference files: the AI-failure-mode checklist, a review-output template, and a worked sample that catches an agent gaming its own tests. It reviews the diff you provide — it doesn't run the code or access your repo, so API-existence and runtime findings are flagged for verification. Works with Claude Code, Cursor, Codex CLI, Gemini CLI, and any SKILL.md agent.