quietreviewer/riskdev

RiskDiff, sold under the name riskdev, is a command line tool that looks at the uncommitted changes in a Git repository and tells you which parts actually need a careful human review. It was built with AI coding assistants like Claude Code in mind: when an AI writes a large diff, most of it is usually safe to skim, but a small number of lines can be risky, and this tool tries to point out exactly which ones. When you run it, RiskDiff looks at the shape of the change first, checking whether things like authentication, permissions, payments, database migrations, or public interfaces were touched. It also runs your project's own test, lint, and type-check commands as supporting evidence, and compares what you said you wanted to do against what the change actually does, since AI tools sometimes solve a different problem than the one asked for. Every change gets sorted into one of three tiers: mandatory line by line review, a recommended focused look, or optional skimming, with reasons, a possible failure scenario, and an estimated number of minutes to review attached to each item. By default, if you give it an Anthropic API key, it sends your diff and some context to the Anthropic API for the AI analysis part. Without a key, or with the no-ai flag, it falls back to rule based checks using only the shape of the diff, and makes no network requests at all. The tool is explicit that it never claims code is safe. It only tries to help a human reviewer focus their limited attention where it matters most, and passing checks reduce uncertainty without removing it. RiskDiff needs Node.js 18 or newer and Git, and is installed with npm. Right now it only fully supports JavaScript and TypeScript projects, since it looks for test, lint, and type-check scripts in package.json. It fits developers who already review AI generated code and want a faster, more targeted way to decide what deserves close attention, rather than reading every line themselves.