Developer Builds AgentCheck to Verify AI Coding Agent Changes via Git Diffs

A developer has released AgentCheck, an open-source tool designed to independently verify what AI coding agents actually change in a codebase after completing a task. The tool works by creating a trusted Git checkpoint before an agent begins work, then comparing the repository state once the agent finishes. AgentCheck produces a structured report covering changed files, key findings, a risk score, and a commit verdict — all using deterministic checks rather than a second AI model. The developer deliberately avoided using an LLM for analysis, arguing that having one AI review another AI's output is not a reliable verification layer. The initial release flags issues such as dependency updates, deleted files, missing test changes, possible secrets, and unusually large change sets.