what is agent-audit fr?

AgentAudit is a command line tool that reviews the actions taken by an AI coding assistant during a session, checking for anything that looks risky from a security standpoint. When you use an AI agent like Claude Code to help write software, that agent reads files, edits files, runs shell commands, and sometimes reaches out to external services. AgentAudit looks back over a recorded session and checks each of those actions against a set of rules to flag things like touching a sensitive file, running a dangerous command, or making an unexpected outside connection. The README notes this is an early prototype and things may change. Importantly, the tool does not use any AI itself to do this checking, and it makes no network calls. It works entirely by comparing the recorded actions against pattern rules, using simple text matching or regular expressions, which makes its results predictable and repeatable every time you run it. Right now it only knows how to read session files from Claude Code, though the design allows for supporting other AI coding tools later. You install it by cloning the repository and running a couple of make commands, which builds the tool and links it into your local command path. Once installed, you can run it to analyze your most recent AI coding session, list all the sessions available on your machine, or check which rules and settings are currently active. Rules can be turned off, adjusted in severity, or added on top of the built in ones, either globally for your whole computer or scoped to a single project using a simple JSON configuration file. This is aimed at developers and teams who want a lightweight, automated way to double check what an AI coding assistant actually did during a session before trusting that work.