wtf is agent-wiretap?

agent-wiretap is a local tool that lets you watch exactly what your AI coding agent sends over the internet while it works. If you have given an agent like Claude Code or Codex access to your files and an API key, this tool shows you, live in your terminal, every file it reads or writes, every shell command it runs, every website it talks to, and any secrets like API keys or passwords that show up in that traffic. Nothing captured is sent anywhere else. Everything stays on your own machine, and none of the traffic itself is written to disk, only decoded in memory and shown to you. You run it with a single command using npx and no separate install step, requiring Node version 22 or newer. You can either wrap your agent so it runs normally with its usual interface while agent-wiretap watches quietly in the background and prints a summary when it finishes, or you can run it in a mode that takes over your terminal to show a live dashboard while your agent runs in a second terminal window. Under the hood, the tool works by generating a local security certificate and starting a proxy on your machine, then setting standard environment variables that tell your coding agent to route its internet traffic through that proxy and trust the certificate. This lets the tool decode encrypted web traffic just long enough to inspect it before passing every request through to its real destination completely unmodified. Nothing is blocked or delayed. The dashboard also flags common secret patterns such as AWS keys, private key files, GitHub and Slack tokens, and sensitive filenames like .env files, and it estimates how much your agent's usage would cost at standard pay as you go API pricing, though this is only an estimate and does not actually charge you anything. The project is written in TypeScript, includes a test suite, and is released under the MIT license.