Audit log and secret scanner for AI coding agents: see what Claude Code, Codex, Gemini CLI and 27 more ran on your machine, and find API keys leaked into their transcripts. Local, MIT.
ranwhat checks whether any of your secrets have leaked into your coding agents' history, and where. It also checks whether your agent ran anything sketchy: read your SSH key, force-pushed, piped a script from the internet into a shell, published a package.
It reads the history that Claude Code and twenty-nine other coding agents already keep on your disk, including Codex, Gemini CLI, Cursor and GitHub Copilot (full list). It's open source (MIT), runs locally and has no dependencies. No account needed, no telemetry.
Use it as an audit log for your agents after the fact: see what commands Claude Code ran on your machine, and which API keys and passwords ended up in its transcripts.
Try it once without installing anything:
uvx ranwhat check
To keep it, install with pip install ranwhat, uv tool install ranwhat or
brew install MatijaMiki/ranwhat/ranwhat. Needs Python 3.9 or later.
ranwhat check
It changes none of your files. It lists the risky things your agents ran in
the last 30 days, and the secrets it finds in those transcripts and where
each one was found (--days 365 looks further back):
ranwhat watch · local agent flight recorder
...
* Credential material accessed 2026-09-30 18:13:00 Bash 9c41e07b2d5a8f13
cat ~/.ssh/id_rsa
-> The agent read a file whose only purpose is to hold secrets. Whatever
it read is now in a model context you do not control.
* Destructive git operation 2026-09-15 17:30:00 Bash e181acd29cf199ab
git push --force origin main
-> History rewriting or branch deletion. This is the class of action
that destroys the record of what else happened.
ranwhat clean · secrets sitting in local transcripts
...
* Stripe live secret key sk_…dc 32 chars seen 8x
read from api/.env
in /Users/you/Desktop/app
Then ranwhat clean walks you through the secrets it found. You can mask
the ones in plain transcript files, see what to rotate, and open each
provider's page to do it. Agents that keep a database, like Cursor, are only
read, and it tells you how to remove the secret there. Masking stops a key
leaking a second time but doesn't un-leak it, so rotate it.
If you want Claude Code to ask you before a high-risk command runs, there's
an opt-in hook: ranwhat hook install.
To keep a record of what watch flagged that shows if it was changed later,
ranwhat ledger record appends it to a hash chain on your machine, and
ranwhat ledger verify names the first line that no longer fits.
Everything that runs on your machine is free and stays free. Plus is €12 a
month per organisation, for the things that need a server: findings from
all your machines and CI runners in one place (ranwhat push), CI tokens
and teammates on your account, and a catalogue of what each token scope
allows, kept up to date. See pricing.
ranwhat check in CI after the steps where an agent ran; it can fail
the job on findings and uploads SARIFMIT licence. Built by cenner.