Part of the AI coding agent security checklist

Claude Code audit log tools compared: ranwhat, suvadu and cc-audit-log

Three free tools answer "which commands did Claude Code run?". They differ in one way that decides the choice: ranwhat and cc-audit-log read the transcripts already on your disk, so they work on sessions from before you installed anything. suvadu records through hooks, so it starts from the day you set it up and keeps more about each command from then on. This page is written by ranwhat's maker; every claim about the other two comes from their own pages, listed under Sources.

$ uvx ranwhat watch
Install

Checked against each tool's site, README and changelog under Sources, and ranwhat's source. We did not run suvadu or cc-audit-log. Published .

01 / Answer

The short answer

Something already went wrong ranwhat. It reads the history your agents already wrote, so it can show a session from last week with nothing installed beforehand. uvx ranwhat watch runs it once without installing.
A record from now on suvadu. Once its hooks are in, it keeps every command in its own database with the exit code, how long it took, the folder and the git branch, next to the commands you typed yourself.
A quick timeline cc-audit-log. One small script, run with npx, that prints what Claude Code did today. It reads Claude Code only, and its repository was archived in July 2026.
Both ranwhat and suvadu do not get in each other's way. One reads transcripts, the other writes its own database, and neither changes what the other sees.
02 / Compare

Side by side

What each tool's own pages say, as of 12 November 2026. "Not described" means we found no mention of it there.

ToolGets its record fromSessions before installAgentsLeaked keysCan block
ranwhat The history agents write to disk Yes 30 Finds them in transcripts and says what to rotate Opt-in hook, Claude Code only
suvadu Hooks it adds to each agent, into its own database No; a past session's conversation can be imported 6 with setup, 10 more tagged Masks secrets in the commands it stores Only if you wire its guard into a hook
cc-audit-log Claude Code's transcripts on disk Yes 1 Not described No
ToolLicencePriceRuns withLatest release
ranwhat MIT Free; Plus €12 a month per organisation Python 3.9+, uvx or pipx 0.10.0, November 2026
suvadu MIT Free A Rust binary from Homebrew or cargo; macOS and Linux, zsh or bash 0.5.3, November 2026
cc-audit-log MIT Free Node, npx 1.0.7, March 2026; archived
03 / suvadu

suvadu: a shell history that also records your agents

suvadu replaces your shell history with a local SQLite database and adds the commands AI agents run to it. For Claude Code, suv init claude-code writes hooks into ~/.claude/settings.json, and recording starts with the next session.

Does well Each command is stored with its exit code, duration, working directory and git branch, which a transcript does not always give you. Your own commands and the agent's sit in one searchable history, and an agent can query that history over suvadu's MCP server.
Keeps longer Claude Code deletes transcripts after 30 days by default. suvadu's database is its own, so what it recorded stays after the transcript is gone.
Agents Dedicated setup for Claude Code, Cursor, Codex, OpenCode, pi.dev and Antigravity. Ten more environments, among them Aider, Copilot and Windsurf, are recognised and tagged without their prompts.
Limits Installing it does not record anything until the hooks are added, and commands from before that are not there. For Claude Code its integration page describes capturing Bash calls, so a file the Read tool opened or an MCP call is outside it. It is for macOS and Linux.
Secrets It masks secret-shaped strings in command text before storing them. Its pages do not describe a report of which credentials an agent saw.
04 / cc-audit-log

cc-audit-log: a timeline from one script

cc-audit-log scans ~/.claude/projects/ and prints a readable timeline of files, commands, commits and subagents, for today, a date or the last few sessions.

Does well No install and no dependencies: npx cc-audit-log --today. It is one small file, so you can read all of it before you run it. It sends nothing anywhere.
Flags Seven kinds of command, among them rm -rf, a force push, a hard reset, npm publish and sudo.
Limits Claude Code only. Its README describes no secret detection. The last release was in March 2026 and the repository was archived by its owner on 2 July 2026, so a change in Claude Code's transcript format will not be followed.
05 / ranwhat

ranwhat: what it adds, and what it does not do

ranwhat reads the history Claude Code and twenty-nine other coding agents already write to disk. Nothing has to be installed before the session you want to look at.

What ran ranwhat watch runs sixteen rules over every tool call, not only shell commands: credential reads, recursive deletes, force pushes, package publishing, scripts downloaded and run unread, and more. The list is on the watch page.
What leaked ranwhat clean finds credentials sitting in plaintext in transcripts, says which to rotate first, and masks them only when you ask.
What a key allows ranwhat scan scores what a token can do from a catalogue of 204 scopes across twelve providers. Neither of the other two describes this.
Checkable ranwhat bench runs the detection over a made-up agent home and scores it against an answer key, and lists what ranwhat is known not to catch. Anyone can rerun it.
Limits It reports after the fact, and only what is still on disk: a Claude Code session older than cleanupPeriodDays is gone for ranwhat too, where a recorder with its own database would still have it. It does not add your own shell commands to the picture. The one thing it can block is a call you opt in to with its Claude Code hook.
06 / Choose

Which one to use

If youStart withWhy
Need to know what an agent did in a session that already happened ranwhat It reads what is on disk. A hook installed today recorded nothing last week.
Use Cursor, Codex or Gemini CLI as well as Claude Code ranwhat One pass over thirty agents' history, with no setup for each.
Want to know whether a key leaked into a transcript ranwhat It finds the credential, says what to rotate and what the token allows.
Want a lasting record with exit codes and timings, next to your own shell history suvadu Its database outlives the transcripts and holds detail they do not.
Want your agent to look up what ran earlier suvadu Its MCP server lets an agent query the history.
Want a one-line look at today in Claude Code cc-audit-log Nothing to install, and small enough to read first. Check that it still parses your version's transcripts.

Next

To read the record by hand, with no tool at all, see a Claude Code audit log from what is already on disk. For tools that stop a call before it runs, Stroq and HOL Guard among them, read the best AI coding agent security tools. Every other control is in the AI coding agent security checklist, and the rest of the guides are listed here.