---
title: "Trace Claude Code · Ataris docs"
description: "Run Claude Code through the ataris CLI on a laptop, in CI, or in your own sandbox."
canonical: "https://ataris.dev/docs/guides/claude-code"
last-updated: 2026-10-12
---
# Trace Claude Code

Run Claude Code through the ataris CLI on a laptop, in CI, or in your own sandbox.

This guide traces Claude Code with the `ataris` CLI. Everything Claude Code does, including its tool calls, file edits, test runs, and the processes and connections underneath them, lands on the session's timeline.

## Install

Install the CLI (see the [quickstart](/docs/quickstart#2-install-the-cli)) and set `ATARIS_API_KEY`. Claude Code runs with its own login or `ANTHROPIC_API_KEY`, as it normally does.

## Run Claude Code through the CLI

Put `ataris` in front of `claude`:

```bash
ataris --agent test-fixer claude -p "Fix the failing tests"
```

Interactive sessions work too: `ataris claude` opens Claude Code as usual, and the session lasts until you quit. Every flag after `claude` is passed to Claude Code unchanged.

Your own Claude Code settings keep working unchanged.

## What lands on the timeline

- Each tool call Claude Code makes, such as `Bash`, `Read`, `Edit`, `Write`, and `WebFetch`, with its result.
- The skills it loads and the MCP servers and tools it calls.
- Every process its commands start, every file they open, and every connection they make.
- The models it used, read from its transcript when the run ends.

## In CI or your own sandbox

Install the CLI in the job or image and run the same command. Use `-p` for a non-interactive run:

```bash
ataris --agent test-fixer claude --dangerously-skip-permissions -p "Fix the failing tests"
```

See [Trace an agent CLI](/docs/guides/cli#run-it-in-ci).

## Block risky tool calls

On Pro and above, a [policy](/docs/concepts/policies) can refuse a tool call before Claude Code runs it, such as `git push --force*` or a read of `~/.ssh/**`. Claude Code is told why the call was blocked, and the attempt is recorded on the timeline.