TLDR is the best test runner for Claude Code

A couple years ago, Aaron and I had an idea for a satirical test runner that enforced fast feedback by giving up on running your tests after 1.8 seconds. It's called TLDR.

I kept pulling on the thread until TLDR could stand as a viable non-satirical test runner and a legitimate Minitest alternative. Its 1.0 release sported a robust CLI, configurable (and disable-able) timeouts, and a compatibility mode that makes TLDR a drop-in replacement for Minitest in most projects.

Anyway, as I got started working with Claude Code and learned about how hooks work, I realized that a test runner with a built-in concept of a timeout was suddenly a very appealing proposition. To make TLDR a great companion to agentic workflows, I put some work into a new release this weekend that allows you to do this:

tldr --timeout 0.1 --exit-0-on-timeout --exit-2-on-failure

The above command does several interesting things:

• Runs as many tests in random order and in parallel as it can in 100ms
• If some tests don't run inside 100ms, TLDR will exit cleanly (normally a timeout fails with exit code 3)
• If a test fails, the command fails with status code 2 (normally, failures exit code 1)

These three flags add up to a really interesting combination when you configure them as a Claude Code hook:

• A short timeout means you can add TLDR to run as an after-write hook for Claude Code without slowing you or Claude down very much
• By exiting with code 0 on a timeout, Claude Code will happily proceed so long as no tests fail. Because Claude Code tends to edit a lot of files relatively quickly, the hook will trigger many randomized test runs as Claude works—uncovering any broken tests reasonably quickly
• By exiting code 2 on test failures, Claude will—according to the docs—block Claude from proceeding until the tests are fixed

Here's an example Claude Code configuration you can drop into any project that uses TLDR. My .claude/settings.json file on todo_or_die looks like this:

{
  "hooks": {
    "PostToolUse": [
      {
        "matcher": "Edit|MultiEdit|Write",
        "hooks": [
          {
            "type": "command",
            "command": "bundle exec tldr --timeout 0.1 --exit-0-on-timeout --exit-2-on-failure"
          }
        ]
      }
    ]
  }
}

If you maintain a linter or a test runner, you might want to consider exposing configuration for timeouts and exit codes in a similar way. I suspect demand for hook-aware CLI tools will become commonplace soon.