See what your agent actually did.

Agent Debugger opens the session log your coding agent left on your computer — the GitHub Copilot CLI, Copilot Chat in VS Code, Claude Code or Codex — and lays the work out step by step: the files it opened, the commands it ran, how its context filled up, and where it went round in circles.

It names seven kinds of wasted effort: a file opened again with nothing changed, a whole data file pulled in for one value, a script read to work out how to run it, a call that failed and failed again, a tool result so large that every later model call carries it, a run of searches with no result opened, and the same skill loaded twice. Each one says which steps were involved, what they cost in tokens, whether that count is measured or estimated, and a fix to try. The log is parsed by code in this browser tab and is never sent anywhere; How it works has the detail, including what each agent's log does and doesn't record.