agenttrail

Why we included this project

Anyone who has let a coding agent run for half an hour has wondered whether it is making real progress or quietly stuck. agenttrail runs entirely on your machine and shows you the answer: it watches what Claude Code, Codex, or Cursor actually does, the plans it declares, the tool calls it makes, the files it edits, and renders all of that as one zoomable map of your repo, so you can see which component the agent is touching and notice when it reopens work it already called done. Setup takes a single npx command in your repo, hooks are wired only with your consent, and there is no account or telemetry, so it is easy to try on an existing project without changing your workflow. The map shows real components with dependency arrows and done, working, and blocked states, which helps when you run several agents at once and need to catch one drifting from its declared plan. If you want to see what your agents actually do instead of trusting their final summary, this is a lightweight way to get that visibility.