agenttrail Catches AI Coding Agents Quietly Reopening Finished Work

A dependency-free Node tool turns any repo into a live zoomable map showing exactly what Claude Code, Codex, and Cursor are doing.

·
·
agenttrail Catches AI Coding Agents Quietly Reopening Finished WorkPRO
  • agenttrail is a local, MIT-licensed observability canvas for Claude Code, Codex, and Cursor sessions (repo).
  • It compares an agent's declared plan against observed file changes to catch stalls and silent regressions.
  • The daemon is one dependency-free Node file (~470 lines) binding to 127.0.0.1 with no telemetry.
  • Install by running npx agenttrail --open inside any repo, no account required.
  • Claude Code gets full run cards via local hooks; Codex and Cursor use file-watcher based views.
  • Uses a small PLAN.md convention with components, dependencies, and file globs maintained by the agents themselves.

If you have left a coding agent running for half an hour and returned wondering whether it made real progress or quietly reopened work it already marked done, agenttrail is a small open-source tool built for exactly that situation. It runs locally, watches AI coding agents in real time, and renders what they are doing onto a zoomable canvas of your repo.

It turns plans, tool calls, file changes, and progress from Claude Code, OpenAI Codex, Cursor, or any agent that edits files into a live project map. One command opens it on localhost, giving you a live view of what your agents are actually touching versus what they said they would touch.

agenttrail live demo showing a Claude session, plan cards, and streaming file edits

Declared intent versus observed behavior

agenttrail runs on two signals. A plan says what the agent intends to do. The filesystem says what it actually touched. When those signals disagree, that is where you look.

  • Declared: The component and task the agent says it is working on
  • Observed: The files it is changing right now, including revisions to finished work

A card marked done lights up again if the agent quietly reopens it. That reopening behavior is one of the more common failure modes in long-running sessions.

Pro article

This story is for Pro members

You've reached the end of the free preview. Upgrade to AlphaSignal Pro to read the full article - and everything else behind the paywall.

Trending
  • No trending articles

Comments

avatar

Next Reads