GitHub - viktor-com/tracequest: Trace Intelligence for Claude Code, Codex, Cursor, OpenCode, Grok and Droid sessions.

Follow your coding agents. Find what happened.

A local web app and CLI for the sessions your coding agents already write.

_{Claude Code · Codex · Cursor · Droid · OpenCode · Grok · vktr · fx · DeepSeek Harness
MIT · runs on your machine · no telemetry · no account}

What it does

Agentic analysis

Ask your coding agent why runs in a repo keep failing. With TraceQuest it searches every agent's past sessions, finds the cause (sometimes another run already worked it out) and writes the fix into AGENTS.md, so the next run starts with it.

tracequest search "No module named"

One index

TraceQuest puts every coding agent's sessions in one index: Claude Code, Codex, Cursor, OpenCode, Grok and Droid, from this machine, your other machines over SSH, and Cursor Cloud. You can search it, and so can your agents.

tracequest find "errors:>0"
tracequest import ssh gpu

Insights

It also keeps the numbers: which failures come up most, error rates per agent, retries and loops, active time and estimated spend. Narrow it to a project, an agent, a machine or a time range. It's all computed on your machine, with no model calls.

tracequest insights --refresh

The app

There's an app too. Watch a run while the agent is still working, send follow-ups to runs you started from it, start a new run in any folder, or continue an earlier session. Run npm start and open localhost:7777; starting and continuing runs needs tmux.

What falls out

Some things come for free once the sessions are indexed. Yesterday's standup, written from the sessions by the agent you already use. A handoff that gives a new run the relevant parts of earlier sessions, with the commands to open each one.

tracequest standup --yesterday
tracequest handoff "No module"

Also in the box

• Compare two sessions side by side
• Export a session as one HTML report or Markdown, or share it to a Gist or Hugging Face
• A read-only hub that collects sessions from your machines over SSH

_{The clips run on sample sessions written for them; in the analysis and standup clips the agent's replies are acted out. Costs are estimated from token counts and model prices, not billing data.}

Try it in a minute

TraceQuest is open source (MIT), runs on your machine and sends no telemetry. You need Node.js 22.5+ and git. Paste this into your coding agent, in the project you want to look into:

Add the TraceQuest skill to this project and use it.
1. Clone https://github.com/viktor-com/tracequest.git next to this project (../tracequest).
2. Copy ../tracequest/skills/tracequest into this project's skills folder
   (.claude/skills/tracequest for Claude Code, or your agent's own skills folder).
3. Follow that SKILL.md: install TraceQuest, then find out what keeps failing in this
   project's past agent sessions and propose a fix for AGENTS.md.

To browse it yourself instead, run npm run demo in the clone and open localhost:7777 for two bundled sample sessions, or npm start for your own.

Documentation · Contributing · Changelog · MIT