Your agent knows what your team already learned and decided.
- Maps your code
- Learns as you work
- Shares with your team
- Mark asks his Claude Code agent to fix the retry bug. Without a map, it opens file after file. With the graft map, it goes straight to client.go.
- It saves what it learned for the whole team: keep retries at 4.
- The next day, Jeff asks about the same code. His Codex agent starts from Mark's learning and saves about 27k tokens.
- Then Jeff tries to raise retries to 10. trail check flags it: Mark decided to keep retries at 4.
- Meanwhile, Sam starts on the same file. His Cursor agent sees that Jeff's branch also changes client.go.
- Every agent on the team adds to one trail.
Same agent. Less wandering.
60%less time
32%lower cost
33vs 27SWE-bench Verified issues solved, of 50
Same model on both sides, with and without the graft map. Time and cost over 162 runs. How we measured
A learning is a file.
Markdown in your repo, reviewed in the pull request like the code it is about.
--- title: Keep the default retry count at 4 author: Mark date: 2026-10-09 cost: { tokens: 27400 } touches: - http/client.go --- ## Decided - Keep defaultRetryMax = 4: ten retries with the default backoff can hold one Do() call for over an hour. ## Tried and ruled out - Raising the default to 10 for one customer: worst case, a single Do() call blocks for over an hour. ## Watch out - RetryWaitMax caps each wait at 30s, not the total: the waits add up with every retry.
Two commands. Then work as usual.
npm install -g @trailhq/trail-codetrail initClaude CodeCodexCursorKiroany agent that reads files
MIT licensedNo server, just filesNo API keyAnonymous usage stats, opt-out