neosigmaai/auto-harness
A loop that runs an agent on a benchmark, reads the failed traces, has a coding agent edit the agent's code, checks the change against a regression suite, then repeats. It ships setups for tau-bench, Terminal-Bench and BIRD-Interact.
Not tested yetKey files read, not run
How we know
We read the README and key files, not every source file, and did not run it, so the most we can say is Idea adapted or Risky.
Good for
Researchers or builders who want to tune an agent overnight against a benchmark they already trust.
Not for
Anyone who wants a finished product. It is marked experimental, and the score gain it reports comes from its own README run.
Repo
- Stars
- 536
- License
- MITUse it, change it, sell it, closed-source included. Keep the author's copyright notice.
- Maturity
- experimental
- Last push
- 2026-07-08
- Last checked
- 2026-11-14