#816 · Primary category: AI Agents & Automation

meta-harness-tbench2-artifact

Meta-Harness: 76.4% on Terminal-Bench 2.0 (Claude Opus 4.6)

Project last updated:03/26/26

GitHub Stars

1.2K

Forks

165

Contributors

1

License

Other

Why we included this project

Meta-Harness is a small agent scaffold that shows a simple trick worth stealing: before the agent loop starts, it snapshots the sandbox environment, working directory, available tools and package managers, and injects that into the initial prompt. That saves the agent a handful of early exploration turns that would otherwise go to `ls` and `which python3`. The README includes a one-command way to run it against Terminal-Bench 2.0, where it scores 76.4% with Claude Opus 4.6. It builds on KRAFTON's Terminus-KIRA and Harbor's Terminus-2 framework, so treat it as a research artifact and a starting point rather than a finished product. Teams building CLI or shell agents can borrow the same idea to cut wasted tool calls.

Articles for this project

No articles for this project yet.

To suggest a topic or contribute an article, contact us.

Related projects in this category