The authors automatically curated roughly 3,000 difficult terminal tasks spanning more than 50 domains, including science, software, hardware and data.
Rewriting harness-assisted solutions improves general terminal-task performance
The authors convert successful solutions discovered under specialized agent harnesses into verified training trajectories executable under a general harness.
Independent
Zongxia Li · Yucheng Shi · Zhongzhi Li · Junyao Yang · Ruhan Wang · Chengsong Huang · +4 more
Research Digest··2 min read
Li and colleagues introduce Recursive Self-Rewrite, a pipeline for turning solutions found with specialized agent controllers into supervised training data compatible with a general-purpose controller.
Why this paper
Independent · Part of Agent Self-Improvement, now 32 papers
In one line
Rewriting successful multi-harness agent traces into verified general-harness trajectories substantially improves a base model on complex terminal tasks.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks (6 benchmarks)
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§