TBNAI
SkillScriptBench: Benchmarking Self-Evolution of Executable Agent Skill Packages Beyond Markdown
Publication: cs.AI updates on arXiv.org
Author: Yuxuan Liu, Haoran Li, Yuhao Zhang, Jiahe Guo, Hongyu Luo, Wenbin Hu, Huihao Jing, Kawai Chung, Junle Chen, Changxuan Fan, Qing Zong, Lingyun Xie, Yangqiu Song
Source published: 06/10/2026, 04:00:00 UTC
Briefing created: 06/10/2026, 05:16:22 UTC
Based on a feed excerpt; full article content has not been acquired. Source evidence retrieved 06/10/2026, 04:18:24 UTC.
Briefings
SkillScriptBench evaluates self-evolution of executable agent skills by distinguishing between documentation, script, and preservation repair tasks. Researchers selected 100 packages from 35,000 GitHub roots and created 150 repair tasks. A controlled track tested 200 tasks across four states: clean, documentation faults, script faults, and combined faults. AST-Guided Skill Revision improved repair success by 21.9% for Raw Package and 27.7% for CoEvoSkills, with 20.8% and 31.5% task completion rates respectively. The benchmark aims to separate skill evolution from Markdown-only revision.
Key points
- SkillScriptBench introduces a 350-task benchmark evaluating executable agent skills separately for documentation, script, and preservation repair.
- AST-Guided Skill Revision uses abstract syntax trees to link maintenance requirements to code locations, improving repair success by 21.9% for Raw Package and 27.7% for CoEvoSkills.
Context and limitations
The study covers 100 packages from 35,000 GitHub roots and 200 tasks from 50 packages, with a snapshot of 150 repair tasks and 4 LLM evaluations.
Generated by Bittrees-hosted qwen35-2b sha256:aaf42c8b7c3c on 06/10/2026, 05:16:22 UTC. Based on publisher-provided feed evidence; may not cover the full original article. Check the source for details.
View archived briefing on IPFS
bafybeiaj4zpfpq2hsqrw57vqy6m7rqhivwl2sqopitaq5pnwwokfvwlbua