A systematic study of observability in agent-generated systems reveals a consistent gap between diagnostic semantics at the source level and fault signals at runtime, suggesting that current evaluations focusing primarily on functional correctness may overlook observability as an important dimension of practical softwa...
This work shows that integrating structured abstractions with LLM-guided reasoning enables effective, reliable, and interpretable diagnosis for agent failures, and introduces the concept of neural invariants to specify agent behavior properties.
Jia-Yi Bi, Yanjie Gao, Yuan-Min Xie et al.· 0 citations
A differential analysis framework that attributes a failure or regression to a skill by comparing a target skill-guided run against a no-skill or semantically matched skill reference run that solves the same task, or solves it more cheaply.
Gen Dong, Yanjie Gao, Li-Qun Li et al.· 8 citations· ⚡1
This work introduces ToolAtlas, a graph-based framework that builds a persistent provider-side tool memory of tool capabilities, failure boundaries, and cross-tool compositions through execution-verified probing and establishes provider-side tool memory as an effective and reusable paradigm for tool servers.
Yue Fang, Zhibang Yang, Fangkai Yang et al.· arXiv.org· 0 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.
