Unverified50% confidenceFactExact time
The research team constructed an auditing mechanism to identify silent failures in agent–tool interactions, and selected 15 scientific tools from the ToolUniverse environment as audit targets.
1
Sources
50%
Confidence
Long-term
Relevance
9/24/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedThe project plans to give agents access to scientific information and tools, enabling them to form hypotheses, test predictions, learn from mistakes, and pass on discoveries.70% similarUnverifiedCo-Scientist系统内部包含多个专职智能体，分别负责文献检索、假设生成、批判性评估和实验方案设计，通过协作辩论方式对科学假设进行迭代优化67% similarUnverified工具动态召回机制先从大规模工具库中通过向量检索筛选最相关的候选工具，再交由模型最终决策，以缓解工具描述质量参差带来的准确率衰减65% similarUnverifiedThe test initiators explicitly stated that they value failure cases the most, and hope the community will provide feedback on which information is ambiguous, what hinders tool selection, and what information agents looked for but could not find.65% similarUnverifiedThis Skill has built-in auxiliary scripts for scanning, scoring, and failure analysis, used to grade existing selector strategies and identify high-risk fragile selectors.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/946013API
curl https://kongchang.com/api/v1/knowledge/claims/946013MCP
get_claim(id=946013)