Automata from Agent Traces: Failure and Next-Step Prediction
- 类型:arxiv
- 标识:2608.23670
- 链接:https://arxiv.org/abs/2608.23670
- 主分类:agent
- 形态:application
- 被引:3
- 被引来源:Semantic Scholar
- S2被引:3
- OpenAlex被引:0
- 影响力被引:0
- TLDR:Behavioral topology is shaped more by the deployment harness than by the LLM, providing a model-agnostic structural primitive for safety auditing and runtime monitoring, and addresses both prediction goals.
- OpenAlex ID:W7204266910
- OpenAlex DOI:10.48550/arxiv.2608.23670
- DOI:10.48550/arxiv.2608.23670
- DOI来源:OpenAlex
- 开放获取:green
- 开放获取链接:https://arxiv.org/pdf/2608.23670
- OpenAlex更新:2026-09-01
- 待LLM分类:否
- 标题中文:来自 Agent 轨迹的自动机:失败与下一步预测
- TLDR中文:行为拓扑更多由部署 harness 决定而非 LLM 本身,为安全审计和运行时监控提供一种与模型无关的结构化 primitive,并同时满足两类预测目标。
- 来源文件:
- /inbox/tom/_candidates/2026-08-27-agent-rag-longcontext-candidates.json
- [S2 enrich]
- [OpenAlex backfill]