AgentGuard: Learning Execution Guardrails from Anomalous Coding-Agent Trajectories
Source
Wuyang Dai, Song Wang
Author
Wuyang Dai, Song Wang
Date
Terms in this piece · Glossary
guardrails — The checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
Why it matters
AgentGuard mines 642 documented coding-AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → failure traces to auto-generate instruction-level guardrailsThe checks around a model that block bad inputs and outputs — filters, validators, and permission rules the model itself can't override.Full definition → against unrelated file edits, rewritten tests and unsafe commands, activating only the rules relevant to the current task.
articleOpenAgentFlow: Enabling System-Wide Safety Boundaries for Heterogeneous AI Agent FleetsDongsheng Chen, Xiangyu Zhao, Xin Yao, Xuetao Wei
articleEngineering Reliable Coding Agents: Evaluating and Operating the System Around the ModelStephanie Jarmak
articleXAgent: eXecution-guided Agentic AI for Effective Localization and Resolution of GitHub IssuesHieu Huynh, Patanamon Thongtanunam, Michael Fu, Bach Le, Kla Tantithamthavorn