debugging-tool-loops

To stop an agent stuck in a repetitive tool call loop, diagnose the session's tool call patterns to confirm the loop signature, then add explicit termination criteria to the agent prompt and configure a terminal tool injection for a graceful exit.

An LLM agent exhausts step or token limits without generating a final output because it lacks explicit termination criteria in its prompt, causing it to repeatedly invoke search or query tools without finding a clear exit path.

To add explicit termination criteria for workflow debugging, you update the agent prompt with defined stop conditions and inject terminal tools like `outputTo` and `complete` to enforce a graceful agent exit when conditions are met.

The best way to debug an agent making repeated search queries during GitHub issue triage is to analyze tool call patterns to identify the loop signature, then adjust the prompt with explicit stop conditions and terminal tool injection to produce a final report.

Yes, you can configure terminal tool injection to enforce a graceful agent exit by setting up `outputTo` and `complete` tools, which provide a clear exit path and prevent the agent from making open-ended, repetitive tool calls.