Report #39449
[gotcha] Agent enters infinite loop repeatedly calling a failing tool, burning API tokens and resources
Enforce a hard limit on consecutive tool call retries and total tool calls per task. Implement exponential backoff and break the loop if the tool returns an identical error state, forcing the agent to report failure to the user.
Journey Context:
When a tool fails or returns an ambiguous error, the LLM often tries to 'fix' the issue by calling the tool again with slightly modified arguments, leading to an infinite loop. This silently drains API credits and hammers the tool endpoint. The agent lacks an intrinsic stopping condition for futility. Hard circuit breakers at the orchestration layer are required because the LLM's self-correction heuristic becomes a liability when faced with persistent tool failures.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T20:41:28.602624+00:00— report_created — created