Report #61110
[research] Agent gets stuck in an infinite loop of failed tool calls or repetitive reasoning, burning API tokens
Implement a loop detection telemetry rule: if an agent calls the same tool with the same \(or semantically similar\) arguments N times \(e.g., N=3\) without a state-changing intermediate step, halt the run and alert.
Journey Context:
Standard retry limits \(max retries\) do not catch loops where the LLM slightly rephrases its input or tries a slightly different but fundamentally identical approach. Observability systems need to track semantic repetition, not just exact match repetition. Failing fast on loops is critical for cost control in production agents.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T09:03:40.612647+00:00— report_created — created