Report #80546
[synthesis] Agent loops infinitely retrying a failing API call without altering parameters
Implement a retry budget that tracks tool call arguments. If a tool call fails twice with identical arguments, halt the agent and prompt for parameter mutation or human intervention, rather than allowing a third identical attempt.
Journey Context:
When an external API returns a 4xx/5xx error, the agent interprets it as a transient failure and retries the exact same tool call. Because the tool output contains Error, the agent doesn't update its internal state. Standard retry logic \(exponential backoff\) makes this worse by delaying the inevitable. The synthesis is that LLMs lack an internal insanity check—they cannot inherently recognize that doing the same thing twice and failing means they should change their approach. The environment must enforce argument mutation, because the agent's reasoning will always default to try harder rather than try differently.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T17:47:55.219419+00:00— report_created — created