Same Wall, Different Angle

An agent calling an external API received a validation error on a malformed request. Rather than reading the error message, it retried the same malformed request nine times with minor, unrelated variations before stopping.

A transient network issue — considered and ruled out; the response was a consistent 400-level validation error, not a timeout or connection failure, and it was identical on every attempt.

Missing the tool needed to inspect the error — considered and ruled out; the agent had access to read the full response body and simply didn't use it before retrying.

Perseveration (the inappropriate repetition of a response after it has stopped being useful or correct) — most consistent with the trace: the agent's retry logic triggered on any non-success response without first reading what the response body said was wrong, and kept firing past the point where repetition could plausibly help.

Perseveration: repeated retries of a failing action without diagnosing or adjusting the underlying cause.

Perseveration is the inappropriate repetition of a word, gesture, or action well past the point where it's still relevant or useful. It's a well-documented sign of frontal lobe dysfunction, and also appears in some forms of dementia and obsessive-compulsive disorder.

This is a low-stakes case, but it's an extremely common one, and it previews how the same pattern becomes expensive at larger scale — repeated charged API calls, rate-limit bans, or, in worse cases, repeated real-world actions like sending duplicate messages. The safeguard is a retry policy that requires the agent to read and act on the specific error before trying again, and a hard cap on identical retries.