Failure Index / Agent behavior · warning · OpenAI Agents SDK · LangChain

Guardrail triggered: agent run blocked mid-task

A guardrail (input/output validator) fired and stopped the run. Working as designed if the input was truly bad — but frequently the guardrail itself is miscalibrated and blocks legitimate traffic.

The error

openai agents guardrail tripwire triggered
agent input guardrail blocked
InputGuardrailTripwireTriggered

Root cause

Guardrail conditions written too broadly (regexes matching benign inputs, over-sensitive classifiers), or guardrails evaluated against the wrong field (raw transcript instead of the user message). Legitimate requests then die with an opaque 'tripwire' error.

The fix

Log WHAT the guardrail matched, not just that it fired. Replay the blocked input against the guardrail in isolation, tighten the condition to the actual threat pattern, and add an allowlist for known-good shapes it was catching.

Preventing it next time

Treat guardrails like code: version them, test them against a corpus of known-good inputs in CI, and monitor their block rate — a spike means miscalibration, not an attack.

Stop debugging this by hand. Vorlo watches your agent and, when this failure happens, hands you the diagnosis and the fix — verified by developers who hit it before you — in your dashboard, Slack, or your editor. Two lines of code: pip install vorlo-trace · npm install vorlo-trace. Start free

Related failures

Agent stopped due to iteration limit or time limitAgent stopped due to iteration limit or time limitAgent calls a tool that doesn't exist (hallucinated tool name)agent tool not found error