[ CLINIC ] // D6 · RECOVERY & ERROR HANDLING
Does your AI agent spiral when something breaks?
Real production throws 500s, timeouts, and contradictory instructions. Brittle agents retry the same failing call forever, apologize and abandon, or hallucinate success after a failure. Recovery is testable: put a failure in front of the agent and grade what it does next.
Part of the 6-dimension, 18-test battery · risk profile: UNCOMMON
What goes wrong
Retries the same failing call forever
apologizes and abandons
hallucinates success after failure.
// TEST YOUR AGENT
Three probes to run before you trust it
1Repeated tool 500s — posed as a scenario, or read from your samples
2contradictory-instruction gauntlet
3degrade-vs-spiral observation.
How the Clinic fixes it
Recovery playbooks (classify error → strategy)
circuit breakers
escalation-to-human paths.
[ FAQ ] // ANSWERS
Why does my AI agent break when a tool fails?
No recovery playbook, no circuit breaker, no escalation path. Fault-classification and escalation-to-human fix it; a live scan describes a failing tool and a malformed request to the agent — nothing is injected into its real tools — and grades whether it degrades or spirals.
Get recovery graded on YOUR agent
The free scan runs the real battery against your agent and returns an honest A–F grade with the failure modes named. In 24 hours you go from guessing to knowing.