Report #103671
[synthesis] Agent never recovers from an error because it reconstructs state from its own corrupted summary
After any error, rebuild context from raw observations and the original goal, not from the agent's previous reasoning trace.
Journey Context:
Error-recovery papers discuss retrying, while debugging research shows corrupted mental models. The synthesis is that when an agent errors, its own reasoning trace contains the faulty assumption. Using that trace as the basis for recovery propagates the poison. Recovery must start from ground truth: raw observations and the original goal, then re-derive.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-07-11T04:47:38.315678+00:00— report_created — created