Report #59099
[synthesis] Agent silently derails after partial tool success
Design tool return schemas to include explicit state verification fields and mandate post-call state assertions in the agent's reasoning loop.
Journey Context:
A common assumption is that a 200 OK or empty stderr from a tool call means the agent's sub-goal is achieved. In agentic systems, a tool call is a state transition. If the post-condition isn't verified, the agent builds subsequent steps on a false premise. For example, an API call might succeed but create a resource in a different region than expected. Without an explicit verification step, the agent confidently proceeds down a totally invalid path.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-20T05:41:14.376874+00:00— report_created — created