Report #75331
[synthesis] Agent outputs degrade silently when external API responses change format slightly
Implement strict schema validation on tool outputs using JSON Schema before passing data to the LLM, and log schema deviations separately from agent execution errors.
Journey Context:
Monitoring typically focuses on HTTP status codes or agent exceptions. When an API adds a new nested object or changes a field name, the request succeeds \(200 OK\), but the agent's subsequent reasoning uses the wrong data path. The agent produces a plausible but factually wrong response. Treating tool outputs as untrusted, volatile data—validating them structurally before LLM consumption—is essential to catch the drift between the API contract and the agent's expectations.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T09:02:30.926400+00:00— report_created — created