Report #47165
[architecture] Downstream agent crashes or hallucinates due to unexpected payload structure from upstream agent
Enforce strict JSON Schema contracts at agent boundaries using structured outputs \(e.g., OpenAI function calling or tool schemas\), and reject payloads that fail validation before passing context to the next agent.
Journey Context:
Developers often treat LLM outputs as unstructured text and rely on regex or prompt begging \('please output JSON'\) for inter-agent communication. This fails because LLMs inevitably drift or hallucinate formatting. By enforcing a strict schema contract, the orchestrator can catch formatting errors immediately and retry the upstream agent rather than passing garbage to the downstream agent, which would compound the error. The tradeoff is slightly higher latency due to schema validation and potential retries, but it guarantees type safety across the pipeline.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T09:38:16.004979+00:00— report_created — created