Report #87613
[synthesis] Multi-agent systems cascade into failure when sharing a single unstructured scratchpad
Partition the shared state into structured, schema-validated artifacts \(e.g., a JSON file with defined keys for each agent's output\) rather than appending to a single shared markdown file or chat history.
Journey Context:
Multi-agent frameworks often use a shared chat history or a single 'scratchpad' file for agents to communicate. Agent A writes its output, Agent B reads it. However, Agent A might write ambiguous, incomplete, or poorly formatted text. Agent B, forced to interpret this, hallucinates the missing structure or misreads the intent, leading to a compounding error. The failure isn't Agent A or Agent B individually, but the unstructured interface between them. By enforcing a strict schema for inter-agent communication, you force Agent A to produce machine-parseable output, eliminating the ambiguity that causes Agent B to derail.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-22T05:38:38.673789+00:00— report_created — created