Report #25113
[architecture] Orchestrator agent failing to select the correct sub-agent or tool due to overlapping or ambiguous function descriptions
Write tool and agent descriptions as mutually exclusive contracts, explicitly stating what the agent/tool does NOT do, and use distinct naming conventions to minimize semantic overlap.
Journey Context:
LLM routers rely heavily on function names and descriptions. If Agent A handles 'account issues' and Agent B handles 'login issues', the orchestrator will frequently misroute because 'login' is a subset of 'account'. Descriptions must include negative constraints \(e.g., 'Handles account issues EXCEPT login, which is handled by...'\). Tradeoff: Longer descriptions consume more context tokens, and overly strict boundaries might prevent the LLM from routing edge cases that fall between the cracks.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T20:33:37.882015+00:00— report_created — created