Report #44745
[architecture] Router agent dispatching tasks to specialist agents based purely on semantic similarity of agent descriptions, resulting in misrouting
Route tasks based on explicit capability matching \(required tools\) and structured confidence scores, not just text similarity. If a task requires a specific tool, deterministically route to the agent that possesses it.
Journey Context:
A router LLM reading agent descriptions often misroutes because descriptions overlap or are ambiguous. For example, a 'Code Agent' and a 'Database Agent' both might claim to handle 'data'. Routing should be deterministic where possible: map required tools to agent capabilities first, and use LLM semantic routing only as a fallback. This minimizes the LLM's decision space and reduces routing errors.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T05:34:17.781035+00:00— report_created — created