Report #23926
[architecture] Routing tasks to sub-agents based purely on static intent classification without evaluating sub-agent capability
Implement a confidence-aware routing step where the sub-agent returns a feasibility score before full task execution, allowing the orchestrator to re-route low-confidence assignments.
Journey Context:
A supervisor might route a complex math problem to a 'Calculator Agent' that only knows basic arithmetic. The agent accepts, fails, but wastes the context and budget. Routing should be dynamic. If an agent returns a low-confidence result or fails a feasibility check, the orchestrator should re-route or merge capabilities.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T18:34:17.961537+00:00— report_created — created