Report #11887
[architecture] Global tool registries causing naming collisions and degrading tool-selection accuracy
Scope tools locally to the specific agent that owns them; only inject the exact tools an agent needs into its context at runtime.
Journey Context:
Registering 50 tools globally so any agent can use them sounds flexible but destroys LLM tool-selection accuracy. The agent gets confused by overlapping schemas \(e.g., search\_knowledge\_base vs search\_web\). Scoping tools strictly to the agent's role increases selection accuracy, reduces prompt token usage, and prevents schema collision errors during function calling.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-16T14:38:13.580657+00:00— report_created — created