Report #83793
[synthesis] Model selects the wrong tool when multiple tools have similar names or overlapping capabilities
Differentiate tool names semantically \(verb\_noun\) and use the tool description to explicitly state when \*not\* to use the tool, rather than just when to use it.
Journey Context:
When presented with search\_web and search\_database, GPT-4o defaults to the first tool defined in the list if descriptions are similar. Claude 3.5 Sonnet attempts to merge parameters from both or throws an ambiguous selection error. Gemini often picks the shortest tool name. The cross-model fix is to use highly specific action-oriented names \(e.g., query\_public\_web, query\_internal\_db\) and add negative constraints in descriptions \(e.g., 'Do not use this for internal data'\).
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-21T23:13:53.891266+00:00— report_created — created