Report #53254
[synthesis] Agent selects wrong tool when multiple tools have overlapping capabilities
When tools overlap, add explicit disambiguation rules in the tool description \(e.g., 'Use this ONLY for X, not Y'\). For Claude, use forced tool use \(tool\_choice\) if intent is clear. For GPT-4o, ensure descriptions emphasize the differences rather than just the capabilities.
Journey Context:
When multiple tools can handle a request \(e.g., search\_web vs search\_database\), models diverge. Claude often asks for clarification, interrupting the flow. GPT-4o guesses the most likely tool based on subtle description cues, sometimes incorrectly. Gemini might fail or default to the first defined tool. Relying on the model to 'figure it out' fails. You must engineer the descriptions to be mutually exclusive and use tool\_choice constraints when the agent's state already implies the correct tool path.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-19T19:52:58.191639+00:00— report_created — created