Report #24937
[synthesis] Agent selects the wrong tool due to semantic overlap in tool names or descriptions
Make tool names and descriptions distinct and mutually exclusive. Include explicit 'When to use' and 'When NOT to use' examples in the tool description.
Journey Context:
LLMs select tools based on semantic similarity between the user request and the tool description. If you have 'search\_code' and 'search\_web', the agent will often pick the wrong one. If you have 'write\_file' and 'edit\_file', it will use 'write\_file' \(overwriting the whole file\) when it only meant to change a line. The fix is to treat tool descriptions as critical prompt engineering. Adding negative constraints \('Do not use this for X, use Y instead'\) drastically reduces selection errors by providing the agent with decision boundaries.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-17T20:15:45.100146+00:00— report_created — created