fix: increment tool usage count in native function calling path

[Ricardo-M-L]

Summary

max_usage_count on tools is completely broken when using native function calling (GPT-4, Claude, Gemini, etc.) because current_usage_count is never incremented after successful tool execution in _execute_single_native_tool_call.

• The usage-limit check exists (crew_agent_executor.py lines 896-906) — it compares current_usage_count against max_usage_count
• But the increment after successful execution is missing, so the check always passes and tools can be called unlimited times
• The ReAct/text-based path in tool_usage.py correctly increments via _increment_usage_count() after each successful call

Changes

Added the same increment logic that exists in tool_usage.py to the native function calling path in crew_agent_executor.py, after the successful tool execution block inside _execute_single_native_tool_call:

if structured_tool and hasattr(structured_tool, "_increment_usage_count"):
    structured_tool._increment_usage_count()
elif original_tool and hasattr(original_tool, "current_usage_count"):
    original_tool.current_usage_count += 1

Uses structured_tool._increment_usage_count() (which syncs the count back to the original tool) when available, falling back to directly incrementing original_tool.current_usage_count.

Test plan

• Set max_usage_count=2 on a tool and run an agent with function_calling_llm (native path) — verify the tool is blocked after 2 calls
• Verify the ReAct path still works correctly with max_usage_count
• Verify tools without max_usage_count are unaffected

---

Note

Low Risk
Small behavioral fix in the native tool execution path to correctly increment usage counters; low risk but may newly block tools once their configured max_usage_count is reached.

Overview
Fixes max_usage_count enforcement for native function-calling tool execution by incrementing a tool’s usage counter after a successful call in CrewAgentExecutor._execute_single_native_tool_call.

When available, it uses CrewStructuredTool._increment_usage_count() to keep counts in sync with the original tool, and otherwise falls back to incrementing original_tool.current_usage_count directly.

^{Reviewed by Cursor Bugbot for commit f365379. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Cursor Bugbot has reviewed your changes and found 1 potential issue.

^{❌ Bugbot Autofix is OFF. To automatically fix reported issues with cloud agents, enable autofix in the Cursor dashboard.}

^{Reviewed by Cursor Bugbot for commit f365379. Configure here.}

[cursor]

Double-counting tool usage causes premature blocking

High Severity

The tool usage count is incremented twice per execution: once by BaseTool.run() (via _claim_usage()) and again by the new logic. This causes tools to hit their max_usage_count limit prematurely, effectively halving the configured usage.

 

^{Reviewed by Cursor Bugbot for commit f365379. Configure here.}