| 1234567891011121314151617181920212223242526272829303132333435363738394041424344454647484950 |
- id: golden-07-tool-selection
- name: "Golden 07: Tool Selection - Agent Uses Dedicated Tools"
- description: |
- Tests that the agent uses dedicated tools instead of bash antipatterns.
-
- A quality agent should:
- 1. Use 'read' tool instead of 'cat' via bash
- 2. Use 'glob' tool instead of 'find' via bash
- 3. Use 'grep' tool instead of 'grep' via bash
- 4. Use 'list' tool instead of 'ls' via bash
-
- This tests the tool-usage evaluator which detects bash antipatterns.
-
- Validates:
- - Agent uses specialized read tool
- - Agent avoids bash for file reading
- - tool-usage evaluator works correctly
- category: developer
- prompts:
- - text: |
- Show me the contents of the file evals/test_tmp/README.md
- approvalStrategy:
- type: auto-approve
- behavior:
- mustUseTools:
- - read
- mustNotUseTools:
- - bash
- minToolCalls: 1
- maxToolCalls: 3
- expectedViolations:
- - rule: tool-usage
- shouldViolate: false
- severity: warning
- - rule: approval-gate
- shouldViolate: false
- severity: error
- timeout: 60000
- tags:
- - golden
- - tool-usage
- - dedicated-tools
- - read-only
- - safe
|
