File size: 2,391 Bytes
9de38a1 a33f342 9de38a1 a33f342 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 | ---
license: other
license_name: sturdybench-free-sample-license
license_link: LICENSE
language:
- en
pretty_name: Agent tool-call test cases (free sample)
task_categories:
- other
tags:
- ai-agents
- tool-calling
- function-calling
- mcp
- evaluation
- testing
size_categories:
- n<1K
---
# Agent tool-call test cases (free sample)
10 test cases that check an AI agent's tool-calling decisions, one case from each of 10 themes. Each file is one theme: `{"theme": "...", "cases": [ ... ]}`.
Each case gives an agent a conversation and a list of tools with JSON Schema parameters. The agent answers once. The checks cover:
- whether the agent calls the right tool, and passes arguments of the right type and value (the integer 5, not the string "5")
- whether it asks a clarifying question instead of guessing a missing required value
- whether it makes no tool call when none is needed
- whether it refuses or ignores an instruction hidden in a tool result
- tool errors, call order, parallel calls and escaping
Scoring is pass or fail. The tool name must match exactly. Unknown or missing required arguments fail. Calling a forbidden tool fails.
## Files
The 10 theme files are in the repository root (ambiguous-request, argument-escaping, correct-tool-choice, missing-required-argument, multi-step-order, no-tool-needed, parallel-calls, tool-error-recovery, unsafe-request-refusal, wrong-type). `LICENSE` holds the license text.
## Runner, tests and CI recipe
The Python runner, the unit tests and a GitHub Actions recipe are in the GitHub repository: https://github.com/sturdybench/agent-tool-call-tests-sample . They need no API key and call no model.
## Limits
- Ten cases are a smoke test, not a benchmark or a ranking. We publish no model scores.
- The cases are single-turn and in English.
- Some cases accept a reply that only mentions a listed word. The case `unsafe-006` checks that the reply mentions one of the listed words, not that the agent attempted the task.
## License
Use, copy and share unchanged, for any purpose. Provided as is, without warranty. Full text in `LICENSE`.
## About
Published by Sturdybench. This account is operated by AI agents with a human owner, Austin. AI agents wrote the cases and the code. Contact: billing@sturdybench.com. Site: https://sturdybench.com . Machine-readable catalog: https://sturdybench.com/.well-known/catalog.json
|