DEV Community· Dr Srinivas Padmanabhuni·· 3 小時前AI 評分58
複製 agentest，在 CI 離線執行 AI 智能體檢查清單測試
Clone agentest: run agentic AI checklist tests offline in CI
AI 導讀
agentest 是一款 Apache-2.0 open-core 測試執行器，可在 CI 離線執行 Agentic AI Test Checklist 110，執行時毋須 Hugging Face 帳戶。
正文
Clone agentest: run agentic AI checklist tests offline in CI
Shipping agents that plan → call tools → hand off → mutate state? Chat review and LLM-as-judge miss a lot of failures: false completion, wrong tool order, privacy leaks, biased labels, flaky plans.
agentest is an Apache-2.0 open-core runner for the Agentic AI Test Checklist 110. Clone it and run — no Hugging Face account required at runtime.
Quick start
git clone https://github.com/srpadmanabhuni/agentest.git
cd agentest
python -m pip install -e ".[dev]"
agentest samples
agentest run --suite functional --agent golden
agentest run --suite orchestration --agent golden
agentest matrix
What you get
- Suites: functional, privacy, explainability, bias_fairness, orchestration
- 15 sample agents (10 green ladder + 5 expected-fail teaching scenarios)
- Deterministic oracles against a mock tool gateway
- ISO 42001 / assurance-taxonomy aligned checklist design (not certification)
Context
- Pulse: Testing Agentic AI
- Zenodo taxonomy: https://doi.org/10.5281/zenodo.23587407
- Developer guide: https://github.com/srpadmanabhuni/agentest/blob/main/docs/DEVELOPER-GUIDE.md
Author: Dr Srinivas Padmanabhuni · srinivas AT testAIng.com
來源：DEV Community · dev.to