Promptfoo — проверка промптов и защита ИИ-приложений

★ 8.0 · testing

Promptfoo is a plugin for Claude Code that helps you connect providers and targets, write eval suites, and set up or run red team scans for apps built on language models. Promptfoo itself is an open source command line tool and library under the MIT license: it tests prompts, agents and RAG pipelines, compares models side by side (OpenAI, Anthropic, Azure, Bedrock, Ollama and more), and runs red teaming, meaning it looks for weak spots such as guardrail bypasses. Use it when you need to check prompt quality before a release, compare which model answers better, add checks to CI/CD, or review pull requests for LLM-related security issues. Evals run locally, so your prompts never leave your machine. To install the tool itself: npm install -g promptfoo, then promptfoo init --example getting-started; most providers need your own API key.