Сквозное тестирование AI-агента с Ollama и облачными API

testing

kosmos-e2e-testing is a Claude Code skill that automates end-to-end testing for the Kosmos autonomous AI scientist project, supporting local Ollama models, Anthropic and OpenAI APIs, and a Docker sandbox for full workflow validation. Tests are organized into four tiers: sanity (~30 s, fast local model qwen3:4b), smoke (~2 min), e2e (~10 min, deepseek-r1:8b reasoning model), and production (~20 min, external API with Claude or GPT-4). The skill ships with shell scripts (run-tests.sh, setup-docker.sh, health-check.sh), a Python library (provider_detector, test_runner, config_manager, report_generator), test templates, and per-provider config files. It is built for teams that need to switch between local and cloud inference, benchmark model performance, and test the ResearchWorkflow component in an isolated environment.