agent-qa

About agent-qa

agent-qa is an open-source QA test harness that lets developers write tests in natural language for web and mobile applications, with agents working from visible UI roles, labels, and screen state rather than fixed selectors. When a sub-action such as a click, fill, or select fails during execution, it re-observes the UI and attempts a different path within the same run to recover from UI drift or flaky interactions. It builds execution memory from product, suite, and test observations across runs, incorporating context from previously healed steps to inform future test execution, and caches validated action plans to reduce planner work and token usage on similar subsequent runs.

The harness includes a dashboard and CLI for developers, along with MCP and skills support for coding agents. It can run Node, Bun, Python, or Bash hooks in sandboxed Docker containers for tasks like environment setup, API calls, fixture seeding, and state teardown. Tests, configs, hooks, memory, and suite logic are stored as version-controlled code. It supports OpenAI- and Anthropic-compatible endpoints, Gemini, local or open-source models, and subscription-based auth for Codex and Claude Code.

The product is aimed at software teams building automated QA/end-to-end test suites for web and mobile applications, distributed as an npm package (`agent-qa`) with a separate package for Codex/Claude Code subscription authentication, and installed and run via the `npx agent-qa` CLI.

Key features

• Write tests in natural language for web and mobile apps
• Self-healing test execution that re-observes the UI and retries failed actions
• Builds execution memory across runs to improve future test accuracy
• Runs sandboxed Node, Bun, Python, or Bash hooks in isolated Docker containers
• Action cache reuses validated plans to reduce token usage and runtime
• Bring your own LLM via OpenAI- and Anthropic-compatible endpoints, Gemini, or local models

Frequently asked questions