vostride/agent-qa

What it solves

It eliminates the fragility of traditional automated testing for web and mobile apps. Instead of relying on static selectors that break when the UI changes, it allows developers to write tests in natural language and uses AI agents to navigate the interface and handle UI drift automatically.

How it works

The system uses LLMs to interpret natural language test definitions and execute actions based on the current screen state, roles, and labels. It features a self-healing mechanism that re-observes the UI and attempts alternative paths if a step fails. To improve efficiency, it maintains an execution memory of past runs and a smart cache to reuse validated plans, reducing token costs and runtime. It also supports sandboxed hooks (via Docker) for environment setup and API calls.

Who it’s for

QA engineers and developers who want to reduce the maintenance burden of automated UI tests for web and mobile applications.

Highlights

• Natural Language Testing: Define actions and assertions using human language rather than rigid code selectors.
• Self-Healing Execution: Automatically recovers from flaky interactions or UI changes during a test run.
• Execution Memory: Learns from past successes and failures to optimize future test runs.
• Sandboxed Hooks: Run isolated Node, Bun, Python, or Bash scripts for setup and teardown.
• LLM Flexibility: Compatible with OpenAI, Anthropic, Gemini, and local open-source models.