ReqToTest Agentic Test Design for UiPath Test Cloud

Inspiration

I'm a first-time hackathon builder and college freshman. I wanted to build something that solved a real problem — not just a demo. QA teams hand-write test cases from specs, and the ones they forget (negative paths, boundary conditions) are exactly the ones that ship bugs. I wanted agents that force honest coverage, not just volume.

What it does

ReqToTest is a closed-loop, risk-aware agentic testing system built on UiPath Test Cloud.

Triage → Generate (provable risk coverage) → Human gate → Create in Test Cloud

1. An input-triage agent judges whether the requirement is testable. If it's vague, it pushes back with clarifying questions instead of guessing.
2. Four specialist agents (happy / edge / negative / boundary) design scenarios grounded in the Agile Testing Quadrants (Crispin & Gregory). Every scenario includes a traced Condition of Satisfaction.
3. A coverage-critic agent audits the full set and names concrete missing cases — triggering a targeted refine round.
4. A human approves in UiPath Action Center before anything touches Test Cloud.
5. Approved scenarios are created as live test cases in UiPath Test Manager via the v2 REST API.

How I built it

• Reasoning core: Python + Pydantic v2 schema-forced structured output, provider-agnostic LLM client (Groq / Gemini / Ollama — no paid key required)
• Agent orchestration: LangGraph multi-agent pipeline with a coverage critic + one refine round
• UiPath integration: Coded Agent (Python SDK + uipath-langchain), Action Center interrupt for human approval, Test Manager v2 REST API for live test case creation
• Built with Claude Code — scaffolded the schema, coverage-forcing prompts, CLI, Test Cloud integration, and LangGraph port

Challenges

• Getting schema-forced structured output to reliably populate all four risk buckets across LLM providers
• Wiring the Action Center interrupt so the human gate is real, not cosmetic
• Verifying the exact Test Manager v2 API payload shape from the live Swagger before writing the integration

What I learned

The coverage critic is the key differentiator — without it, specialists overlap and miss buckets. One targeted refine round after critique consistently produces better coverage than running more specialists. As a first-time builder, learning to ship a working integration (not just a prototype) was the biggest lesson.

Built With

• action
• agents
• center
• claude
• cloud
• coded
• groq
• langgraph
• pydantic
• python
• rest
• test
• uipath