Inspiration
AI agents can repair tests, but a repaired test can become green for the wrong reason: weakened assertions, skipped checks, or evidence reused from another candidate. RunProof AI was built to make UiPath Test Cloud evidence safer for agentic software testing.
What it does
RunProof AI is an anti-false-green evidence governor for AI-repaired tests. It checks whether the repaired test still belongs to the same candidate, preserves defect detection, keeps certification evidence clean, and allows only human release review.
How we built it
The prototype uses a public-safe RunProof validator, synthetic UiPath Test Cloud evidence packets, a UiPath report adapter, a demo dashboard, tests, and submission audit scripts. Merge and deploy are always blocked.
Challenges we ran into
The main challenge was separating a real repair from a false-green repair. A test can pass because the system is fixed, but it can also pass because an assertion was weakened, a check was skipped, or evidence from another candidate was reused. We also had to keep the public demo safe by excluding credentials, private validation banks, seeds, and thresholds.
Accomplishments that we're proud of
We built a working public-safe prototype with HOLD, BLOCK, and human-review-only outcomes. RunProof AI blocks assertion weakening, blocks subject mismatch, keeps merge and deploy disabled, and produces an honest claim boundary for live UiPath tenant evidence.
What we learned
We learned that agentic testing needs more than a green check. UiPath Test Cloud evidence becomes much more useful when it is bound to the exact candidate, report digest, requirement links, and assertion/log references.
What's next for RunProof AI
Next, we want to bind a real UiPath Test Cloud execution through RUN014, publish the GitHub repository, upload the demo video, and connect RunProof AI to a Studio Web or Automation Cloud workflow for human release-review routing.
Log in or sign up for Devpost to join the conversation.