Finalrun – Spec-driven testing using English and vision for mobile apps
Vision-based execution beats brittle XPath selectors that break on every UI change.

Interesting concept, but landing page is broken and nothing is actually open source yet.
Mobile app developers and QA engineers
TestRail · BrowserStack · Applitools
Finalrun replaces locators with human-readable specs. You write plain-English instructions (e.g., “Tap Settings, search ‘Spanish’, verify ‘Español’”), and a vision-based QA agent drives the app visually like a human — so renamed IDs don’t break tests.
With plain english agent for testing mobile apps, Spec driven approach works smoothly without much maintainance: 1. With skills, The agent can analyze your repo, 2. Auto-generate edge-case specs (with setup/cleanup), and export them as .md tests 3. Now you can run the test via CLI (./mobile-cli run ./test/search.md).
If you want early access or the code when it’s open-sourced, I’m collecting emails: https://docs.google.com/forms/d/1EwHjqK6t1pBQgsKWih1Z_hQqP83...
(Also demo: https://youtu.be/SsVHRDWk_ss)
Vision-based execution beats brittle XPath selectors that break on every UI change.
Natural language test generation via Claude; OpenAPI scanner auto-detects specs from source.
Spec-to-code pipeline, but Cursor and Continue already do this with better traction.
It extracts focused, executable operations from giant OpenAPI files (the GitHub REST YAML is shown) to shrink context and avoid sidecar adapter sprawl — a pragmatic answer to token bloat and brittle ad-hoc integrations. Useful and concrete: if it actually generates tidy, updateable skill units and runtime hooks it saves a lot of maintenance. That said, the idea competes with existing LangChain/openai-function patterns; the repo will need clear runtime, versioning, and update strategies to feel like more than a nicer converter.
Another config standard for AI agents when each harness already has its own format.
Natural language to live API in 14 seconds with chaos injection and team share links.