How OrangePro works
Proof that your tests catch what just changed.
After merge and before release, OrangePro finds the changed behavior your tests would not notice if it broke. It proves each new test by breaking that behavior on purpose, and hands you a test-only patch with a pass and fail receipt for every test. The engine is open source and runs on your machine.
Five steps from a merged change to proof you keep
- 1
Map the change
The engine reads your repository and finds the functions that recent merges touched, and how much real test evidence each one has. This step needs no model.
- 2
Rank the risk
Changed behavior with weak or missing test evidence goes to the top, so effort lands where a bug would reach customers.
- 3
Break it on purpose
For each risky function, OrangePro breaks that behavior in an isolated copy and reruns your tests. This is mutation testing, aimed at the code that just changed.
- 4
Keep only tests that catch it
Where no test notices, it can draft a focused test with the model provider you choose. A test counts only if it passes on your code and fails on the broken copy.
- 5
Hand back a test-only patch
Counted tests arrive as a patch to your test files, with a pass and fail receipt for each one. No production code changes. They live in your repo and run in your own CI.
The receipt is the product
A coverage number says a line ran. A receipt says a test failed when the behavior it guards was wrong, and passed when it was right. Each counted test carries both results, so a reviewer can check the claim instead of trusting it.
A test that never fails is not counted, however good it looks. A name match between a test and a function is not counted either.
receipt for one new test
- Original code pass
- Same code, behavior broken on purpose fail
The test caught the fault, so it is counted.
Patch touches test files only.
How it behaves
After merge, before release
OrangePro looks at what has already merged and is about to ship. It is not a pull request check and not a release gate. Your team decides what ships.
Independent of the code author
Whether a person, Claude Code, Cursor, Codex or Copilot wrote the change, the same check applies. The proof does not depend on the agent that wrote the code.
Honest about what it could not check
If OrangePro cannot run your tests, the result is unknown, with the blocker named and a manual next step. An unchecked change is never reported as low risk.
Your code, your machine, your key
The engine is MIT licensed and runs locally. Mapping and ranking need no model. Optional test drafting sends only the selected context to the provider you pick, with your key.
The local engine
The MIT-licensed engine, @orangepro/mcp-server, runs on your machine and works with Claude Code, Cursor, Codex and Copilot. The one pull request evaluation uses the same engine.
The platform for teams
We are building the hosted side with our first evaluation teams. It is not generally available yet. Today it starts with one real pull request, run with your team.
- Runs on your merged base build and the combined release build, with history across releases.
- Receipts and test patches collected in one place for the team.
- Private repositories connected through the OrangePro GitHub App.
See it on your own code
Run the free engine on your own repository to see where test protection is weakest. No account, and your source stays on your machine. How it fits next to your other tools.