opro

Use the OrangePro MCP tools for local evidence-graph analysis and grounded test generation.

1. Start with orangepro_start for the current checkout.
2. If OrangePro reports a large-repo scope breakdown, prefer a focused scope for AI/generation; full deterministic analysis is still allowed.
3. For PR work, call orangepro_generate_tests with base_ref=main.
4. For baseline work, call orangepro_find_test_gaps, pick one high-priority gap, then call orangepro_generate_tests for that target.
5. Write only generated tests that include run_hints; drafts are context, not runnable claims.
6. Run the suggested command from the owning package directory.
7. After a pass, take the prove_run args returned by orangepro_generate_tests and pass them to orangepro_prove so OrangePro can dynamically prove the target. Use record_run only for static diagnostics.
8. Report status as Proven, Reproven, Runs under tests, Linked, Unconfirmed, or No test found. Never promote Linked, Unconfirmed or AI links to Proven. A successful orangepro_prove returns proven: true and stores the ledger status reproven, including the first time. Proven is the report count of behaviors with that certificate. Reproven (this run) is the traceability label for the same certificate. It does not mean the function was proved before.

For JavaScript or TypeScript, orangepro_prove needs target_symbol in the form sym:<file>#<Symbol>, test_path, and replacement set to an empty body or a single literal return such as return null;. An async function also needs replacement_mode promise-json. Set runner to the repo's runner (jest, vitest, or mocha). Example: target_symbol sym:index.js#run, test_path index.test.js, replacement return null;, replacement_mode promise-json, runner jest.

OrangePro may use weak AI grounding when a provider key is configured, but AI links are suggestions for generation only and never change Proven coverage.