swe-review-verifylisted

# swe-review-verify > Wraps `swe_review.VerifySkill` and `VerifierSubAgent`. ## When to use When you need to know whether a given PR resolves an issue, beyond what the LLM reviewer says. Do **not** use inside the review/revise prompt — the verifier CAN see oracle (ground-truth patch) for offline evaluation, but **the reviewer MUST NOT**. ## Hard constraints 1. Sandbox by default: clones the repo into a `tempfile.mkdtemp` and applies the patch there. User's working tree is untouched. 2. Optional `oracle` (gold patch) is only used for offline similarity scoring, never injected to a review prompt. 3. Test runner template uses `{test}` placeholder; default tries `pytest -q`, `pytest -x -q`, `unittest`, `npm test`. ## How to invoke ```python from swe_review import VerifySkill verify = VerifySkill(repo_path=".") res = await verify.execute( pr_diff=open("pr.diff").read(), test_info={ "fail_to_pass": ["tests/test_x.py::test_foo"], "pass_to_pass": ["tests/test_x.py::test_bar"], }, # oracle is optional AND restricted to evaluator-only paths: oracle=open("gold.diff").read(), ) print(res.payload["resolution_status"], res.payload["confidence"]) ``` ## Output keys - `passed: bool` - `test_results: [{test, type, should_pass, passed, stdout_tail, stderr_tail}]` - `resolution_status`: resolved | not_resolved | partially_resolved | unknown - `confidence: float` - `details: str` - `patch_applied: bool`, `sandbox_used: bool` - `oracle_similarity: flo