initial impl

tests/test_code_review.py

@@ -0,0 +1,121 @@
+from pathlib import Path
+from types import SimpleNamespace
+
+from agentci.domain.models import (
+    AgentResult,
+    Job,
+    JobKind,
+    ReviewFinding,
+    ReviewReport,
+    ReviewSeverity,
+    Workflow,
+    WorkflowKind,
+)
+from agentci.workflows.code_review import CodeReviewLoop
+
+
+def serious_report() -> ReviewReport:
+    return ReviewReport(
+        summary="Needs work",
+        findings=[
+            ReviewFinding(
+                severity=ReviewSeverity.MAJOR,
+                title="Missing check",
+                detail="A check is absent.",
+                recommendation="Add it.",
+            )
+        ],
+    )
+
+
+class FakeCodex:
+    def __init__(self, reports: list[ReviewReport]) -> None:
+        self.reports = iter(reports)
+        self.reviews = 0
+        self.revisions = 0
+
+    async def start(self, **_kwargs):
+        self.reviews += 1
+        return "reviewer", next(self.reports)
+
+    async def resume(self, **kwargs):
+        if kwargs["result_type"] is ReviewReport:
+            self.reviews += 1
+            return next(self.reports)
+        self.revisions += 1
+        return AgentResult(summary_markdown=f"revision {self.revisions}", tests=[])
+
+
+class FakeStorage:
+    async def update_job(self, *_args, **_kwargs):
+        return None
+
+    async def update_workflow(self, *_args, **_kwargs):
+        return None
+
+
+class FakePrompts:
+    def render(self, name, **_kwargs):
+        return name
+
+
+def objects(rounds: int, reports: list[ReviewReport]):
+    codex = FakeCodex(reports)
+    settings = SimpleNamespace(
+        implement_review_rounds=rounds,
+        implement_model="model",
+        implement_reasoning="high",
+    )
+    deps = SimpleNamespace(
+        settings=settings,
+        codex=codex,
+        storage=FakeStorage(),
+        prompts=FakePrompts(),
+    )
+    workflow = Workflow(
+        id="flow",
+        kind=WorkflowKind.IMPLEMENT,
+        repo_owner="org",
+        repo_name="repo",
+        issue_number=1,
+        workspace_path=Path("."),
+        base_sha="abc",
+        primary_session_id="primary",
+    )
+    job = Job(
+        id="job",
+        kind=JobKind.IMPLEMENT,
+        target_key="target",
+        repo_owner="org",
+        repo_name="repo",
+        issue_number=1,
+        pr_number=None,
+        requester="alice",
+        message="",
+        comment_id=1,
+    )
+    return CodeReviewLoop(deps), codex, workflow, job  # type: ignore[arg-type]
+
+
+async def test_stops_after_clean_second_review() -> None:
+    clean = ReviewReport(summary="Ready", findings=[])
+    loop, codex, workflow, job = objects(4, [serious_report(), clean])
+    _, report = await loop.run(
+        job, workflow, "context", AgentResult(summary_markdown="initial", tests=[])
+    )
+    assert not report.has_serious_findings
+    assert codex.reviews == 2
+    assert codex.revisions == 1
+
+
+async def test_does_not_make_unreviewed_final_revision() -> None:
+    loop, codex, workflow, job = objects(
+        3, [serious_report(), serious_report(), serious_report()]
+    )
+    _, report = await loop.run(
+        job, workflow, "context", AgentResult(summary_markdown="initial", tests=[])
+    )
+    assert report.has_serious_findings
+    assert codex.reviews == 3
+    assert codex.revisions == 2
+