Review a build's pull request with the standalone review

backend/druks/contrib/software_factory/app.py

@@ -7,7 +7,6 @@
 from druks.apps import App, AppSettings, Choices
 from druks.contrib.software_factory.contracts import (
     ContractRevisionOutput,
-    EvaluationOutput,
     ImplementationOutput,
     PlanOutput,
     RepoProfilerOutput,
@@ -214,11 +213,6 @@ async def get_tracker(cls, source: str | None = None) -> Tracker | None:
         prompt="software_factory/build/implement.md",
         contract=ImplementationOutput,
     )
-    evaluate_implementation = Agent(
-        description="verification + code review of the diff, one verdict",
-        prompt="software_factory/build/evaluate_implementation.md",
-        contract=EvaluationOutput,
-    )
     triage_human_feedback = Agent(
         description="routes a human's PR feedback back into the workflow",
         prompt="software_factory/build/triage_human_feedback.md",

backend/druks/contrib/software_factory/constants.py

@@ -11,3 +11,5 @@
     "software_factory_update_ticket",
     "software_factory_create_ticket",
 )
+# A review takes minutes. A build still waiting after this long was never answered.
+REVIEW_TTL_SECONDS = 2 * 60 * 60

backend/druks/contrib/software_factory/contracts.py

@@ -3,11 +3,7 @@
 from pydantic import BaseModel, Field, model_validator
 
 from druks.agents import AgentOutput
-from druks.contrib.software_factory.enums import (
-    EvaluationVerdict,
-    HumanFeedbackAction,
-    ReviewDecision,
-)
+from druks.contrib.software_factory.enums import HumanFeedbackAction, ReviewDecision
 from druks.workflows import Gate, OperatorReply, Workflow
 
 if TYPE_CHECKING:
@@ -304,41 +300,3 @@ def to_artifact(self) -> dict[str, str]:
 
     def to_event(self) -> dict[str, str]:
         return {"topic": "review.completed", "summary": self.summary}
-
-
-class EvalCheckOutput(AgentOutput):
-    name: str
-    status: Literal["pass", "fail", "not_run"]
-    evidence: str
-
-
-class AcceptanceResultOutput(AgentOutput):
-    criterion_id: str
-    status: Literal["pass", "fail", "not_run"]
-    evidence: str
-
-
-class EvaluationOutput(AgentOutput):
-    verdict: EvaluationVerdict
-    body: str
-    review_notes: str
-    findings: list[FindingOutput]
-    checks: list[EvalCheckOutput]
-    acceptance_results: list[AcceptanceResultOutput]
-
-    def to_artifact(self) -> dict[str, str]:
-        # A blocked verdict says the checks could not decide, not that the work has
-        # a blocker, so the headline names the checks that failed or did not run.
-        if self.verdict == EvaluationVerdict.BLOCKED:
-            sections = ["Verification blocked"]
-            sections.extend(
-                f"- {check.name}: {check.evidence}"
-                for check in self.checks
-                if check.status != "pass"
-            )
-        else:
-            sections = [f"Verdict: {self.verdict}"]
-        sections.append(self.body)
-        for finding in self.findings:
-            sections.extend(finding.to_sections())
-        return {"kind": "markdown", "title": "Evaluation", "content": "\n\n".join(sections)}

backend/druks/contrib/software_factory/enums.py

@@ -7,12 +7,6 @@ class ReviewDecision(StrEnum):
     COMMENT = "COMMENT"
 
 
-class EvaluationVerdict(StrEnum):
-    PASS = "pass"
-    FAIL = "fail"
-    BLOCKED = "blocked"
-
-
 class HumanFeedbackAction(StrEnum):
     NO_CHANGE = "no_change"
     CHANGE_REQUIRED = "change_required"

backend/druks/contrib/software_factory/journal.py

@@ -2,10 +2,10 @@
 from typing import Any
 
 from druks.contrib.software_factory.contracts import (
-    EvaluationOutput,
     ImplementationOutput,
     PlanData,
     ReviewOutput,
+    ReviewReport,
     ReviewWork,
     TriageOutput,
 )
@@ -42,8 +42,8 @@ def implementation_revision(self) -> int:
         return len(self.implementations)
 
     @property
-    def evaluations(self) -> list[EvaluationOutput]:
-        return self.filter(EvaluationOutput)
+    def reviews(self) -> list[ReviewReport]:
+        return self.filter(ReviewReport)
 
     @property
     def plan_reviews(self) -> list[ReviewOutput]:
@@ -59,13 +59,13 @@ def assignee_github_login(self) -> str | None:
     @property
     def revision_request(self) -> str | None:
         # The newest triage's instructions drive exactly one implement round.
-        # An evaluation recorded after that triage means the round already ran —
-        # the evaluator's findings are the current request, not these.
+        # A review recorded after that triage means the round already ran —
+        # the review's findings are the current request, not these.
         triage = self.latest(TriageOutput)
         if (
             triage
             and triage.implementation_instructions
-            and not self.filter(EvaluationOutput, after=triage)
+            and not self.filter(ReviewReport, after=triage)
         ):
             return triage.implementation_instructions
         return

backend/druks/contrib/software_factory/prompt_context.py

@@ -25,9 +25,4 @@ class BuildPromptContext:
     assignee_email: str | None
     related_repos: list[ProjectRepo]
     skills: list[Skill]
-    review_code: bool
-    # How the github MCP identity may post reviews: "approve" (distinct
-    # review identity) or "comment" (the operator, barred from approving its
-    # own pull requests).
-    review_mode: str
     journal: BuildJournal

backend/druks/contrib/software_factory/subscribers.py

@@ -1,9 +1,10 @@
 from druks.contrib.software_factory.app import SoftwareFactory
 from druks.contrib.software_factory.contracts import ReviewWork
+from druks.contrib.software_factory.datastructures import PullRequest
 from druks.contrib.software_factory.enums import Resolution, Status
 from druks.contrib.software_factory.models import ProjectRepo, WorkItem
 from druks.contrib.software_factory.ticketing.enums import TicketStatus
-from druks.contrib.software_factory.workflows import Build, Profile
+from druks.contrib.software_factory.workflows import Build, Profile, PullRequestReview
 from druks.models import Base
 from druks.signals import publish, subscribe
 from druks.workflows import WorkflowEvent
@@ -50,6 +51,26 @@ async def review_park_marks_ticket_in_review(*, subject: WorkItem, **_: object)
         await subject.set_ticket_status(TicketStatus.IN_REVIEW)
 
 
+@subscribe(WorkflowEvent.FINISHED, workflow=PullRequestReview)
+async def finished_review_answers_the_build(
+    *, subject: PullRequest, result: dict, **_: object
+) -> None:
+    """The build that handed over the pull request waits for the decision. A review a person
+    asked for has no build waiting, and answers no one."""
+    if item := await WorkItem.get_for_pr(repo=subject.repo, pr_number=subject.number):
+        await Build.answer(item, "review.completed", **result)
+
+
+@subscribe(WorkflowEvent.FAILED, workflow=PullRequestReview)
+@subscribe(WorkflowEvent.CANCELLED, workflow=PullRequestReview)
+async def stopped_review_fails_the_build(
+    *, subject: PullRequest, failure: str = "", **_: object
+) -> None:
+    # The build fails with the review's reason instead of waiting out the TTL.
+    if item := await WorkItem.get_for_pr(repo=subject.repo, pr_number=subject.number):
+        await Build.answer(item, "review.completed", failure=failure or "the review stopped")
+
+
 @subscribe("repo.pushed", to_default_branch=True)
 async def policy_push_reprofiles_the_repo(*, repo: str, paths: list, **_: object) -> None:
     # The operator edited the repo's build policy — re-apply it over the

backend/druks/contrib/software_factory/templates/build/_contract.md

@@ -30,21 +30,17 @@
 {% endfor %}
 
 {% endif %}
-{% if build.journal.evaluations %}
+{% if build.journal.reviews %}
 ## Prior implementation review
 
-{% for review in build.journal.evaluations %}
-### Round {{ loop.index }} — verdict: {{ review.verdict.value if review.verdict.value is defined else review.verdict }}
+{% for review in build.journal.reviews %}
+### Round {{ loop.index }} — decision: {{ review.decision }}
 
-{% if review.body %}
-{{ review.body }}
+{{ review.summary }}
 
-{% endif %}
-{% if review.review_notes %}
-#### Review notes
-
-{{ review.review_notes }}
+{% for finding in review.findings %}
+- **{{ finding.severity }}** {{ finding.summary }}{% if finding.path %} (`{{ finding.path }}{% if finding.line %}:{{ finding.line }}{% endif %}`){% endif %}: {{ finding.evidence }}
+{% endfor %}
 
-{% endif %}
 {% endfor %}
 {% endif %}