paperclip/server/src/__tests__/run-liveness.test.ts

import { describe, expect, it } from "vitest";
import { classifyRunLiveness } from "../services/run-liveness.ts";

const baseInput = {
  runStatus: "succeeded",
  issue: {
    status: "in_progress",
    title: "Implement feature",
    description: "Add the requested behavior.",
  },
  resultJson: null,
  stdoutExcerpt: null,
  stderrExcerpt: null,
  error: null,
  errorCode: null,
  continuationAttempt: 0,
  evidence: null,
};

describe("run liveness classifier", () => {
  it("classifies text-only future work as plan_only", () => {
    const classification = classifyRunLiveness({
      ...baseInput,
      resultJson: {
        summary: "I will inspect the repo next and then implement the fix.",
      },
    });

    expect(classification.livenessState).toBe("plan_only");
    expect(classification.nextAction).toContain("inspect the repo");
  });

  it("classifies empty successful output as empty_response", () => {
    const classification = classifyRunLiveness(baseInput);

    expect(classification.livenessState).toBe("empty_response");
  });

  it("treats issue comments, documents, products, and actions as progress", () => {
    const latestEvidenceAt = new Date("2026-04-18T12:00:00Z");
    const classification = classifyRunLiveness({
      ...baseInput,
      resultJson: {
        summary: "Updated implementation.",
      },
      evidence: {
        issueCommentsCreated: 1,
        documentRevisionsCreated: 1,
        workProductsCreated: 1,
        toolOrActionEventsCreated: 1,
        latestEvidenceAt,
      },
    });

    expect(classification.livenessState).toBe("advanced");
    expect(classification.lastUsefulActionAt).toBe(latestEvidenceAt);
  });

  it("does not treat workspace operations alone as concrete progress", () => {
    const classification = classifyRunLiveness({
      ...baseInput,
      resultJson: {
        summary: "I will inspect the repo next.",
      },
      evidence: {
        workspaceOperationsCreated: 1,
        latestEvidenceAt: new Date("2026-04-18T12:00:00Z"),
      },
    });

    expect(classification.livenessState).toBe("plan_only");
    expect(classification.lastUsefulActionAt).toBeNull();
  });

  it("exempts planning/document tasks from plan-only retry classification", () => {
    const classification = classifyRunLiveness({
      ...baseInput,
      issue: {
        status: "in_progress",
        title: "Draft implementation plan",
        description: "Create a plan for the work.",
      },
      resultJson: {
        summary: "Plan:\n- Inspect files\n- Implement after approval",
      },
    });

    expect(classification.livenessState).toBe("advanced");
  });

  it("exempts runs that update the plan document from plan-only classification", () => {
    const classification = classifyRunLiveness({
      ...baseInput,
      resultJson: {
        summary: "Next steps:\n- inspect files\n- implement the service",
      },
      evidence: {
        documentRevisionsCreated: 1,
        planDocumentRevisionsCreated: 1,
        latestEvidenceAt: new Date("2026-04-18T12:00:00Z"),
      },
    });

    expect(classification.livenessState).toBe("advanced");
  });

  it("classifies done issues as completed", () => {
    const classification = classifyRunLiveness({
      ...baseInput,
      issue: {
        ...baseInput.issue,
        status: "done",
      },
      resultJson: {
        summary: "Finished the implementation.",
      },
    });

    expect(classification.livenessState).toBe("completed");
  });

  it("classifies declared blockers as blocked", () => {
    const classification = classifyRunLiveness({
      ...baseInput,
      resultJson: {
        summary: "I cannot proceed because I need access credentials.",
      },
    });

    expect(classification.livenessState).toBe("blocked");
  });
});