gsd-pi 2.81.0 → 2.82.0-dev.20138ae42

package/src/resources/extensions/gsd/tests/plan-milestone-sketch-render.test.ts

@@ -0,0 +1,157 @@
+// ADR-011 #5750: ROADMAP.md renders sketch slices with a `[sketch]` badge.
+//
+// Locks in the visual distinction so an auditor scanning the rendered roadmap
+// can tell which slices are sketches awaiting refine-slice expansion vs which
+// already carry a full plan. Sits alongside `plan-milestone.test.ts` which
+// covers the full-plan render path.
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, mkdirSync, rmSync, readFileSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+
+import { openDatabase, closeDatabase } from "../gsd-db.ts";
+import { handlePlanMilestone, type PlanMilestoneParams } from "../tools/plan-milestone.ts";
+
+function makeTmpBase(): string {
+  const base = mkdtempSync(join(tmpdir(), "gsd-plan-sketch-render-"));
+  mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });
+  openDatabase(join(base, ".gsd", "gsd.db"));
+  return base;
+}
+
+function cleanup(base: string): void {
+  try {
+    closeDatabase();
+  } catch {
+    /* noop */
+  }
+  try {
+    rmSync(base, { recursive: true, force: true });
+  } catch {
+    /* noop */
+  }
+}
+
+function planMilestoneWithSketches(): PlanMilestoneParams {
+  return {
+    milestoneId: "M001",
+    title: "Progressive Planning Demo",
+    vision: "Demonstrate sketch slices in ROADMAP rendering.",
+    successCriteria: ["S01 full, S02 sketch", "ROADMAP distinguishes them"],
+    keyRisks: [{ risk: "Visual collision", whyItMatters: "Auditors need to spot sketches." }],
+    proofStrategy: [{ riskOrUnknown: "Render correctness", retireIn: "S01", whatWillBeProven: "Roadmap shows the badge." }],
+    verificationContract: "Contract verification text",
+    verificationIntegration: "Integration verification text",
+    verificationOperational: "Operational verification text",
+    verificationUat: "UAT verification text",
+    definitionOfDone: ["Renderer emits badge", "Test asserts it"],
+    requirementCoverage: "Covers ADR-011 #5750.",
+    boundaryMapMarkdown: "| From | To | Produces | Consumes |\n|------|----|----------|----------|\n| S01 | S02 | scaffold | nothing |",
+    slices: [
+      {
+        sliceId: "S01",
+        title: "Fully planned scaffold",
+        risk: "medium" as const,
+        depends: [],
+        demo: "Scaffold is in place.",
+        goal: "Lay down the structural foundation.",
+        successCriteria: "Scaffold tests pass.",
+        proofLevel: "integration" as const,
+        integrationClosure: "Downstream slices depend on this scaffold.",
+        observabilityImpact: "No new telemetry.",
+        // No isSketch flag — defaults to full plan.
+      },
+      {
+        sliceId: "S02",
+        title: "Refinement candidate",
+        risk: "low" as const,
+        depends: ["S01"],
+        demo: "Sketched until S01 ships.",
+        goal: "Refine into a full plan after S01 lands.",
+        successCriteria: "",
+        proofLevel: "",
+        integrationClosure: "",
+        observabilityImpact: "",
+        isSketch: true,
+        sketchScope: "Pick up the scaffold from S01 and add the demo behavior. Stay inside the existing module boundaries.",
+      },
+    ],
+  };
+}
+
+test("ROADMAP renders sketch slices with [sketch] badge and full slices without", async () => {
+  const base = makeTmpBase();
+  try {
+    const params = planMilestoneWithSketches();
+    const result = await handlePlanMilestone(params, base);
+    if ("error" in result) {
+      assert.fail(`handlePlanMilestone failed: ${result.error}`);
+    }
+
+    const roadmapPath = join(base, ".gsd", "milestones", "M001", "M001-ROADMAP.md");
+    const roadmap = readFileSync(roadmapPath, "utf-8");
+
+    // S01 is a full slice — no sketch badge.
+    const s01Line = roadmap.split("\n").find((line) => line.includes("**S01:"));
+    assert.ok(s01Line, "S01 slice line must exist in roadmap");
+    assert.equal(
+      s01Line.includes("`[sketch]`"),
+      false,
+      "fully-planned S01 must NOT carry the sketch badge",
+    );
+    assert.match(s01Line, /`risk:medium`/);
+
+    // S02 is a sketch — badge required, positioned before risk.
+    const s02Line = roadmap.split("\n").find((line) => line.includes("**S02:"));
+    assert.ok(s02Line, "S02 slice line must exist in roadmap");
+    assert.ok(
+      s02Line.includes("`[sketch]`"),
+      `sketch slice S02 must carry the sketch badge, got: ${s02Line}`,
+    );
+    // Badge sits before risk so it stays visible if the line truncates.
+    const sketchIdx = s02Line.indexOf("`[sketch]`");
+    const riskIdx = s02Line.indexOf("`risk:");
+    assert.ok(
+      sketchIdx >= 0 && riskIdx >= 0 && sketchIdx < riskIdx,
+      "sketch badge must appear before the risk tag",
+    );
+  } finally {
+    cleanup(base);
+  }
+});
+
+test("ROADMAP omits sketch badge when no slices are sketches", async () => {
+  const base = makeTmpBase();
+  try {
+    const params = planMilestoneWithSketches();
+    // Strip the sketch designation from S02 so both slices are fully planned.
+    params.slices[1] = {
+      ...params.slices[1],
+      isSketch: false,
+      successCriteria: "Demo behavior works.",
+      proofLevel: "unit" as const,
+      integrationClosure: "S02 closes the demo behavior.",
+      observabilityImpact: "No new telemetry.",
+    };
+
+    const result = await handlePlanMilestone(params, base);
+    if ("error" in result) {
+      assert.fail(`handlePlanMilestone failed: ${result.error}`);
+    }
+
+    const roadmap = readFileSync(
+      join(base, ".gsd", "milestones", "M001", "M001-ROADMAP.md"),
+      "utf-8",
+    );
+
+    assert.equal(
+      roadmap.includes("`[sketch]`"),
+      false,
+      "roadmap must not carry the sketch badge when no slice is a sketch",
+    );
+  } finally {
+    cleanup(base);
+  }
+});

package/src/resources/extensions/gsd/tests/plan-milestone.test.ts

@@ -116,6 +116,32 @@ test('handlePlanMilestone rejects invalid payloads', async () => {
   }
 });
 
+test('handlePlanMilestone rejects delimiter characters in milestone and slice titles', async () => {
+  const base = makeTmpBase();
+  const dbPath = join(base, '.gsd', 'gsd.db');
+  openDatabase(dbPath);
+
+  try {
+    const milestoneResult = await handlePlanMilestone({ ...validParams(), title: 'Client/Server split' }, base);
+    assert.ok('error' in milestoneResult);
+    assert.match(milestoneResult.error, /validation failed: title is invalid: .*forward slash/);
+    assert.equal(getMilestone('M001'), null, 'invalid milestone title must not persist');
+
+    const sliceResult = await handlePlanMilestone({
+      ...validParams(),
+      slices: [
+        validParams().slices[0],
+        { ...validParams().slices[1], title: 'Client/Server migration' },
+      ],
+    }, base);
+    assert.ok('error' in sliceResult);
+    assert.match(sliceResult.error, /validation failed: slices\[1\]\.title is invalid: .*forward slash/);
+    assert.equal(getMilestoneSlices('M001').length, 0, 'invalid slice title must not persist partial roadmap state');
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanMilestone surfaces render failures and does not clear parse-visible state on failure', async () => {
   const base = makeTmpBase();
   const dbPath = join(base, '.gsd', 'gsd.db');

package/src/resources/extensions/gsd/tests/plan-slice-prompt.test.ts

@@ -65,7 +65,9 @@ test("plan-slice prompt: DB-backed tool names survive template substitution", ()
   const result = loadPrompt("plan-slice", { ...BASE_VARS, commitInstruction: "Do not commit." });
   assert.ok(result.includes("gsd_plan_slice"), "gsd_plan_slice should appear in rendered prompt");
   assert.ok(result.includes("gsd_plan_task"), "gsd_plan_task should appear in rendered prompt");
+  assert.ok(result.includes("gsd_decision_save"), "structural decisions should use DB-backed decision tool");
   assert.ok(result.includes("canonical write path"), "canonical write path language should survive substitution");
+  assert.doesNotMatch(result, /append them to `.gsd\/DECISIONS\.md`/);
 });
 
 test("plan-slice prompt: compact planning gates survive template substitution", () => {

package/src/resources/extensions/gsd/tests/plan-slice.test.ts

@@ -6,7 +6,7 @@ import { mkdtempSync, mkdirSync, rmSync, readFileSync, existsSync, writeFileSync
 import { join } from 'node:path';
 import { tmpdir } from 'node:os';
 
-import { openDatabase, closeDatabase, insertMilestone, insertSlice, getSlice, getSliceTasks, getTask } from '../gsd-db.ts';
+import { openDatabase, closeDatabase, insertMilestone, insertSlice, getSlice, getSliceTasks, getTask, getGateResults, updateTaskStatus } from '../gsd-db.ts';
 import { handlePlanSlice } from '../tools/plan-slice.ts';
 import { parsePlan } from '../parsers-legacy.ts';
 import { parseTaskPlanFile } from '../files.ts';
@@ -15,6 +15,10 @@ import { deriveState, invalidateStateCache } from '../state.ts';
 function makeTmpBase(): string {
   const base = mkdtempSync(join(tmpdir(), 'gsd-plan-slice-'));
   mkdirSync(join(base, '.gsd', 'milestones', 'M001', 'slices', 'S02', 'tasks'), { recursive: true });
+  mkdirSync(join(base, 'src', 'resources', 'extensions', 'gsd', 'tools'), { recursive: true });
+  writeFileSync(join(base, 'src', 'resources', 'extensions', 'gsd', 'tools', 'plan-milestone.ts'), '// fixture\n', 'utf-8');
+  writeFileSync(join(base, 'src', 'resources', 'extensions', 'gsd', 'tools', 'plan-task.ts'), '// fixture\n', 'utf-8');
+  writeFileSync(join(base, 'stale-input.py'), '# fixture\n', 'utf-8');
   return base;
 }
 
@@ -125,6 +129,31 @@ test('handlePlanSlice advances DB-derived state out of planning immediately', as
   }
 });
 
+test('handlePlanSlice clears sketch flag so DB-derived state leaves refining', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    insertMilestone({ id: 'M001', title: 'Milestone', status: 'active' });
+    insertSlice({ id: 'S02', milestoneId: 'M001', title: 'Planning slice', status: 'pending', demo: 'Rendered plans exist.', isSketch: true });
+
+    invalidateStateCache();
+    const before = await deriveState(base);
+    assert.equal(before.phase, 'refining');
+
+    const result = await handlePlanSlice(validParams(), base);
+    assert.ok(!('error' in result), `unexpected error: ${'error' in result ? result.error : ''}`);
+    assert.equal(getSlice('M001', 'S02')?.is_sketch, 0, 'planned slice must no longer be treated as a sketch');
+
+    invalidateStateCache();
+    const after = await deriveState(base);
+    assert.notEqual(after.phase, 'refining');
+    assert.equal(after.progress?.tasks?.total, 2);
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanSlice leaves omitted enrichment fields empty instead of rendering placeholders', async () => {
   const base = makeTmpBase();
   openDatabase(join(base, '.gsd', 'gsd.db'));
@@ -172,6 +201,28 @@ test('handlePlanSlice rejects invalid payloads', async () => {
   }
 });
 
+test('handlePlanSlice explains string task IO fields must be arrays', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+    const result = await handlePlanSlice({
+      ...validParams(),
+      tasks: [
+        {
+          ...validParams().tasks[0],
+          inputs: 'src/index.ts' as unknown as string[],
+        },
+      ],
+    }, base);
+    assert.ok('error' in result);
+    assert.match(result.error, /validation failed: tasks\[0\]\.inputs must be an array of strings, not string/);
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanSlice rejects absolute task IO paths outside the active worktree', async () => {
   const base = makeTmpBase();
   openDatabase(join(base, '.gsd', 'gsd.db'));
@@ -198,6 +249,63 @@ test('handlePlanSlice rejects absolute task IO paths outside the active worktree
   }
 });
 
+test('handlePlanSlice rejects missing task input paths before persisting tasks', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+    const result = await handlePlanSlice({
+      ...validParams(),
+      tasks: [
+        {
+          ...validParams().tasks[0],
+          inputs: ['fixtures/missing-source.json'],
+        },
+      ],
+    }, base);
+
+    assert.ok('error' in result);
+    assert.match(result.error, /pre-execution validation failed:/);
+    assert.match(result.error, /fixtures\/missing-source\.json/);
+    assert.equal(getSliceTasks('M001', 'S02').length, 0, 'invalid planning IO must not persist tasks');
+  } finally {
+    cleanup(base);
+  }
+});
+
+test('handlePlanSlice rejects task input paths created by later tasks before persisting tasks', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+    const params = validParams();
+    const result = await handlePlanSlice({
+      ...params,
+      tasks: [
+        {
+          ...params.tasks[0],
+          inputs: ['generated/report.json'],
+          expectedOutput: ['generated/summary.json'],
+        },
+        {
+          ...params.tasks[1],
+          inputs: [],
+          expectedOutput: ['generated/report.json'],
+        },
+      ],
+    }, base);
+
+    assert.ok('error' in result);
+    assert.match(result.error, /pre-execution validation failed:/);
+    assert.match(result.error, /sequence violation/);
+    assert.equal(getSliceTasks('M001', 'S02').length, 0, 'invalid task ordering must not persist tasks');
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanSlice accepts absolute task IO paths inside the active worktree', async () => {
   const base = makeTmpBase();
   openDatabase(join(base, '.gsd', 'gsd.db'));
@@ -304,3 +412,119 @@ test('handlePlanSlice reruns idempotently and refreshes parse-visible state', as
     cleanup(base);
   }
 });
+
+test('handlePlanSlice removes omitted pending tasks when replanning a smaller task set', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+    const fourTaskPlan = {
+      ...validParams(),
+      tasks: [
+        ...validParams().tasks,
+        { ...validParams().tasks[0], taskId: 'T03', title: 'Third task' },
+        { ...validParams().tasks[0], taskId: 'T04', title: 'Stale task', inputs: ['stale-input.py'] },
+      ],
+    };
+
+    const first = await handlePlanSlice(fourTaskPlan, base);
+    assert.ok(!('error' in first), `unexpected error: ${'error' in first ? first.error : ''}`);
+    const staleTaskPlanPath = join(base, '.gsd', 'milestones', 'M001', 'slices', 'S02', 'tasks', 'T04-PLAN.md');
+    assert.ok(existsSync(staleTaskPlanPath), 'initial plan should render T04');
+
+    const second = await handlePlanSlice({
+      ...validParams(),
+      tasks: fourTaskPlan.tasks.filter((task) => task.taskId !== 'T04'),
+    }, base);
+    assert.ok(!('error' in second), `unexpected error: ${'error' in second ? second.error : ''}`);
+
+    assert.deepEqual(getSliceTasks('M001', 'S02').map((task) => task.id), ['T01', 'T02', 'T03']);
+    assert.equal(getGateResults('M001', 'S02', 'task').some((gate) => gate.task_id === 'T04'), false);
+    assert.equal(existsSync(staleTaskPlanPath), false, 'omitted task plan artifact should be removed');
+  } finally {
+    cleanup(base);
+  }
+});
+
+test('handlePlanSlice rejects omitted completed tasks without changing slice or task state', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+    const fourTaskPlan = {
+      ...validParams(),
+      tasks: [
+        ...validParams().tasks,
+        { ...validParams().tasks[0], taskId: 'T03', title: 'Third task' },
+        { ...validParams().tasks[0], taskId: 'T04', title: 'Stale task', inputs: ['stale-input.py'] },
+      ],
+    };
+
+    const first = await handlePlanSlice(fourTaskPlan, base);
+    assert.ok(!('error' in first), `unexpected error: ${'error' in first ? first.error : ''}`);
+    const staleTaskPlanPath = join(base, '.gsd', 'milestones', 'M001', 'slices', 'S02', 'tasks', 'T04-PLAN.md');
+    assert.ok(existsSync(staleTaskPlanPath), 'initial plan should render T04');
+
+    updateTaskStatus('M001', 'S02', 'T04', 'complete', '2026-05-12T00:00:00.000Z');
+    const tasksBefore = getSliceTasks('M001', 'S02');
+    const gatesBefore = getGateResults('M001', 'S02', 'task');
+
+    const second = await handlePlanSlice({
+      ...validParams(),
+      goal: 'Rejected replan should not persist.',
+      tasks: fourTaskPlan.tasks.filter((task) => task.taskId !== 'T04'),
+    }, base);
+    assert.deepEqual(second, { error: 'cannot remove completed task T04' });
+
+    assert.equal(getSlice('M001', 'S02')?.goal, 'Persist slice planning through the DB.');
+    assert.deepEqual(getSliceTasks('M001', 'S02'), tasksBefore);
+    assert.deepEqual(getGateResults('M001', 'S02', 'task'), gatesBefore);
+    assert.ok(existsSync(staleTaskPlanPath), 'completed task plan artifact should remain after rejected replan');
+  } finally {
+    cleanup(base);
+  }
+});
+
+test('regression: validateTasks surfaces clean per-field errors for non-array IO inputs', async () => {
+  // Regression for the bug fixed in PR #5872: an earlier refactor on main
+  // (0b0e1a901) re-added validateStringArray() calls inside validateTasks
+  // without re-adding its import. The catch around validateParams swallowed
+  // the ReferenceError into a generic "validation failed: validateStringArray
+  // is not defined" message, so silent runtime breakage was possible.
+  //
+  // Exercise every validateStringArray call site (files, inputs, expectedOutput)
+  // so a future missing-import would surface as a per-field assertion failure
+  // here, not a deep ReferenceError that's easy to mis-diagnose.
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParentSlice();
+
+    for (const field of ['files', 'inputs', 'expectedOutput'] as const) {
+      const result = await handlePlanSlice({
+        ...validParams(),
+        tasks: [{
+          ...validParams().tasks[0],
+          [field]: 'not-an-array' as unknown as string[],
+        }],
+      }, base);
+      assert.ok('error' in result, `${field}: expected validation error, got success`);
+      assert.match(
+        result.error,
+        new RegExp(`tasks\\[0\\]\\.${field} must be an array`),
+        `${field}: expected per-field validation message, got: ${result.error}`,
+      );
+      assert.doesNotMatch(
+        result.error,
+        /is not defined/,
+        `${field}: validation surfaced ReferenceError — likely a missing import in plan-slice.ts`,
+      );
+      assert.equal(getSliceTasks('M001', 'S02').length, 0, `${field}: invalid input must not persist`);
+    }
+  } finally {
+    cleanup(base);
+  }
+});

package/src/resources/extensions/gsd/tests/plan-task.test.ts

@@ -79,6 +79,23 @@ test('handlePlanTask rejects invalid payloads', async () => {
   }
 });
 
+test('handlePlanTask explains string IO fields must be arrays', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParent();
+    const result = await handlePlanTask({
+      ...validParams(),
+      expectedOutput: 'src/output.ts' as unknown as string[],
+    }, base);
+    assert.ok('error' in result);
+    assert.match(result.error, /validation failed: expectedOutput must be an array of strings, not string/);
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanTask rejects absolute task IO paths outside the active worktree', async () => {
   const base = makeTmpBase();
   openDatabase(join(base, '.gsd', 'gsd.db'));

package/src/resources/extensions/gsd/tests/post-exec-retry-bypass.test.ts

@@ -1,3 +1,5 @@
+// Project/App: GSD-2
+// File Purpose: Tests for post-execution retry bypass and verification gate failure handling.
 /**
  * post-exec-retry-bypass.test.ts — Tests for post-execution blocking failure retry bypass.
  *
@@ -9,7 +11,7 @@
 import { describe, test, mock, beforeEach, afterEach } from "node:test";
 import assert from "node:assert/strict";
 import { tmpdir } from "node:os";
-import { mkdirSync, writeFileSync, rmSync, existsSync } from "node:fs";
+import { mkdirSync, readFileSync, writeFileSync, rmSync } from "node:fs";
 import { join } from "node:path";
 
 import { runPostUnitVerification, type VerificationContext } from "../auto-verification.ts";
@@ -71,6 +73,7 @@ function setupTestEnvironment(): void {
   mkdirSync(milestonesDir, { recursive: true });
 
   process.chdir(tempDir);
+  invalidateAllCaches();
   _clearGsdRootCache();
 
   dbPath = join(gsdDir, "gsd.db");
@@ -140,6 +143,34 @@ function createBasicTask(): void {
   });
 }
 
+function createTaskWithoutVerify(): void {
+  insertMilestone({ id: "M001" });
+  insertSlice({
+    id: "S01",
+    milestoneId: "M001",
+    title: "Test Slice",
+    risk: "low",
+  });
+
+  insertTask({
+    id: "T01",
+    sliceId: "S01",
+    milestoneId: "M001",
+    title: "Task without host verification",
+    status: "pending",
+    planning: {
+      description: "Task intentionally missing runnable verification",
+      estimate: "1h",
+      files: [],
+      verify: "",
+      inputs: [],
+      expectedOutput: [],
+      observabilityImpact: "",
+    },
+    sequence: 0,
+  });
+}
+
 function createPostExecFailureTask(): void {
   insertMilestone({ id: "M001" });
   insertSlice({
@@ -250,14 +281,14 @@ describe("Post-execution blocking failure retry bypass", () => {
     const s = makeMockSession(tempDir, { type: "execute-task", id: "M001/S01/T01" });
     
     // Pre-set some retry state
-    s.verificationRetryCount.set("M001/S01/T01", 2);
+    s.verificationRetryCount.set("execute-task:M001/S01/T01", 2);
 
     const vctx: VerificationContext = { s, ctx, pi };
     const result = await runPostUnitVerification(vctx, pauseAutoMock);
 
     // On success, retry count should be cleared
     assert.equal(result, "continue");
-    assert.equal(s.verificationRetryCount.has("M001/S01/T01"), false);
+    assert.equal(s.verificationRetryCount.has("execute-task:M001/S01/T01"), false);
   });
 
   test("post-exec failure notification mentions cross-task consistency", async () => {
@@ -327,6 +358,53 @@ describe("Post-execution blocking failure retry bypass", () => {
     assert.equal(row?.outcome, "fail");
     assert.equal(row?.failure_class, "artifact");
   });
+
+  test("execute-task with no host-owned verification pauses fail-closed", async () => {
+    createTaskWithoutVerify();
+
+    const ctx = makeMockCtx();
+    const pi = makeMockPi();
+    const pauseAutoMock = mock.fn(async () => {});
+    const s = makeMockSession(tempDir, { type: "execute-task", id: "M001/S01/T01" });
+
+    const result = await runPostUnitVerification({ s, ctx, pi }, pauseAutoMock);
+
+    assert.equal(result, "pause");
+    assert.equal(pauseAutoMock.mock.callCount(), 1);
+    assert.equal(s.pendingVerificationRetry, null);
+
+    const evidencePath = join(tempDir, ".gsd", "milestones", "M001", "slices", "S01", "tasks", "T01-VERIFY.json");
+    const evidence = JSON.parse(readFileSync(evidencePath, "utf-8"));
+    assert.equal(evidence.passed, false);
+    assert.equal(evidence.discoverySource, "none");
+    assert.ok(!("retryAttempt" in evidence), "no-host-checks evidence must not include retryAttempt");
+    assert.ok(!("maxRetries" in evidence), "no-host-checks evidence must not include maxRetries");
+  });
+
+  test("auto-discovered package.json verification failure retries instead of continuing", async () => {
+    createTaskWithoutVerify();
+    writeFileSync(
+      join(tempDir, "package.json"),
+      JSON.stringify({ scripts: { test: "exit 1" } }),
+      "utf-8",
+    );
+    writePreferences({
+      verification_auto_fix: true,
+      verification_max_retries: 2,
+    });
+
+    const ctx = makeMockCtx();
+    const pi = makeMockPi();
+    const pauseAutoMock = mock.fn(async () => {});
+    const s = makeMockSession(tempDir, { type: "execute-task", id: "M001/S01/T01" });
+
+    const result = await runPostUnitVerification({ s, ctx, pi }, pauseAutoMock);
+
+    assert.equal(result, "retry");
+    assert.equal(pauseAutoMock.mock.callCount(), 0);
+    assert.equal(s.pendingVerificationRetry?.unitId, "M001/S01/T01");
+    assert.match(s.pendingVerificationRetry?.failureContext ?? "", /npm run test/);
+  });
 });
 
 describe("Post-execution retry behavior", () => {