gsd-pi 2.81.0 → 2.82.0-dev.20138ae42

package/src/resources/extensions/gsd/tests/post-execution-checks.test.ts

@@ -127,6 +127,61 @@ import { b } from './b';
     assert.equal(imports.length, 2);
   });
 
+  test("ignores import-looking string literals in test fixtures", () => {
+    const source = `
+const rewritten = source.replace(
+  'import { normalizeZagrebBusinessDeadline } from "./cutoff";',
+  'const helper = true;'
+);
+
+import realThing from "./real-thing";
+`;
+    const imports = extractRelativeImports(source);
+    assert.deepEqual(imports, [
+      { importPath: "./real-thing", lineNum: 7 },
+    ]);
+  });
+
+  test("ignores import-looking lines inside template literals", () => {
+    const source = [
+      "const fixture = `",
+      "import missingThing from './missing-thing';",
+      "`;",
+      "",
+      "import realThing from './real-thing';",
+    ].join("\n");
+    const imports = extractRelativeImports(source);
+    assert.deepEqual(imports, [
+      { importPath: "./real-thing", lineNum: 5 },
+    ]);
+  });
+
+  test("ignores require() inside string literals", () => {
+    const source = [
+      'const fixture = "const x = require(\'./missing\');";',
+      "const otherFixture = 'const y = require(\"./also-missing\");';",
+      "const real = require('./real');",
+    ].join("\n");
+    const imports = extractRelativeImports(source);
+    assert.deepEqual(imports, [
+      { importPath: "./real", lineNum: 3 },
+    ]);
+  });
+
+  test("ignores require() inside template literals", () => {
+    const source = [
+      "const fixture = `",
+      "const x = require('./missing');",
+      "`;",
+      "",
+      "const real = require('./real');",
+    ].join("\n");
+    const imports = extractRelativeImports(source);
+    assert.deepEqual(imports, [
+      { importPath: "./real", lineNum: 5 },
+    ]);
+  });
+
   test("handles empty source", () => {
     const imports = extractRelativeImports("");
     assert.deepEqual(imports, []);
@@ -810,6 +865,37 @@ describe("runPostExecutionChecks", () => {
     }
   });
 
+  test("does not fail on import-looking strings in task key files", () => {
+    tempDir = join(tmpdir(), `post-exec-test-${Date.now()}`);
+    mkdirSync(tempDir, { recursive: true });
+    mkdirSync(join(tempDir, "tests"), { recursive: true });
+    writeFileSync(join(tempDir, "tests", "real-thing.ts"), "export default true;");
+    writeFileSync(
+      join(tempDir, "tests", "source-verifier.test.ts"),
+      `
+const rewritten = source.replace(
+  'import { normalizeZagrebBusinessDeadline } from "./cutoff";',
+  'const helper = true;'
+);
+
+import realThing from "./real-thing";
+assert.ok(realThing);
+`
+    );
+
+    try {
+      const task = createTask({
+        id: "T03",
+        key_files: ["tests/source-verifier.test.ts"],
+      });
+      const result = runPostExecutionChecks(task, [], tempDir);
+      assert.equal(result.status, "pass");
+      assert.deepEqual(result.checks, []);
+    } finally {
+      rmSync(tempDir, { recursive: true, force: true });
+    }
+  });
+
   test("returns fail status when blocking failure exists", () => {
     tempDir = join(tmpdir(), `post-exec-test-${Date.now()}`);
     mkdirSync(tempDir, { recursive: true });

package/src/resources/extensions/gsd/tests/post-unit-git-failure.test.ts

@@ -11,7 +11,7 @@ const source = readFileSync(
 
 test("postUnitPreVerification blocks on git action failure", () => {
   const failureBlock = extractSourceRegion(source, 'if (gitResult.status === "failed")');
-  assert.ok(failureBlock.includes('ctx.ui.notify(failureMsg, "error")'));
+  assert.ok(failureBlock.includes('ctx.ui.notify(failureMsg, opts?.softFailure ? "warning" : "error")'));
   assert.ok(failureBlock.includes("await pauseAuto(ctx, pi)"));
   assert.ok(failureBlock.includes('return "dispatched"'));
   assert.ok(!failureBlock.includes("git-action-failed-nonblocking"));

package/src/resources/extensions/gsd/tests/pre-execution-checks.test.ts

@@ -9,6 +9,7 @@
  *   2. File path consistency — files exist vs prior expected_output
  *   3. Task ordering — detect impossible read-before-create
  *   4. Interface contracts — contradictory function signatures
+ *   5. Verify commands — reject unsafe or non-runnable task verification
  */
 
 import { describe, test, mock } from "node:test";
@@ -22,6 +23,7 @@ import {
   checkFilePathConsistency,
   checkTaskOrdering,
   checkInterfaceContracts,
+  checkVerificationCommands,
   runPreExecutionChecks,
   normalizeFilePath,
   type PreExecutionResult,
@@ -812,6 +814,33 @@ function process(a: number): number
   });
 });
 
+describe("checkVerificationCommands", () => {
+  test("accepts pipe-free pytest Verify command", () => {
+    const results = checkVerificationCommands([
+      createTask({
+        id: "T01",
+        verify: "python3 -m pytest tests/ -q --tb=short",
+      }),
+    ]);
+
+    assert.deepEqual(results, []);
+  });
+
+  test("rejects piped pytest Verify command", () => {
+    const results = checkVerificationCommands([
+      createTask({
+        id: "T01",
+        verify: "python3 -m pytest tests/ -q --tb=short 2>&1 | tail -5",
+      }),
+    ]);
+
+    assert.equal(results.length, 1);
+    assert.equal(results[0]?.category, "tool");
+    assert.equal(results[0]?.blocking, true);
+    assert.match(results[0]?.message ?? "", /shell control syntax/);
+  });
+});
+
 // ─── runPreExecutionChecks Integration Tests ─────────────────────────────────
 
 describe("runPreExecutionChecks", () => {
@@ -847,6 +876,30 @@ describe("runPreExecutionChecks", () => {
     }
   });
 
+  test("returns fail status for unsafe Verify command before execution", async () => {
+    tempDir = join(tmpdir(), `pre-exec-test-${Date.now()}`);
+    mkdirSync(tempDir, { recursive: true });
+
+    try {
+      const tasks = [
+        createTask({
+          id: "T01",
+          verify: "python3 -m pytest tests/ -q --tb=short 2>&1 | tail -5",
+        }),
+      ];
+
+      const result = await runPreExecutionChecks(tasks, tempDir);
+
+      assert.equal(result.status, "fail");
+      assert.equal(result.checks.length, 1);
+      assert.equal(result.checks[0]?.category, "tool");
+      assert.equal(result.checks[0]?.blocking, true);
+      assert.match(result.checks[0]?.message ?? "", /Unsafe or non-runnable Verify command/);
+    } finally {
+      rmSync(tempDir, { recursive: true, force: true });
+    }
+  });
+
   test("returns fail status when blocking failure exists", async () => {
     tempDir = join(tmpdir(), `pre-exec-test-${Date.now()}`);
     mkdirSync(tempDir, { recursive: true });

package/src/resources/extensions/gsd/tests/progressive-planning.test.ts

@@ -14,9 +14,9 @@ import {
   insertMilestone,
   insertSlice,
   setSliceSketchFlag,
-  autoHealSketchFlags,
   getSlice,
 } from "../gsd-db.ts";
+import { autoHealSketchFlags } from "../state-reconciliation/drift/sketch-flag.ts";
 import { deriveStateFromDb } from "../state.ts";
 import { resolveDispatch } from "../auto-dispatch.ts";
 import type { DispatchContext } from "../auto-dispatch.ts";

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

@@ -55,6 +55,14 @@ test("system prompt hard rules forbid fabricating user responses", () => {
   assert.match(prompt, /ask_user_questions.*only valid structured user input/i);
 });
 
+test("system prompt requires reading before edit or overwrite", () => {
+  const prompt = readPrompt("system");
+  assert.match(prompt, /Read before edit or overwrite/i);
+  assert.match(prompt, /Before any write that creates or replaces a file/i);
+  assert.match(prompt, /confirm whether the path exists; if it does, `read` it first/i);
+  assert.match(prompt, /For truly new files, confirm the path does not already exist/i);
+});
+
 test("discuss prompt allows implementation questions when they materially matter", () => {
   const prompt = readPrompt("discuss");
   assert.match(prompt, /Lead with experience, but ask implementation when it materially matters/i);
@@ -64,6 +72,16 @@ test("discuss prompt allows implementation questions when they materially matter
   assert.doesNotMatch(prompt, /Questions must be about the experience, not the implementation/i);
 });
 
+test("discuss prompt ends milestone planning with next-step handoff", () => {
+  const prompt = readPrompt("discuss");
+  assert.match(prompt, /Next steps:/);
+  assert.match(prompt, /\/gsd auto/);
+  assert.match(prompt, /\/gsd status/);
+  assert.match(prompt, /\/gsd visualize/);
+  assert.match(prompt, /\/gsd notifications/);
+  assert.doesNotMatch(prompt, /nothing else\. Auto-mode will start automatically/);
+});
+
 test("guided discussion prompts avoid wrap-up prompts after every round", () => {
   const milestonePrompt = readPrompt("guided-discuss-milestone");
   const slicePrompt = readPrompt("guided-discuss-slice");
@@ -217,6 +235,17 @@ test("complete-slice prompt does not instruct LLM to toggle checkboxes manually"
   assert.doesNotMatch(prompt, /change \[ \] to \[x\]/);
 });
 
+test("complete-slice prompt keeps source fixes in execution units", () => {
+  const prompt = readPrompt("complete-slice");
+  assert.match(prompt, /Do not use direct `bash` for verification commands/i);
+  assert.match(prompt, /do \*\*not\*\* edit source files in this unit/i);
+  assert.match(prompt, /do \*\*not\*\* call `gsd_slice_complete`/i);
+  assert.match(prompt, /gsd_task_reopen/);
+  assert.match(prompt, /gsd_replan_slice/);
+  assert.match(prompt, /needs execution follow-up/i);
+  assert.doesNotMatch(prompt, /Fix failures before marking done/i);
+});
+
 test("complete-slice prompt instructs writing summary and UAT files before tool call", () => {
   const prompt = readPrompt("complete-slice");
   assert.match(prompt, /\{\{sliceSummaryPath\}\}/);
@@ -363,7 +392,9 @@ test("execute-task prompt uses camelCase parameter names matching TypeBox schema
 test("complete-slice prompt uses camelCase parameter names matching TypeBox schema", () => {
   const prompt = readPrompt("complete-slice");
   // The gsd_complete_slice tool schema uses camelCase: milestoneId, sliceId
-  const toolCallLine = prompt.split("\n").find((l) => /gsd_complete_slice/.test(l) || /gsd_slice_complete/.test(l));
+  const toolCallLine = prompt.split("\n").find((l) =>
+    (/gsd_complete_slice/.test(l) || /gsd_slice_complete/.test(l)) && /milestoneId/.test(l) && /sliceId/.test(l)
+  );
   assert.ok(toolCallLine, "prompt must contain a gsd_complete_slice or gsd_slice_complete tool call line");
   assert.doesNotMatch(toolCallLine!, /milestone_id/, "must use milestoneId, not milestone_id");
   assert.doesNotMatch(toolCallLine!, /slice_id/, "must use sliceId, not slice_id");

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

@@ -0,0 +1,23 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { loadPrompt } from "../prompt-loader.ts";
+
+test("loadPrompt reports missing template variables with balanced braces", () => {
+  assert.throws(
+    () => loadPrompt("guided-discuss-milestone", {
+      milestoneId: "M001",
+      milestoneTitle: "Missing working directory",
+      structuredQuestionsAvailable: "false",
+      fastPathInstruction: "",
+      inlinedTemplates: "context template",
+      commitInstruction: "Do not commit during this test.",
+    }),
+    (error) => {
+      assert.ok(error instanceof Error);
+      assert.match(error.message, /template declares \{\{workingDirectory\}\} but no value was provided/);
+      assert.doesNotMatch(error.message, /\{\{workingDirectory\}\}\}/);
+      return true;
+    },
+  );
+});

package/src/resources/extensions/gsd/tests/provider-errors.test.ts

@@ -10,7 +10,7 @@ import assert from "node:assert/strict";
 import { classifyError, isTransient, isTransientNetworkError } from "../error-classifier.ts";
 import { pauseAutoForProviderError } from "../provider-error-pause.ts";
 import { resumeAutoAfterProviderDelay } from "../bootstrap/provider-error-resume.ts";
-import { MAX_TRANSIENT_AUTO_RESUMES, resetTransientRetryState } from "../bootstrap/agent-end-recovery.ts";
+import { MAX_TRANSIENT_AUTO_RESUMES, isTerminalDeletedWorktreeProviderError, resetTransientRetryState } from "../bootstrap/agent-end-recovery.ts";
 import { _buildCancelledUnitStopReason } from "../auto/phases.ts";
 import { getNextFallbackModel } from "../preferences.ts";
 // Zero-import module — imported by path rather than through the package
@@ -399,6 +399,25 @@ test("pauseAutoForProviderError falls back to indefinite pause when not rate lim
   ]);
 });
 
+test("isTerminalDeletedWorktreeProviderError matches removed auto-worktree paths only", () => {
+  assert.equal(
+    isTerminalDeletedWorktreeProviderError('Path "/Users/dev/.gsd/projects/abc123/worktrees/M005" does not exist'),
+    true,
+  );
+  assert.equal(
+    isTerminalDeletedWorktreeProviderError('Path "/Users/dev/app/.gsd/worktrees/M005" does not exist'),
+    true,
+  );
+  assert.equal(
+    isTerminalDeletedWorktreeProviderError('Path "/Users/dev/app/src/file.ts" does not exist'),
+    false,
+  );
+  assert.equal(
+    isTerminalDeletedWorktreeProviderError('Path "/Users/dev/.gsd/projects/abc123/worktrees/M005" failed with EACCES'),
+    false,
+  );
+});
+
 // ── resumeAutoAfterProviderDelay ────────────────────────────────────────────
 
 test("resumeAutoAfterProviderDelay restarts paused auto-mode from the recorded base path", async () => {

package/src/resources/extensions/gsd/tests/provider-switch-observer.test.ts

@@ -0,0 +1,252 @@
+// GSD-2 — ADR-005 Phase 3b: ProviderSwitchObserver Tests
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, readFileSync, rmSync, existsSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+
+import { setProviderSwitchObserver, type ProviderSwitchReport } from "@gsd/pi-ai";
+
+import { autoSession } from "../auto-runtime-state.ts";
+import {
+  initNotificationStore,
+  readNotifications,
+  _resetNotificationStore,
+} from "../notification-store.ts";
+import {
+  _resetProviderSwitchStats,
+  getProviderSwitchStats,
+  installProviderSwitchObserver,
+  uninstallProviderSwitchObserver,
+} from "../provider-switch-observer.ts";
+
+function makeReport(overrides: Partial<ProviderSwitchReport> = {}): ProviderSwitchReport {
+  return {
+    fromApi: "anthropic-messages",
+    toApi: "openai-responses",
+    thinkingBlocksDropped: 0,
+    thinkingBlocksDowngraded: 0,
+    toolCallIdsRemapped: 0,
+    syntheticToolResultsInserted: 0,
+    thoughtSignaturesDropped: 0,
+    ...overrides,
+  };
+}
+
+function withTempBasePath(): { basePath: string; cleanup: () => void } {
+  const basePath = mkdtempSync(join(tmpdir(), "gsd-provider-switch-"));
+  return {
+    basePath,
+    cleanup: () => rmSync(basePath, { recursive: true, force: true }),
+  };
+}
+
+test.beforeEach(() => {
+  _resetProviderSwitchStats();
+  _resetNotificationStore();
+  autoSession.currentTraceId = null;
+  autoSession.basePath = "";
+});
+
+test.afterEach(() => {
+  uninstallProviderSwitchObserver();
+  _resetNotificationStore();
+  autoSession.currentTraceId = null;
+  autoSession.basePath = "";
+});
+
+test("installProviderSwitchObserver wires the pi-ai observer hook", () => {
+  installProviderSwitchObserver();
+  // Indirect: a second install is a no-op. We verify by checking the counter
+  // increments exactly once when we fire a single report directly into pi-ai's
+  // observer slot.
+  installProviderSwitchObserver();
+
+  // Drive the pi-ai observer directly. The install above pointed it at our
+  // handler; firing here exercises the same code path as a real transform.
+  // We can't reach the installed handler reference from here, so we re-install
+  // a sentinel and confirm setProviderSwitchObserver accepts undefined.
+  setProviderSwitchObserver(undefined);
+  assert.ok(true); // install/uninstall did not throw
+});
+
+test("non-empty report increments the in-memory counter", () => {
+  installProviderSwitchObserver();
+
+  // Reach into the installed handler via setProviderSwitchObserver re-binding.
+  // We snapshot the handler by re-installing, capturing nothing — instead we
+  // use the public API: drive a report through the observer hook directly by
+  // calling setProviderSwitchObserver with a wrapper that proxies into the
+  // installed handler. This is awkward without a separate seam, so we test
+  // the recordReport path end-to-end by firing through pi-ai's transform helper
+  // pattern in a sibling test below. Here we exercise install idempotency only.
+
+  // Fire by setting our own observer that forwards to the module API isn't
+  // possible without exporting handleReport. Instead, verify install state
+  // doesn't crash; the real fire path is covered by the integration test
+  // below that exercises transformMessagesWithReport.
+  assert.deepEqual(getProviderSwitchStats().totalSwitches, 0);
+});
+
+test("end-to-end: transformMessagesWithReport fires the observer and updates stats + notifications", async () => {
+  const { transformMessagesWithReport } = await import("@gsd/pi-ai");
+
+  const { basePath, cleanup } = withTempBasePath();
+  try {
+    initNotificationStore(basePath);
+    installProviderSwitchObserver();
+
+    // Construct a cross-API transform that will drop a redacted thinking block.
+    const targetModel = {
+      id: "gpt-5",
+      name: "GPT-5",
+      api: "openai-responses",
+      provider: "openai",
+      baseUrl: "",
+      reasoning: false,
+      input: ["text"],
+      cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+      contextWindow: 128000,
+      maxTokens: 8192,
+    } as Parameters<typeof transformMessagesWithReport>[1];
+
+    const messages = [
+      {
+        role: "assistant" as const,
+        content: [
+          { type: "thinking" as const, thinking: "", redacted: true },
+          { type: "text" as const, text: "hi" },
+        ],
+        api: "anthropic-messages",
+        provider: "anthropic",
+        model: "claude-sonnet-4-6",
+        usage: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, totalTokens: 0, cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 } },
+        stopReason: "stop" as const,
+        timestamp: Date.now(),
+      },
+    ];
+
+    transformMessagesWithReport(messages as Parameters<typeof transformMessagesWithReport>[0], targetModel, undefined, "anthropic-messages");
+
+    const stats = getProviderSwitchStats();
+    assert.equal(stats.totalSwitches, 1, "non-empty report should bump the counter");
+    assert.ok(stats.totals.thinkingBlocksDropped >= 1, "thinking block drop should be tallied");
+    assert.ok(stats.lastReport, "last report should be retained");
+    assert.equal(stats.lastReport?.fromApi, "anthropic-messages");
+    assert.equal(stats.lastReport?.toApi, "openai-responses");
+
+    // Notification persistence — the observer is interactive (no traceId), so
+    // the byTrace key falls back to "interactive".
+    assert.ok("interactive" in stats.byTrace, "interactive trace bucket should exist");
+    assert.equal(stats.byTrace.interactive?.switches, 1);
+
+    const notifications = readNotifications(basePath);
+    const switchNotifs = notifications.filter((n) => n.message.includes("Provider switch"));
+    assert.ok(switchNotifs.length >= 1, "a provider-switch notification should be persisted");
+    assert.equal(switchNotifs[0]?.severity, "warning");
+  } finally {
+    cleanup();
+  }
+});
+
+test("end-to-end: audit event is emitted when an auto trace is active", async () => {
+  const { transformMessagesWithReport } = await import("@gsd/pi-ai");
+
+  const { basePath, cleanup } = withTempBasePath();
+  try {
+    initNotificationStore(basePath);
+    installProviderSwitchObserver();
+    autoSession.basePath = basePath;
+    autoSession.currentTraceId = "trace-provider-switch-1";
+
+    const targetModel = {
+      id: "gpt-5",
+      name: "GPT-5",
+      api: "openai-responses",
+      provider: "openai",
+      baseUrl: "",
+      reasoning: false,
+      input: ["text"],
+      cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+      contextWindow: 128000,
+      maxTokens: 8192,
+    } as Parameters<typeof transformMessagesWithReport>[1];
+
+    const messages = [
+      {
+        role: "assistant" as const,
+        content: [
+          { type: "thinking" as const, thinking: "", redacted: true },
+          { type: "text" as const, text: "hi" },
+        ],
+        api: "anthropic-messages",
+        provider: "anthropic",
+        model: "claude-sonnet-4-6",
+        usage: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, totalTokens: 0, cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0, total: 0 } },
+        stopReason: "stop" as const,
+        timestamp: Date.now(),
+      },
+    ];
+
+    transformMessagesWithReport(messages as Parameters<typeof transformMessagesWithReport>[0], targetModel, undefined, "anthropic-messages");
+
+    const auditLogPath = join(basePath, ".gsd", "audit", "events.jsonl");
+    assert.ok(existsSync(auditLogPath), "audit events file should be created");
+    const auditLines = readFileSync(auditLogPath, "utf-8")
+      .trim()
+      .split("\n")
+      .map((line) => JSON.parse(line) as { category: string; type: string; traceId: string; payload: Record<string, unknown> });
+    const switchEvent = auditLines.find((e) => e.category === "model-policy" && e.type === "provider-switch");
+    assert.ok(switchEvent, "a model-policy/provider-switch audit event should be present");
+    assert.equal(switchEvent?.traceId, "trace-provider-switch-1");
+    assert.equal(switchEvent?.payload.fromApi, "anthropic-messages");
+    assert.equal(switchEvent?.payload.toApi, "openai-responses");
+
+    const stats = getProviderSwitchStats();
+    assert.ok("trace-provider-switch-1" in stats.byTrace, "trace-keyed bucket should be populated");
+    assert.equal(stats.byTrace["trace-provider-switch-1"]?.switches, 1);
+  } finally {
+    cleanup();
+  }
+});
+
+test("empty report does not bump counter or emit a notification", async () => {
+  const { transformMessagesWithReport } = await import("@gsd/pi-ai");
+
+  const { basePath, cleanup } = withTempBasePath();
+  try {
+    initNotificationStore(basePath);
+    installProviderSwitchObserver();
+
+    // Same-API transform → no transformations → empty report.
+    const sameApiModel = {
+      id: "claude-sonnet-4-6",
+      name: "Claude Sonnet 4.6",
+      api: "anthropic-messages",
+      provider: "anthropic",
+      baseUrl: "",
+      reasoning: false,
+      input: ["text"],
+      cost: { input: 0, output: 0, cacheRead: 0, cacheWrite: 0 },
+      contextWindow: 200000,
+      maxTokens: 8192,
+    } as Parameters<typeof transformMessagesWithReport>[1];
+
+    transformMessagesWithReport(
+      [
+        {
+          role: "user" as const,
+          content: "plain text — no transforms triggered",
+        },
+      ] as Parameters<typeof transformMessagesWithReport>[0],
+      sameApiModel,
+      undefined,
+      "anthropic-messages",
+    );
+
+    assert.equal(getProviderSwitchStats().totalSwitches, 0);
+    assert.equal(readNotifications(basePath).length, 0);
+  } finally {
+    cleanup();
+  }
+});

package/src/resources/extensions/gsd/tests/remediation-completion-guard.test.ts

@@ -1,6 +1,6 @@
 /**
- * Regression test for #2675: completing-milestone dispatch rule must
- * block completion when VALIDATION verdict is "needs-remediation".
+ * Regression tests for non-passing VALIDATION verdicts: completing-milestone
+ * dispatch must block completion when VALIDATION needs remediation or attention.
  *
  * Without this guard, needs-remediation + allSlicesDone causes a loop:
  * complete-milestone dispatched → agent refuses (correct) → no SUMMARY
@@ -66,6 +66,50 @@ test("completing-milestone blocks when VALIDATION verdict is needs-remediation (
   }
 });
 
+test("completing-milestone blocks when VALIDATION verdict is needs-attention (#5747)", async () => {
+  const base = mkdtempSync(join(tmpdir(), "gsd-attention-"));
+  mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });
+
+  try {
+    writeFileSync(
+      join(base, ".gsd", "milestones", "M001", "M001-VALIDATION.md"),
+      [
+        "---",
+        "verdict: needs-attention",
+        "remediation_round: 0",
+        "---",
+        "",
+        "# Validation Report",
+        "",
+        "Acceptance proof is incomplete and needs human attention.",
+      ].join("\n"),
+    );
+
+    const ctx = {
+      mid: "M001",
+      midTitle: "Test Milestone",
+      basePath: base,
+      state: { phase: "completing-milestone" } as any,
+      prefs: {} as any,
+      session: undefined,
+    };
+
+    const result = await completingRule!.match(ctx);
+
+    assert.ok(result !== null, "rule should match");
+    assert.equal(result!.action, "stop", "should return stop action");
+    if (result!.action === "stop") {
+      assert.equal(result!.level, "warning", "should be warning level (pausable)");
+      assert.ok(
+        result!.reason.includes("needs-attention"),
+        "reason should mention needs-attention",
+      );
+    }
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
 test("completing-milestone proceeds normally when VALIDATION verdict is pass (#2675 guard)", async () => {
   const base = mkdtempSync(join(tmpdir(), "gsd-remediation-"));
   mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });

package/src/resources/extensions/gsd/tests/resume-dispatch-worktree.test.ts

@@ -10,7 +10,11 @@ import { join } from "node:path";
 import { tmpdir } from "node:os";
 import { randomUUID } from "node:crypto";
 
-import { _resolvePausedResumeBasePathForTest } from "../auto.ts";
+// ADR-016 phase 2 / B3 (#5621): the legacy `resolvePausedResumeBasePath`
+// helper was retired and folded into `WorktreeLifecycle.resumeFromPausedSession`.
+// The pure path-resolution function lives in `worktree-lifecycle.ts` for tests
+// that exercise the path-resolution invariant without constructing a session.
+import { resolvePausedResumeBasePath } from "../worktree-lifecycle.ts";
 
 function makeTmpBase(): string {
   const base = join(tmpdir(), `gsd-resume-wt-${randomUUID()}`);
@@ -59,7 +63,7 @@ test("resume base path uses paused-session worktreePath when the worktree exists
   try {
     setupWorktreeOnDisk(wt);
     assert.equal(
-      _resolvePausedResumeBasePathForTest(base, wt),
+      resolvePausedResumeBasePath(base, wt),
       wt,
     );
   } finally {
@@ -72,7 +76,7 @@ test("resume base path falls back to project root when paused worktree is missin
   const wt = makeWorktreePath(base, "M001-test");
   try {
     assert.equal(
-      _resolvePausedResumeBasePathForTest(base, wt),
+      resolvePausedResumeBasePath(base, wt),
       base,
     );
   } finally {