gsd-pi 2.80.0-dev.c5f2443b3 → 2.80.0-dev.e146beb20

package/src/resources/extensions/gsd/tests/pr-evidence-hardening.test.ts

@@ -0,0 +1,165 @@
+// Project/App: GSD-2
+// File Purpose: Hardening tests for buildPrEvidence — HTML-comment stripping, fake commit-trailer removal, and per-item length capping.
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import { buildPrEvidence, type PrEvidenceInput } from "../pr-evidence.ts";
+
+test("pr-evidence hardening: strips HTML comments from summaries", () => {
+  const evidence = buildPrEvidence({
+    milestoneId: "M001",
+    summaries: ["visible<!-- hidden secret -->tail"],
+  });
+  assert.ok(!evidence.body.includes("<!--"), "raw <!-- must not appear");
+  assert.ok(!evidence.body.includes("hidden secret"), "comment contents must be stripped");
+  assert.ok(evidence.body.includes("visibletail"), "non-comment text must remain");
+});
+
+test("pr-evidence hardening: removes Co-Authored-By trailer from why", () => {
+  const evidence = buildPrEvidence({
+    milestoneId: "M001",
+    why: "Real reason here.\nCo-Authored-By: Evil <e@evil.com>\nMore reason.",
+  });
+  assert.ok(!evidence.body.includes("Evil <e@evil.com>"));
+  assert.ok(!/Co-Authored-By:/i.test(evidence.body));
+  assert.ok(evidence.body.includes("Real reason here."));
+  assert.ok(evidence.body.includes("More reason."));
+});
+
+test("pr-evidence hardening: removes Signed-off-by trailer from how", () => {
+  const evidence = buildPrEvidence({
+    milestoneId: "M001",
+    how: "Step one.\nSigned-off-by: Forged <f@x.com>\nStep two.",
+  });
+  assert.ok(!evidence.body.includes("Forged <f@x.com>"));
+  assert.ok(!/Signed-off-by:/i.test(evidence.body));
+  assert.ok(evidence.body.includes("Step one."));
+  assert.ok(evidence.body.includes("Step two."));
+});
+
+test("pr-evidence hardening: trailer-name match is case-insensitive", () => {
+  const evidence = buildPrEvidence({
+    milestoneId: "M001",
+    why: "ok\nco-authored-by: lower <l@l.com>\nSIGNED-OFF-BY: upper <u@u.com>\nend",
+  });
+  assert.ok(!evidence.body.includes("lower <l@l.com>"));
+  assert.ok(!evidence.body.includes("upper <u@u.com>"));
+  assert.ok(evidence.body.includes("ok"));
+  assert.ok(evidence.body.includes("end"));
+});
+
+test("pr-evidence hardening: caps oversize summaries item with truncation suffix", () => {
+  const big = "A".repeat(5 * 1024); // 5 KB
+  const evidence = buildPrEvidence({
+    milestoneId: "M001",
+    summaries: [big],
+  });
+  // Find the truncated A-block in the body and assert it is bounded.
+  const lines = evidence.body.split("\n");
+  const longLine = lines.find((l) => l.startsWith("AAAA"));
+  assert.ok(longLine, "expected truncated A-line in body");
+  assert.ok(longLine!.endsWith(" … [truncated]"), "must end with truncation suffix");
+  assert.ok(
+    Buffer.byteLength(longLine!, "utf8") <= 2048,
+    `truncated item must be within 2 KB cap, got ${Buffer.byteLength(longLine!, "utf8")}`,
+  );
+});
+
+test("pr-evidence hardening: HTML comment split across summary items is preserved literally", () => {
+  // Documented behavior: each item is sanitized independently. A comment
+  // that begins in one item and closes in the next is NOT joined, so the
+  // open/close markers remain as literal text. This is intentional — joining
+  // items before sanitizing would let an attacker straddle items to inject
+  // an aligned comment that hides the second item from rendered view.
+  const evidence = buildPrEvidence({
+    milestoneId: "M001",
+    summaries: ["first item ends <!--", "--> second item begins"],
+  });
+  // The literal markers survive because each item was sanitized alone.
+  assert.ok(evidence.body.includes("<!--"), "open marker preserved as literal");
+  assert.ok(evidence.body.includes("-->"), "close marker preserved as literal");
+  assert.ok(evidence.body.includes("first item ends"));
+  assert.ok(evidence.body.includes("second item begins"));
+});
+
+test("pr-evidence hardening: clean input is byte-identical to pre-hardening output", () => {
+  // This test is the contract that protects the golden fixtures: the
+  // sanitizer must be a true no-op for well-formed input. If this fails,
+  // there is a bug in the sanitizer (not in the goldens).
+  const cleanInput: PrEvidenceInput = {
+    milestoneId: "M001",
+    milestoneTitle: "Authentication",
+    changeType: "feat",
+    linkedIssue: "Closes #123",
+    summaries: ["### S01\nImplemented login flow."],
+    blockers: ["Awaiting design review"],
+    roadmapItems: ["- [x] **S01: Login**"],
+    metrics: ["**Units executed:** 3"],
+    testsRun: ["npm test", "npm run typecheck:extensions"],
+    why: "Users need to authenticate before accessing protected resources.",
+    how: "Added password hash check and session token issuance.",
+    rollbackNotes: ["Revert the merge commit."],
+  };
+
+  const expected = [
+    "## TL;DR",
+    "",
+    "**What:** Ship milestone M001 - Authentication",
+    "**Why:** Users need to authenticate before accessing protected resources.",
+    "**How:** Added password hash check and session token issuance.",
+    "",
+    "## What",
+    "",
+    "### S01\nImplemented login flow.",
+    "",
+    "## Blockers",
+    "",
+    "- Awaiting design review",
+    "",
+    "## Why",
+    "",
+    "Users need to authenticate before accessing protected resources.",
+    "",
+    "## How",
+    "",
+    "Added password hash check and session token issuance.",
+    "",
+    "## Linked Issue",
+    "",
+    "Closes #123",
+    "",
+    "## Roadmap",
+    "",
+    "- [x] **S01: Login**",
+    "",
+    "## Metrics",
+    "",
+    "- **Units executed:** 3",
+    "",
+    "## Tests Run",
+    "",
+    "- npm test",
+    "- npm run typecheck:extensions",
+    "",
+    "## Change Type",
+    "",
+    "- [x] `feat` - New feature or capability",
+    "- [ ] `fix` - Bug fix",
+    "- [ ] `refactor` - Code restructuring",
+    "- [ ] `test` - Adding or updating tests",
+    "- [ ] `docs` - Documentation only",
+    "- [ ] `chore` - Build, CI, or tooling changes",
+    "",
+    "## Rollback And Compatibility",
+    "",
+    "- Revert the merge commit.",
+    "",
+    "## AI Assistance Disclosure",
+    "",
+    "This PR was prepared with AI assistance.",
+  ].join("\n");
+
+  const actual = buildPrEvidence(cleanInput).body;
+  assert.equal(actual, expected, "clean input must produce byte-identical output (sanitizer is no-op)");
+});

package/src/resources/extensions/gsd/tests/right-sized-workflow-prompts.test.ts

@@ -0,0 +1,192 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { execFileSync } from "node:child_process";
+import { mkdtempSync, mkdirSync, readFileSync, rmSync, writeFileSync } from "node:fs";
+import { join } from "node:path";
+import { tmpdir } from "node:os";
+
+import { buildCompleteMilestonePrompt, buildPlanMilestonePrompt } from "../auto-prompts.ts";
+
+function git(cwd: string, args: string[]): string {
+  return execFileSync("git", args, {
+    cwd,
+    stdio: ["ignore", "pipe", "pipe"],
+    encoding: "utf-8",
+    env: { ...process.env, GIT_AUTHOR_NAME: "Test User", GIT_AUTHOR_EMAIL: "test@example.com", GIT_COMMITTER_NAME: "Test User", GIT_COMMITTER_EMAIL: "test@example.com" },
+  }).trim();
+}
+
+function makeRepo(files: Record<string, string>): string {
+  const base = mkdtempSync(join(tmpdir(), "gsd-right-size-"));
+  git(base, ["init", "-b", "main"]);
+  mkdirSync(join(base, ".gsd", "milestones", "M001"), { recursive: true });
+  writeFileSync(join(base, ".gsd", "milestones", "M001", "M001-CONTEXT.md"), "# Context\n\nTest milestone.");
+  for (const [path, content] of Object.entries(files)) {
+    const abs = join(base, path);
+    mkdirSync(join(abs, ".."), { recursive: true });
+    writeFileSync(abs, content);
+  }
+  git(base, ["add", "."]);
+  git(base, ["commit", "-m", "init"]);
+  return base;
+}
+
+function writeCompleteMilestoneFiles(base: string, validation: string): void {
+  const dir = join(base, ".gsd", "milestones", "M001");
+  mkdirSync(join(dir, "slices", "S01"), { recursive: true });
+  writeFileSync(join(dir, "M001-ROADMAP.md"), "# M001\n\n## Slices\n- [x] **S01: One** `risk:low` `depends:[]`\n  > Done\n");
+  writeFileSync(join(dir, "M001-VALIDATION.md"), validation);
+  writeFileSync(join(dir, "slices", "S01", "S01-SUMMARY.md"), "# S01 Summary\n\n**Verification:** passed\n");
+}
+
+function validationMetadata(): string {
+  return [
+    "validation_metadata:",
+    "  covered_artifacts:",
+    "    - `.gsd/milestones/M001/M001-VALIDATION.md`",
+    "    - `.gsd/milestones/M001/M001-ROADMAP.md`",
+    "    - `.gsd/milestones/M001/slices/S01/S01-SUMMARY.md`",
+  ].join("\n");
+}
+
+test("plan-milestone prompt includes tiny untyped project classification and one-slice guidance", async () => {
+  const base = makeRepo({ "index.html": "<!doctype html>\n<title>Test</title>\n" });
+  try {
+    const prompt = await buildPlanMilestonePrompt("M001", "Polish static page", base, "minimal");
+    assert.match(prompt, /\*\*Kind:\*\* untyped-existing/);
+    assert.match(prompt, /\*\*Content files:\*\* 1/);
+    assert.match(prompt, /`index\.html`/);
+    assert.match(prompt, /Prefer exactly one slice/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("plan-milestone prompt includes small untyped project 1-2 slice guidance", async () => {
+  const base = makeRepo({
+    "index.html": "html",
+    "README.md": "readme",
+    "styles.css": "body {}",
+  });
+  try {
+    const prompt = await buildPlanMilestonePrompt("M001", "Polish static files", base, "minimal");
+    assert.match(prompt, /\*\*Kind:\*\* untyped-existing/);
+    assert.match(prompt, /\*\*Content files:\*\* 3/);
+    assert.match(prompt, /Prefer 1-2 slices/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("plan-milestone prompt keeps normal guidance for typed projects", async () => {
+  const base = makeRepo({
+    "package.json": "{\"scripts\":{\"test\":\"node --test\"}}\n",
+    "src/index.js": "console.log('ok');\n",
+  });
+  try {
+    const prompt = await buildPlanMilestonePrompt("M001", "Update app", base, "minimal");
+    assert.match(prompt, /\*\*Kind:\*\* typed-existing/);
+    assert.match(prompt, /Use normal ecosystem-aware planning guidance/);
+    assert.doesNotMatch(prompt, /Prefer exactly one slice/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("workflow docs no longer contain blanket 4-10 slice guidance", () => {
+  const docs = readFileSync(join(process.cwd(), "src", "resources", "GSD-WORKFLOW.md"), "utf-8");
+  assert.doesNotMatch(docs, /4-10 slices/);
+  assert.match(docs, /1-10 slices/);
+  assert.match(docs, /single-file/);
+});
+
+test("prompt templates carry right-sized planning and closeout mode guidance", () => {
+  const planTemplate = readFileSync(join(process.cwd(), "src", "resources", "extensions", "gsd", "prompts", "plan-milestone.md"), "utf-8");
+  const completeTemplate = readFileSync(join(process.cwd(), "src", "resources", "extensions", "gsd", "prompts", "complete-milestone.md"), "utf-8");
+
+  assert.match(planTemplate, /Use 1-10 slices, sized to the work/);
+  assert.match(planTemplate, /tiny\/single-file\/static work should usually be one slice/);
+  assert.match(planTemplate, /untyped-existing/);
+  assert.match(completeTemplate, /Closeout Review Mode/);
+  assert.match(completeTemplate, /passing validation artifact is present/);
+  assert.doesNotMatch(completeTemplate, /^### Delegate Review Work/m);
+});
+
+test("complete-milestone prompt trusts passing validation artifact", async () => {
+  const base = makeRepo({ "index.html": "<!doctype html>\n<title>Test</title>\n" });
+  try {
+    writeCompleteMilestoneFiles(base, `---\nverdict: pass\nremediation_round: 0\n---\n\n# Validation\n${validationMetadata()}\n\nAll checks passed.`);
+    const prompt = await buildCompleteMilestonePrompt("M001", "Polish static page", base, "minimal");
+    assert.match(prompt, /Passing Validation Artifact/);
+    assert.match(prompt, /Treat it as authoritative/);
+    assert.match(prompt, /Do not delegate fresh reviewer\/security\/tester audits/);
+    assert.match(prompt, /All checks passed/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("complete-milestone prompt trusts centralized markdown body pass verdict", async () => {
+  const base = makeRepo({ "index.html": "<!doctype html>\n<title>Test</title>\n" });
+  try {
+    writeCompleteMilestoneFiles(base, `# Validation\n\n**Verdict:** PASS\n\n${validationMetadata()}\n\nAll checks passed.`);
+    const prompt = await buildCompleteMilestonePrompt("M001", "Polish static page", base, "minimal");
+    assert.match(prompt, /Passing Validation Artifact/);
+    assert.match(prompt, /Treat it as authoritative/);
+    assert.match(prompt, /Do not delegate fresh reviewer\/security\/tester audits/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("complete-milestone prompt does not trust stale pass validation without metadata", async () => {
+  const base = makeRepo({ "index.html": "<!doctype html>\n<title>Test</title>\n" });
+  try {
+    writeCompleteMilestoneFiles(base, "---\nverdict: pass\nremediation_round: 0\n---\n\n# Validation\nAll checks passed.");
+    const prompt = await buildCompleteMilestonePrompt("M001", "Polish static page", base, "minimal");
+    assert.match(prompt, /Validation Requires Attention/);
+    assert.match(prompt, /missing freshness metadata/);
+    assert.doesNotMatch(prompt, /Passing Validation Artifact/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("complete-milestone prompt does not trust pass validation missing current summary coverage", async () => {
+  const base = makeRepo({ "index.html": "<!doctype html>\n<title>Test</title>\n" });
+  try {
+    writeCompleteMilestoneFiles(base, [
+      "---",
+      "verdict: pass",
+      "remediation_round: 0",
+      "---",
+      "",
+      "# Validation",
+      "validation_metadata:",
+      "  covered_artifacts:",
+      "    - `.gsd/milestones/M001/M001-VALIDATION.md`",
+      "    - `.gsd/milestones/M001/M001-ROADMAP.md`",
+      "",
+      "All checks passed.",
+    ].join("\n"));
+    const prompt = await buildCompleteMilestonePrompt("M001", "Polish static page", base, "minimal");
+    assert.match(prompt, /Validation Requires Attention/);
+    assert.match(prompt, /does not cover current milestone artifacts/);
+    assert.doesNotMatch(prompt, /Passing Validation Artifact/);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});
+
+test("complete-milestone prompt keeps deeper review path without passing validation", async () => {
+  const base = makeRepo({ "index.html": "<!doctype html>\n<title>Test</title>\n" });
+  try {
+    writeCompleteMilestoneFiles(base, "---\nverdict: needs-attention\nremediation_round: 0\n---\n\n# Validation\nFix gaps.");
+    const prompt = await buildCompleteMilestonePrompt("M001", "Polish static page", base, "minimal");
+    assert.match(prompt, /Validation Requires Attention/);
+    assert.match(prompt, /verdict `needs-attention`/);
+    assert.match(prompt, /Use `subagent` for review work needing fresh context/i);
+  } finally {
+    rmSync(base, { recursive: true, force: true });
+  }
+});

package/src/resources/extensions/gsd/tests/safety-harness-false-positives.test.ts

@@ -144,6 +144,18 @@ test("safety-harness-bug2-race: bash evidence survives mid-unit reset between to
   assert.ok(bash[0].outputSnippet.includes("found"), "output snippet captured");
 });
 
+test("safety-harness: gsd_exec counts as execution evidence", () => {
+  resetEvidence();
+
+  recordToolCall("tc-exec-1", "gsd_exec", { command: "grep -n render index.html" });
+  recordToolResult("tc-exec-1", "gsd_exec", "Command exited with code 0\n1:render\n", false);
+
+  const bash = getEvidence().filter((e): e is BashEvidence => e.kind === "bash");
+  assert.equal(bash.length, 1, "gsd_exec must be tracked as execution evidence");
+  assert.equal(bash[0].command, "grep -n render index.html");
+  assert.equal(bash[0].exitCode, 0);
+});
+
 // ─── Bug 3: git diff HEAD~1 scope check ─────────────────────────────────────
 
 test("safety-harness-bug3: validateFileChanges works on initial commit (no HEAD~1)", (t) => {
@@ -237,3 +249,20 @@ test("safety-harness-bug3: validateFileChanges works on merge commit", (t) => {
   // Must produce a valid result without throwing
   assert.ok(audit !== null, "audit must be produced for merge commit repo");
 });
+
+test("safety-harness: planned changed file avoids unexpected-file warning", (t) => {
+  const base = mkdtempSync(join(tmpdir(), "gsd-planned-file-"));
+  t.after(() => rmSync(base, { recursive: true, force: true }));
+
+  execFileSync("git", ["init"], { cwd: base });
+  execFileSync("git", ["config", "user.email", "test@example.com"], { cwd: base });
+  execFileSync("git", ["config", "user.name", "Test User"], { cwd: base });
+  writeFileSync(join(base, "index.html"), "<main></main>\n");
+  execFileSync("git", ["add", "index.html"], { cwd: base });
+  execFileSync("git", ["commit", "-m", "add static app"], { cwd: base });
+
+  const audit = validateFileChanges(base, [], ["index.html"]);
+  assert.ok(audit !== null, "audit must be produced");
+  assert.deepEqual(audit!.unexpectedFiles, [], "planned index.html must not be unexpected");
+  assert.deepEqual(audit!.missingFiles, [], "planned index.html must not be missing");
+});

package/src/resources/extensions/gsd/tests/smart-entry-complete.test.ts

@@ -47,6 +47,9 @@ test("guided-flow complete branch offers a chooser for next milestone or status"
   const branchChunk = guidedFlowSource.slice(branchIdx, nextBranchIdx === -1 ? branchIdx + 1600 : nextBranchIdx);
 
   assert.match(branchChunk, /showNextAction\(/, "complete branch should present a chooser");
+  assert.match(branchChunk, /id:\s*"quick_task"/, "complete branch should offer quick task before milestone planning");
+  assert.match(branchChunk, /Do a small bounded task without opening a milestone/, "quick task action should explain that it avoids milestones");
+  assert.match(branchChunk, /recommended:\s*true/, "quick task action should be the recommended complete-state action");
   assert.match(branchChunk, /findMilestoneIds\(basePath\)/, "complete branch should compute the next milestone id");
   assert.match(
     branchChunk,
@@ -56,6 +59,41 @@ test("guided-flow complete branch offers a chooser for next milestone or status"
   assert.match(branchChunk, /dispatchWorkflow\(pi, await prepareAndBuildDiscussPrompt\(/, "complete branch should dispatch the prepared discuss prompt");
 });
 
+test("guided-flow quick task choices prompt for text and route through /gsd quick", () => {
+  const guidedFlowSource = readFileSync(join(import.meta.dirname, "..", "guided-flow.ts"), "utf-8");
+
+  assert.match(
+    guidedFlowSource,
+    /async function runQuickTaskChoice\([\s\S]*ctx\.ui\.input\("Quick task"[\s\S]*await import\("\.\/quick\.js"\)[\s\S]*await handleQuick\(task,\s*ctx,\s*pi\)/,
+    "quick task chooser should prompt for task text and route through handleQuick",
+  );
+
+  const notifyOnlyPattern = /if \(choice === "quick_task"\) \{\s*ctx\.ui\.notify\("Run \/gsd quick <task>/;
+  assert.doesNotMatch(
+    guidedFlowSource,
+    notifyOnlyPattern,
+    "quick task chooser must not merely print instructions after selection",
+  );
+
+  const quickChoiceCalls = guidedFlowSource.match(/if \(choice === "quick_task"\) \{\s*await runQuickTaskChoice\(ctx,\s*pi\);/g) ?? [];
+  assert.equal(quickChoiceCalls.length, 3, "all guided-flow quick task choices should dispatch the helper");
+});
+
+test("dispatcher routes multi-word freeform /gsd input through /gsd do", () => {
+  const dispatcherSource = readFileSync(join(import.meta.dirname, "..", "commands", "dispatcher.ts"), "utf-8");
+
+  assert.match(
+    dispatcherSource,
+    /if\s*\(trimmed\.includes\(" "\)\)\s*\{[\s\S]*handleDo\(trimmed,\s*ctx,\s*pi\)/,
+    "dispatcher should treat multi-word unknown input as natural-language /gsd do work",
+  );
+  assert.match(
+    dispatcherSource,
+    /Unknown: \/gsd/,
+    "single-token unknown commands should still report the normal unknown-command warning",
+  );
+});
+
 test("guided-flow needs-discussion skip branch opens the project DB before reserving a new milestone", () => {
   const guidedFlowSource = readFileSync(join(import.meta.dirname, "..", "guided-flow.ts"), "utf-8");
   const laterDbOpenIdx = guidedFlowSource.indexOf("// Ensure DB is open before querying slices (#2560).");

package/src/resources/extensions/gsd/tests/start-auto-detached.test.ts

@@ -3,6 +3,8 @@ import assert from "node:assert/strict";
 import { readFileSync } from "node:fs";
 import { resolve } from "node:path";
 
+import { _withDetachedAutoKeepaliveForTest } from "../auto.ts";
+
 const gsdDir = resolve(import.meta.dirname, "..");
 
 function readGsdFile(relativePath: string): string {
@@ -102,8 +104,8 @@ test("startAutoDetached reports failures asynchronously (#3733)", () => {
     "auto.ts should export startAutoDetached",
   );
   assert.ok(
-    autoSrc.includes("void startAuto(ctx, pi, base, verboseMode, options).catch"),
-    "startAutoDetached should launch startAuto without awaiting it",
+    autoSrc.includes("void withDetachedAutoKeepalive(startAuto(ctx, pi, base, verboseMode, options)).catch"),
+    "startAutoDetached should launch startAuto without awaiting it and keep the process alive",
   );
   assert.ok(
     autoSrc.includes("ctx.ui.notify(`Auto-start failed: ${message}`, \"error\")"),
@@ -111,6 +113,48 @@ test("startAutoDetached reports failures asynchronously (#3733)", () => {
   );
 });
 
+test("detached auto-start keeps a ref'ed handle until the run settles", async () => {
+  const originalSetInterval = globalThis.setInterval;
+  const originalClearInterval = globalThis.clearInterval;
+  let intervalCreated = false;
+  let intervalCleared = false;
+  let createdHandle: NodeJS.Timeout | undefined;
+  let resolveRun!: () => void;
+  const run = new Promise<void>((resolve) => {
+    resolveRun = resolve;
+  });
+
+  globalThis.setInterval = ((handler: TimerHandler, timeout?: number, ...args: unknown[]) => {
+    intervalCreated = true;
+    assert.equal(timeout, 30_000);
+    void handler;
+    void args;
+    createdHandle = originalSetInterval(() => {}, 1_000_000);
+    assert.equal(createdHandle.hasRef(), true, "detached auto keepalive must be ref'ed");
+    return createdHandle;
+  }) as unknown as typeof setInterval;
+
+  globalThis.clearInterval = ((handle?: NodeJS.Timeout | number | string) => {
+    if (handle === createdHandle) intervalCleared = true;
+    return originalClearInterval(handle);
+  }) as unknown as typeof clearInterval;
+
+  try {
+    const heldRun = _withDetachedAutoKeepaliveForTest(run);
+    assert.equal(intervalCreated, true, "keepalive interval should start immediately");
+    assert.equal(intervalCleared, false, "keepalive should remain active while auto-mode is running");
+
+    resolveRun();
+    await heldRun;
+
+    assert.equal(intervalCleared, true, "keepalive interval should clear when auto-mode settles");
+  } finally {
+    if (createdHandle) originalClearInterval(createdHandle);
+    globalThis.setInterval = originalSetInterval;
+    globalThis.clearInterval = originalClearInterval;
+  }
+});
+
 test("detached auto-start preserves milestone lock across pause/stop cleanup (#3733)", () => {
   const autoSrc = readGsdFile("auto.ts");
   const sessionSrc = readGsdFile("auto/session.ts");

package/src/resources/extensions/gsd/tests/unit-runtime.test.ts

@@ -5,6 +5,7 @@ import {
   clearUnitRuntimeRecord,
   formatExecuteTaskRecoveryStatus,
   inspectExecuteTaskDurability,
+  isInFlightRuntimePhase,
   readUnitRuntimeRecord,
   writeUnitRuntimeRecord,
 } from "../unit-runtime.ts";
@@ -22,6 +23,12 @@ writeFileSync(
   "utf-8",
 );
 
+console.log("\n=== in-flight runtime phases ===");
+{
+  assert.equal(isInFlightRuntimePhase("crashed"), true, "crashed records remain recoverable");
+  assert.equal(isInFlightRuntimePhase("finalized"), false, "finalized records are terminal");
+}
+
 console.log("\n=== runtime record write/read/update ===");
 {
   const first = writeUnitRuntimeRecord(base, "execute-task", "M100/S02/T09", 1000, { phase: "dispatched" });

package/src/resources/extensions/gsd/tests/uok-plan-v2-wiring.test.ts

@@ -118,7 +118,7 @@ test("guided flow checks pending deep setup before plan-v2 gate", () => {
   const source = readFileSync(join(gsdDir, "guided-flow.ts"), "utf-8");
   const showSmartEntryIdx = source.indexOf("export async function showSmartEntry");
   assert.notEqual(showSmartEntryIdx, -1);
-  const deepIdx = source.indexOf("hasPendingDeepStage(prefs, basePath)", showSmartEntryIdx);
+  const deepIdx = source.indexOf("shouldRunDeepProjectSetup(state, prefs, basePath)", showSmartEntryIdx);
   const planIdx = source.indexOf("runPlanV2Gate(ctx, basePath, state)", showSmartEntryIdx);
   assert.ok(
     deepIdx > -1 && planIdx > -1 && deepIdx < planIdx,

package/src/resources/extensions/gsd/tests/working-output-messages.test.ts

@@ -0,0 +1,93 @@
+// GSD-2 + src/resources/extensions/gsd/tests/working-output-messages.test.ts - Regression coverage for user-facing working-state message quality.
+
+import test from "node:test";
+import assert from "node:assert/strict";
+
+import {
+  evaluateWorkingOutputMessage,
+  evaluateWorkingOutputMessages,
+  formatAutoUnitWorkingMessage,
+} from "../working-output-messages.ts";
+
+test("auto unit loader messages name the active work", () => {
+  assert.equal(
+    formatAutoUnitWorkingMessage("research-milestone", "M001"),
+    "Researching M001: waiting for provider response",
+  );
+  assert.equal(
+    evaluateWorkingOutputMessage({
+      surface: "loader",
+      message: formatAutoUnitWorkingMessage("research-milestone", "M001"),
+      context: { unitType: "research-milestone", unitId: "M001", health: "waiting" },
+    }).length,
+    0,
+  );
+});
+
+test("generic working messages are flagged", () => {
+  const findings = evaluateWorkingOutputMessage({
+    surface: "loader",
+    message: "Working...",
+    context: { health: "waiting" },
+  });
+
+  assert.deepEqual(findings.map(f => f.code), ["generic-working-message"]);
+});
+
+test("dashboard messages cannot claim healthy progress while recovering", () => {
+  const findings = evaluateWorkingOutputMessage({
+    surface: "dashboard",
+    message: "GSD AUTO Progressing well",
+    context: { health: "recovering", recoveryAttempts: 1 },
+  });
+
+  assert.deepEqual(findings.map(f => f.code), ["misleading-healthy-message"]);
+});
+
+test("pre-roadmap zero-slice counters are flagged", () => {
+  const findings = evaluateWorkingOutputMessage({
+    surface: "dashboard",
+    message: "0/0 slices",
+    context: { health: "waiting" },
+  });
+
+  assert.deepEqual(findings.map(f => f.code), ["fake-zero-progress"]);
+});
+
+test("stalled/provider-error/timeout messages need an action", () => {
+  const noAction = evaluateWorkingOutputMessage({
+    surface: "notification",
+    message: "Provider stalled after 2m",
+    context: { health: "stalled" },
+  });
+  assert.deepEqual(noAction.map(f => f.code), ["missing-action"]);
+
+  const withAction = evaluateWorkingOutputMessage({
+    surface: "notification",
+    message: "Provider stalled after 2m. Type /gsd stop or wait for retry.",
+    context: { health: "stalled" },
+  });
+  assert.equal(withAction.length, 0);
+});
+
+test("current fixed research working surfaces pass the audit", () => {
+  const findings = evaluateWorkingOutputMessages([
+    {
+      surface: "loader",
+      message: "Researching M001: waiting for provider response",
+      context: { unitType: "research-milestone", unitId: "M001", health: "waiting" },
+    },
+    {
+      surface: "dashboard",
+      message: "Recovering\nretry 1 after idle stall",
+      context: { unitType: "research-milestone", unitId: "M001", health: "recovering", recoveryAttempts: 1 },
+    },
+    {
+      surface: "notification",
+      message: "Auto-mode stopped - User requested stop.",
+      context: { health: "stopped" },
+    },
+  ]);
+
+  assert.equal(findings.length, 0);
+});