gsd-pi 2.80.0-dev.c5f2443b3 → 2.80.0-dev.e51d2c88c

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

@@ -79,6 +79,27 @@ test('handlePlanTask rejects invalid payloads', async () => {
   }
 });
 
+test('handlePlanTask rejects absolute task IO paths outside the active worktree', async () => {
+  const base = makeTmpBase();
+  openDatabase(join(base, '.gsd', 'gsd.db'));
+
+  try {
+    seedParent();
+    const outside = join(tmpdir(), 'outside-checkout', 'index.html');
+    const result = await handlePlanTask({
+      ...validParams(),
+      inputs: [outside],
+      expectedOutput: [outside],
+    }, base);
+
+    assert.ok('error' in result);
+    assert.match(result.error, /validation failed: inputs contains absolute path outside working directory/);
+    assert.equal(getTask('M001', 'S02', 'T02'), null, 'invalid planning IO must not persist the task');
+  } finally {
+    cleanup(base);
+  }
+});
+
 test('handlePlanTask rejects missing parent slice', async () => {
   const base = makeTmpBase();
   openDatabase(join(base, '.gsd', 'gsd.db'));

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

@@ -0,0 +1,102 @@
+// Project/App: GSD-2
+// File Purpose: Golden-fixture equivalence tests pinning PR-body output for buildPrEvidence and formatSwarmLanePRBody.
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync, writeFileSync } from "node:fs";
+import { fileURLToPath } from "node:url";
+import { dirname, join } from "node:path";
+
+import { buildPrEvidence, type PrEvidenceInput } from "../pr-evidence.ts";
+import {
+  formatSwarmLanePRBody,
+  type SwarmLanePRData,
+} from "../../github-sync/templates.ts";
+
+const __dirname = dirname(fileURLToPath(import.meta.url));
+const FIXTURES_DIR = join(__dirname, "fixtures", "pr-body");
+
+const UPDATE = process.env.UPDATE_GOLDENS === "1";
+
+function compareGolden(name: string, actual: string): void {
+  const path = join(FIXTURES_DIR, name);
+  if (UPDATE) {
+    writeFileSync(path, actual, "utf8");
+    return;
+  }
+  const expected = readFileSync(path, "utf8");
+  assert.equal(actual, expected, `golden mismatch for ${name}`);
+}
+
+const SHIP_BASIC: PrEvidenceInput = {
+  milestoneId: "M001",
+  milestoneTitle: "Authentication",
+  changeType: "feat",
+  linkedIssue: "Closes #123",
+  summaries: ["### S01\nImplemented login flow."],
+  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 SHIP_EMPTY_OPTIONALS: PrEvidenceInput = {
+  milestoneId: "M001",
+  milestoneTitle: "Authentication",
+  changeType: "feat",
+};
+
+const SWARM_WITH_BLOCKERS: SwarmLanePRData = {
+  lane: {
+    id: "writer",
+    branch: "lane/single-writer",
+    owner: "@owner",
+    latestCommit: "abc1234",
+    changedContracts: ["WriterToken"],
+    testEvidence: ["npm run typecheck:extensions"],
+    blockers: ["Awaiting state-lane writer-sequence merge", "Parity report incomplete"],
+  },
+  impactArea: "Single-writer UOK metadata.",
+  transitionRisks: ["Writer token lifecycle regression"],
+  rollbackPlan: ["Disable writer sequence enrichment"],
+  linkedIssue: 123,
+};
+
+const SWARM_NO_BLOCKERS: SwarmLanePRData = {
+  lane: {
+    id: "writer",
+    branch: "lane/single-writer",
+    owner: "@owner",
+    latestCommit: "abc1234",
+    changedContracts: ["WriterToken"],
+    testEvidence: ["npm run typecheck:extensions"],
+  },
+  impactArea: "Single-writer UOK metadata.",
+  transitionRisks: ["Writer token lifecycle regression"],
+  rollbackPlan: ["Disable writer sequence enrichment"],
+  linkedIssue: 123,
+};
+
+test("pr-evidence golden: commands-ship basic", () => {
+  compareGolden("commands-ship-basic.md", buildPrEvidence(SHIP_BASIC).body);
+});
+
+test("pr-evidence golden: commands-ship empty optionals", () => {
+  compareGolden("commands-ship-empty-optionals.md", buildPrEvidence(SHIP_EMPTY_OPTIONALS).body);
+});
+
+test("pr-evidence golden: swarm-lane with blockers", () => {
+  const body = formatSwarmLanePRBody(SWARM_WITH_BLOCKERS);
+  compareGolden("swarm-lane-with-blockers.md", body);
+  // Cross-check: top-level ## Blockers heading must appear (regression guard
+  // for the silent-drop fixed in this PR).
+  assert.ok(body.includes("\n## Blockers\n"), "swarm-lane body must emit a top-level ## Blockers heading");
+});
+
+test("pr-evidence golden: swarm-lane no blockers (no Blockers heading)", () => {
+  const body = formatSwarmLanePRBody(SWARM_NO_BLOCKERS);
+  compareGolden("swarm-lane-no-blockers.md", body);
+  assert.ok(!body.includes("## Blockers"), "swarm-lane body without blockers must not emit ## Blockers heading");
+});

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/prompt-path-audit.test.ts

@@ -0,0 +1,40 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { readFileSync, readdirSync } from "node:fs";
+import { join } from "node:path";
+import { fileURLToPath } from "node:url";
+import { dirname } from "node:path";
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = dirname(__filename);
+const promptsDir = join(__dirname, "..", "prompts");
+
+test("prompt templates do not reference legacy milestone-root .gsd paths", () => {
+  const offenders: string[] = [];
+  for (const file of readdirSync(promptsDir)) {
+    if (!file.endsWith(".md")) continue;
+    const content = readFileSync(join(promptsDir, file), "utf-8");
+    const legacyPatterns = [
+      /\.gsd\/\{\{(?:milestoneId|mid)\}\}\//g,
+      /\.gsd\/<milestone-id>\//g,
+      /\.gsd\/<ID>\//g,
+    ];
+    for (const pattern of legacyPatterns) {
+      if (pattern.test(content)) {
+        offenders.push(`${file}: ${pattern.source}`);
+      }
+    }
+  }
+
+  assert.deepEqual(
+    offenders,
+    [],
+    "Milestone artifacts must use .gsd/milestones/<MID>/..., not legacy .gsd/<MID>/...",
+  );
+});
+
+test("quick task prompt delegates commit policy to quick.ts", () => {
+  const content = readFileSync(join(promptsDir, "quick-task.md"), "utf-8");
+  assert.match(content, /\{\{commitInstruction\}\}/);
+  assert.doesNotMatch(content, /Stage only relevant files/);
+});

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

@@ -65,6 +65,25 @@ describe('prompt step ordering (#3696)', () => {
     assert.ok(learningsIdx < completeMilestoneIdx, 'learnings extraction must happen before gsd_complete_milestone');
   });
 
+  test('complete-milestone duplicate guard checks milestone status before durable writes', () => {
+    const guardMatch = completeMilestoneMd.match(/^\d+\.\s.*gsd_milestone_status/m);
+    const reqUpdateMatch = completeMilestoneMd.match(/^\d+\.\s.*gsd_requirement_update/m);
+    assert.ok(guardMatch, 'complete-milestone must start with a gsd_milestone_status duplicate guard');
+    assert.ok(reqUpdateMatch, 'gsd_requirement_update should appear in a numbered step');
+
+    const guardIdx = completeMilestoneMd.indexOf(guardMatch![0]);
+    const reqUpdateIdx = completeMilestoneMd.indexOf(reqUpdateMatch![0]);
+    assert.ok(
+      guardIdx < reqUpdateIdx,
+      'duplicate guard must run before requirement/project/learnings writes',
+    );
+    assert.match(
+      completeMilestoneMd,
+      /status(?:`|\*\*)?\s+(?:is\s+)?(?:`complete`|"complete")/i,
+      'duplicate guard must tell the agent to stop when status is complete',
+    );
+  });
+
   test('complete-slice.md uses gsd_requirement_update', () => {
     assert.match(completeSliceMd, /gsd_requirement_update/,
       'complete-slice.md should reference gsd_requirement_update');

package/src/resources/extensions/gsd/tests/quick-external-gsd.test.ts

@@ -0,0 +1,40 @@
+import test from "node:test";
+import assert from "node:assert/strict";
+import { mkdtempSync, mkdirSync, realpathSync, rmSync, symlinkSync, writeFileSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+import { spawnSync } from "node:child_process";
+import { buildQuickCommitInstruction } from "../quick.ts";
+
+function git(cwd: string, args: string[]): void {
+  const result = spawnSync("git", args, { cwd, encoding: "utf-8" });
+  assert.equal(result.status, 0, result.stderr || result.stdout);
+}
+
+test("quick task commit instruction does not ask agents to stage external .gsd quick files", { skip: process.platform === "win32" }, () => {
+  const tempRoot = realpathSync(mkdtempSync(join(tmpdir(), "gsd-quick-ext-")));
+  const repo = join(tempRoot, "repo");
+  const externalGsd = join(tempRoot, "state");
+  mkdirSync(repo);
+  mkdirSync(externalGsd);
+
+  const previousCwd = process.cwd();
+  try {
+    git(repo, ["init"]);
+    git(repo, ["config", "user.email", "test@example.com"]);
+    git(repo, ["config", "user.name", "Test User"]);
+    writeFileSync(join(repo, "README.md"), "# Test\n", "utf-8");
+    git(repo, ["add", "README.md"]);
+    git(repo, ["commit", "-m", "init"]);
+    symlinkSync(externalGsd, join(repo, ".gsd"), "dir");
+
+    const instruction = buildQuickCommitInstruction(repo, join(repo, ".gsd"));
+
+    assert.match(instruction, /do not stage or commit `\.gsd\/quick\/\.\.\.`/);
+    assert.match(instruction, /nothing in the project repo to commit/);
+    assert.match(instruction, /Write the quick summary file directly/);
+  } finally {
+    process.chdir(previousCwd);
+    rmSync(tempRoot, { recursive: true, force: true });
+  }
+});

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");
+});