gsd-pi 2.76.0 → 2.77.0-dev.538325aea

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

@@ -0,0 +1,205 @@
+/**
+ * Regression tests for three false-positive sources in the safety harness.
+ * Issue #4385
+ *
+ * Bug 1: Hardcoded BASH_READ_ONLY_RE — new legitimate commands blocked
+ * Bug 2: Non-persisted evidence — session restart causes false positive on resume
+ * Bug 3: git diff HEAD~1 scope check — fails on initial commits / shallow clones
+ */
+
+import test from "node:test";
+import assert from "node:assert/strict";
+import { execFileSync } from "node:child_process";
+import { mkdtempSync, mkdirSync, writeFileSync, rmSync } from "node:fs";
+import { tmpdir } from "node:os";
+import { join } from "node:path";
+
+import { shouldBlockQueueExecution } from "../bootstrap/write-gate.ts";
+import {
+  resetEvidence,
+  recordToolCall,
+  getEvidence,
+  saveEvidenceToDisk,
+  loadEvidenceFromDisk,
+} from "../safety/evidence-collector.ts";
+import { validateFileChanges } from "../safety/file-change-validator.ts";
+
+// ─── Bug 1: Hardcoded Bash allowlist ────────────────────────────────────────
+
+test("safety-harness-bug1: npm commands are not blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "npm run test", true);
+  assert.strictEqual(r.block, false, "npm run test must be read-only-safe");
+});
+
+test("safety-harness-bug1: npx commands are not blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "npx tsc --noEmit", true);
+  assert.strictEqual(r.block, false, "npx tsc --noEmit must pass");
+});
+
+test("safety-harness-bug1: tsx commands are not blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "tsx src/index.ts", true);
+  assert.strictEqual(
+    r.block,
+    false,
+    "tsx (TypeScript runner — read-only investigative) must pass",
+  );
+});
+
+test("safety-harness-bug1: node --print commands are not blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "node --print 'process.version'", true);
+  assert.strictEqual(r.block, false, "node --print must pass");
+});
+
+test("safety-harness-bug1: python read-only invocations are not blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "python -c 'import sys; print(sys.version)'", true);
+  assert.strictEqual(r.block, false, "python -c read-only must pass");
+});
+
+test("safety-harness-bug1: jq read-only command is not blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "jq '.version' package.json", true);
+  assert.strictEqual(r.block, false, "jq (read-only JSON query) must pass");
+});
+
+test("safety-harness-bug1: destructive commands are still blocked during queue mode", () => {
+  const r = shouldBlockQueueExecution("bash", "rm -rf dist/", true);
+  assert.strictEqual(r.block, true, "rm -rf must still be blocked");
+});
+
+// ─── Bug 2: Non-persisted evidence ──────────────────────────────────────────
+
+test("safety-harness-bug2: evidence survives save/load round-trip (simulates session restart)", (t) => {
+  const base = mkdtempSync(join(tmpdir(), "gsd-evidence-persist-"));
+  t.after(() => rmSync(base, { recursive: true, force: true }));
+
+  resetEvidence();
+
+  // Simulate bash tool calls during unit execution
+  recordToolCall("tc-001", "Bash", { command: "npm run test:unit" });
+  recordToolCall("tc-002", "Bash", { command: "npx tsc --noEmit" });
+  recordToolCall("tc-003", "Write", { file_path: "src/foo.ts" });
+
+  const before = getEvidence();
+  assert.equal(before.length, 3, "three entries before save");
+
+  // Persist to disk
+  saveEvidenceToDisk(base, "M001", "S001", "T001");
+
+  // Simulate session restart: module-level array reset
+  resetEvidence();
+  assert.equal(getEvidence().length, 0, "in-memory cleared after reset");
+
+  // Resume: load from disk
+  loadEvidenceFromDisk(base, "M001", "S001", "T001");
+
+  const after = getEvidence();
+  assert.equal(after.length, 3, "evidence restored from disk after simulated restart");
+
+  const bashEntries = after.filter((e) => e.kind === "bash");
+  assert.equal(bashEntries.length, 2, "both bash entries restored");
+
+  const writeEntries = after.filter((e) => e.kind === "write");
+  assert.equal(writeEntries.length, 1, "write entry restored");
+});
+
+test("safety-harness-bug2: loadEvidenceFromDisk returns empty array when no file exists (fresh unit)", (t) => {
+  const base = mkdtempSync(join(tmpdir(), "gsd-evidence-nopersist-"));
+  t.after(() => rmSync(base, { recursive: true, force: true }));
+
+  resetEvidence();
+  loadEvidenceFromDisk(base, "M001", "S001", "T001");
+  assert.equal(getEvidence().length, 0, "no evidence on fresh unit is correct — not a false positive");
+});
+
+// ─── Bug 3: git diff HEAD~1 scope check ─────────────────────────────────────
+
+test("safety-harness-bug3: validateFileChanges works on initial commit (no HEAD~1)", (t) => {
+  const base = mkdtempSync(join(tmpdir(), "gsd-initial-commit-"));
+  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.ts"), "export const x = 1;\n");
+  execFileSync("git", ["add", "."], { cwd: base });
+  execFileSync("git", ["commit", "-m", "initial"], { cwd: base });
+
+  // On initial commit, HEAD~1 does not exist — must not throw or produce wrong results
+  const audit = validateFileChanges(base, ["index.ts"], []);
+
+  assert.ok(audit !== null, "audit must be produced for initial commit");
+  assert.deepEqual(audit!.unexpectedFiles, [], "no unexpected files on initial commit");
+  assert.deepEqual(audit!.missingFiles, [], "no missing files on initial commit");
+});
+
+test("safety-harness-bug3: validateFileChanges works on shallow clone (shallow repo without full history)", (t) => {
+  // Simulate shallow clone: create a repo, then clone it with depth=1
+  const origin = mkdtempSync(join(tmpdir(), "gsd-origin-"));
+  const shallow = mkdtempSync(join(tmpdir(), "gsd-shallow-"));
+  t.after(() => {
+    rmSync(origin, { recursive: true, force: true });
+    rmSync(shallow, { recursive: true, force: true });
+  });
+
+  // Set up origin with multiple commits
+  execFileSync("git", ["init"], { cwd: origin });
+  execFileSync("git", ["config", "user.email", "test@example.com"], { cwd: origin });
+  execFileSync("git", ["config", "user.name", "Test User"], { cwd: origin });
+  writeFileSync(join(origin, "a.ts"), "export const a = 1;\n");
+  execFileSync("git", ["add", "."], { cwd: origin });
+  execFileSync("git", ["commit", "-m", "first"], { cwd: origin });
+  writeFileSync(join(origin, "b.ts"), "export const b = 2;\n");
+  execFileSync("git", ["add", "."], { cwd: origin });
+  execFileSync("git", ["commit", "-m", "second"], { cwd: origin });
+
+  // Shallow clone with depth=1 — HEAD~1 will not exist
+  execFileSync("git", ["clone", "--depth=1", `file://${origin}`, shallow], {
+    stdio: ["ignore", "pipe", "pipe"],
+  });
+
+  // Verify the shallow clone has no parent (HEAD~1 unavailable)
+  let hasParent = true;
+  try {
+    execFileSync("git", ["rev-parse", "HEAD~1"], {
+      cwd: shallow,
+      stdio: ["ignore", "pipe", "pipe"],
+    });
+  } catch {
+    hasParent = false;
+  }
+  assert.equal(hasParent, false, "shallow clone should not have HEAD~1");
+
+  // validateFileChanges must not throw or give wrong results
+  const audit = validateFileChanges(shallow, ["b.ts"], []);
+  assert.ok(audit !== null, "audit must be produced even in shallow clone");
+});
+
+test("safety-harness-bug3: validateFileChanges works on merge commit", (t) => {
+  const base = mkdtempSync(join(tmpdir(), "gsd-merge-commit-"));
+  t.after(() => rmSync(base, { recursive: true, force: true }));
+
+  execFileSync("git", ["init", "-b", "main"], { cwd: base });
+  execFileSync("git", ["config", "user.email", "test@example.com"], { cwd: base });
+  execFileSync("git", ["config", "user.name", "Test User"], { cwd: base });
+
+  // Main branch: initial commit
+  writeFileSync(join(base, "main.ts"), "export const m = 1;\n");
+  execFileSync("git", ["add", "."], { cwd: base });
+  execFileSync("git", ["commit", "-m", "initial"], { cwd: base });
+
+  // Feature branch
+  execFileSync("git", ["checkout", "-b", "feature"], { cwd: base });
+  writeFileSync(join(base, "feature.ts"), "export const f = 2;\n");
+  execFileSync("git", ["add", "."], { cwd: base });
+  execFileSync("git", ["commit", "-m", "feature work"], { cwd: base });
+
+  // Merge back to main
+  execFileSync("git", ["checkout", "main"], { cwd: base });
+  execFileSync("git", ["merge", "--no-ff", "feature", "-m", "Merge feature"], { cwd: base });
+
+  // HEAD is now a merge commit with two parents — git diff HEAD~1 gives wrong scope
+  const audit = validateFileChanges(base, ["feature.ts"], []);
+
+  // Must produce a valid result without throwing
+  assert.ok(audit !== null, "audit must be produced for merge commit repo");
+});

package/src/resources/extensions/gsd/tests/save-gate-result-render.test.ts

@@ -0,0 +1,95 @@
+/**
+ * Regression test suite for save_gate_result renderResult.
+ *
+ * Verifies that renderResult does not print "undefined: undefined" when
+ * `details` is empty, and that the error fallback does not produce a
+ * duplicated `Error: Error:` prefix when `content[0].text` already starts
+ * with `Error:`.
+ */
+
+import { test } from 'node:test';
+import assert from 'node:assert/strict';
+import { registerDbTools } from '../bootstrap/db-tools.ts';
+
+function makeMockPi() {
+  const tools: any[] = [];
+  return {
+    registerTool: (tool: any) => tools.push(tool),
+    tools,
+  } as any;
+}
+
+const fakeTheme = {
+  fg: (_color: string, text: string) => text,
+  bold: (text: string) => text,
+};
+
+function getSaveGateResultTool() {
+  const pi = makeMockPi();
+  registerDbTools(pi);
+  const tool = pi.tools.find((t: any) => t.name === 'gsd_save_gate_result');
+  assert.ok(tool, 'gsd_save_gate_result should be registered');
+  return tool;
+}
+
+test('save_gate_result renderResult falls back to content text when details is empty', () => {
+  const tool = getSaveGateResultTool();
+  const result = {
+    content: [{ type: 'text', text: 'Gate Q3 result saved: verdict=pass' }],
+    details: {},
+    isError: false,
+  };
+  const rendered = tool.renderResult(result, {}, fakeTheme);
+  const text = String(rendered.content ?? rendered.text ?? rendered);
+  assert.ok(!text.includes('undefined'), `got: ${text}`);
+  assert.ok(
+    text.includes('Gate Q3') || text.includes('verdict=pass'),
+    `expected content summary — got: ${text}`,
+  );
+});
+
+test('save_gate_result renderResult uses structured details when present', () => {
+  const tool = getSaveGateResultTool();
+  const result = {
+    content: [{ type: 'text', text: 'Gate Q3 result saved: verdict=flag' }],
+    details: { operation: 'save_gate_result', gateId: 'Q3', verdict: 'flag' },
+    isError: false,
+  };
+  const rendered = tool.renderResult(result, {}, fakeTheme);
+  const text = String(rendered.content ?? rendered.text ?? rendered);
+  assert.ok(text.includes('Q3'), `got: ${text}`);
+  assert.ok(text.includes('flag'), `got: ${text}`);
+  assert.ok(!text.includes('undefined'), `got: ${text}`);
+});
+
+test('save_gate_result renderResult shows error from content when details.error is missing', () => {
+  const tool = getSaveGateResultTool();
+  const result = {
+    content: [{ type: 'text', text: 'Error: Invalid gateId "Z1"' }],
+    details: {},
+    isError: true,
+  };
+  const rendered = tool.renderResult(result, {}, fakeTheme);
+  const text = String(rendered.content ?? rendered.text ?? rendered);
+  assert.ok(
+    text.includes('Invalid gateId') || text.includes('Error'),
+    `got: ${text}`,
+  );
+  assert.ok(!text.includes('undefined'), `got: ${text}`);
+});
+
+test('save_gate_result renderResult does not duplicate Error: prefix', () => {
+  const tool = getSaveGateResultTool();
+  const result = {
+    content: [{ type: 'text', text: 'Error: Invalid gateId "Z1"' }],
+    details: {},
+    isError: true,
+  };
+  const rendered = tool.renderResult(result, {}, fakeTheme);
+  const text = String(rendered.content ?? rendered.text ?? rendered);
+  assert.ok(
+    !/Error:\s*Error:/i.test(text),
+    `expected a single Error: prefix — got: ${text}`,
+  );
+  assert.ok(text.includes('Invalid gateId'), `got: ${text}`);
+});