ThothII/backend/scripts/p1-acceptance.test.mjs

import assert from "node:assert/strict";
import { execFile } from "node:child_process";
import {
  chmod, lstat, mkdir, mkdtemp, readFile, realpath, rm, symlink, writeFile,
} from "node:fs/promises";
import { tmpdir } from "node:os";
import { dirname, join } from "node:path";
import { promisify } from "node:util";
import { fileURLToPath } from "node:url";
import test from "node:test";

import {
  canonicalIntegrationBase,
  CHECK_IDS,
  buildSafeEnvironment,
  installExternalFetchGuard,
  negativeRequestEvidence,
  cleanupOwnedRun,
  createOwnedRun,
  deriveOverall,
  executeChecks,
  exportArchiveEvidencePath,
  readAndValidateOwnership,
  runCommand,
  runIntegration,
  scalarSecretBytes,
  scanSecrets,
  validateReport,
  validateRunRoot,
} from "./p1-acceptance.mjs";

const execFileAsync = promisify(execFile);
const roots = [];
async function fakeRepository() {
  const root = await mkdtemp(join(tmpdir(), "p1 acceptance repository with spaces-"));
  roots.push(root);
  await mkdir(join(root, ".artifacts", "p1-integration"), { recursive: true });
  return await realpath(root);
}

test.afterEach(async () => {
  await Promise.all(roots.splice(0).map((root) => rm(root, { recursive: true, force: true })));
});

test("run roots are only canonical direct integration children", async () => {
  const repositoryRoot = await fakeRepository();
  const base = canonicalIntegrationBase(repositoryRoot);
  const id = `p1-${"a".repeat(32)}`;
  assert.equal(validateRunRoot(repositoryRoot, join(base, id), id), join(base, id));
  for (const candidate of [
    base,
    join(repositoryRoot, ".artifacts", "manual-acceptance", id),
    join(base, id, "nested"),
    join(base, "foreign"),
    join(dirname(base), id),
  ]) assert.throws(() => validateRunRoot(repositoryRoot, candidate, id));
  assert.throws(() => validateRunRoot(repositoryRoot, join(base, `p1-${"A".repeat(32)}`), `p1-${"A".repeat(32)}`));
});

test("cleanup refuses every unowned or ambiguous root", async () => {
  const repositoryRoot = await fakeRepository();
  const base = canonicalIntegrationBase(repositoryRoot);
  const cases = [
    ["missing ownership", async (run) => rm(join(run.root, "ownership.json"))],
    ["malformed ownership", async (run) => writeFile(join(run.root, "ownership.json"), "{")],
    ["mismatched root", async (run) => {
      const value = JSON.parse(await readFile(join(run.root, "ownership.json"), "utf8"));
      value.root = join(base, `p1-${"b".repeat(32)}`);
      await writeFile(join(run.root, "ownership.json"), JSON.stringify(value));
    }],
    ["mismatched pid", async (run) => {
      const value = JSON.parse(await readFile(join(run.root, "ownership.json"), "utf8"));
      value.pid += 1;
      await writeFile(join(run.root, "ownership.json"), JSON.stringify(value));
    }],
    ["wrong resource list", async (run) => {
      const value = JSON.parse(await readFile(join(run.root, "ownership.json"), "utf8"));
      value.resources.push(join(repositoryRoot, "foreign"));
      await writeFile(join(run.root, "ownership.json"), JSON.stringify(value));
    }],
  ];
  for (const [, mutate] of cases) {
    const run = await createOwnedRun({ repositoryRoot });
    await mutate(run);
    await assert.rejects(cleanupOwnedRun({ repositoryRoot, runRoot: run.root, expectedNonce: run.nonce }));
    assert.equal((await lstat(run.root)).isDirectory(), true);
  }
  const wrongNonce = await createOwnedRun({ repositoryRoot });
  await assert.rejects(cleanupOwnedRun({ repositoryRoot, runRoot: wrongNonce.root, expectedNonce: "0".repeat(64) }));
  const symlinkRun = await createOwnedRun({ repositoryRoot });
  const target = `${symlinkRun.root}-target`;
  await rm(symlinkRun.root, { recursive: true });
  await mkdir(target);
  await symlink(target, symlinkRun.root);
  await assert.rejects(cleanupOwnedRun({ repositoryRoot, runRoot: symlinkRun.root, expectedNonce: symlinkRun.nonce }));
  for (const bad of [base, join(repositoryRoot, ".artifacts", "manual-acceptance"), join(base, "foreign")]) {
    await assert.rejects(cleanupOwnedRun({ repositoryRoot, runRoot: bad, expectedNonce: "0".repeat(64) }));
  }
});

test("cleanup atomically removes one owned root and preserves siblings", async () => {
  const repositoryRoot = await fakeRepository();
  const run = await createOwnedRun({ repositoryRoot });
  const sibling = join(canonicalIntegrationBase(repositoryRoot), `p1-${"c".repeat(32)}`);
  await mkdir(sibling);
  await writeFile(join(sibling, "sentinel"), "foreign");
  await readAndValidateOwnership({ repositoryRoot, runRoot: run.root, expectedNonce: run.nonce });
  await cleanupOwnedRun({ repositoryRoot, runRoot: run.root, expectedNonce: run.nonce });
  await assert.rejects(lstat(run.root));
  assert.equal(await readFile(join(sibling, "sentinel"), "utf8"), "foreign");
});

function resultFor(id) {
  return {
    id, status: "PASS", startedAt: "2026-08-09T00:00:00.000Z",
    finishedAt: "2026-08-09T00:00:01.000Z", commands: ["git"],
    artifacts: [{ path: `logs/${id}.json`, sha256: "a".repeat(64) }],
  };
}
function validReport(checks = CHECK_IDS.map(resultFor)) {
  return {
    schemaVersion: 1, runId: `p1-${"d".repeat(32)}`, startedAt: "2026-08-09T00:00:00.000Z",
    finishedAt: "2026-08-09T00:00:02.000Z", command: "p1-acceptance integration --keep",
    overall: deriveOverall(checks), checks,
  };
}

test("report validation enforces uniqueness, derivation, safe evidence, hashes, times, and commands", () => {
  assert.doesNotThrow(() => validateReport(validReport()));
  const mutations = [
    (r) => r.checks.push(structuredClone(r.checks[0])),
    (r) => { r.checks[0].attempt = 1; },
    (r) => { r.checks[0].artifacts[0].path = "../secret"; },
    (r) => { r.checks[0].artifacts[0].sha256 = "bad"; },
    (r) => { r.checks[0].startedAt = "today"; },
    (r) => { r.checks[0].commands = ["git status"]; },
    (r) => { r.overall = "PASS"; r.checks[0].status = "FAIL"; },
    (r) => { r.nested = { retries: 2 }; },
  ];
  for (const mutate of mutations) {
    const report = validReport(); mutate(report); assert.throws(() => validateReport(report));
  }
});

function exactScenarios(run = async () => ({ commands: [], artifacts: [] })) {
  return CHECK_IDS.map((id) => ({ id, run: () => run(id) }));
}

test("injected failure executes once, retains a complete ordered diagnostic report, and returns nonzero", async () => {
  const repositoryRoot = await fakeRepository();
  const calls = [];
  const failAt = CHECK_IDS[3];
  const result = await runIntegration({
    repositoryRoot, keep: false, failAt,
    checks: exactScenarios(async (id) => { calls.push(id); return { commands: [], artifacts: [] }; }),
  });
  assert.equal(result.exitCode, 1);
  assert.deepEqual(calls, CHECK_IDS.slice(0, 4));
  assert.equal((await lstat(result.runRoot)).isDirectory(), true);
  const report = JSON.parse(await readFile(join(result.runRoot, "report.json"), "utf8"));
  assert.deepEqual(report.checks.map(({ id }) => id), CHECK_IDS);
  assert.equal(report.checks.filter((check) => check.status === "FAIL").length, CHECK_IDS.length - 3);
  assert.equal(report.checks[3].error, "Acceptance scenario failed safely.");
  assert.equal(report.checks[4].error, "Not executed after earlier failure.");
});

test("executeChecks never repeats or executes after first failure but emits the exact check set", async () => {
  const calls = new Map();
  const result = await executeChecks({
    checks: exactScenarios(async (id) => { calls.set(id, (calls.get(id) ?? 0) + 1); return {}; }),
    failAt: CHECK_IDS[1],
  });
  assert.deepEqual(result.map(({ id }) => id), CHECK_IDS);
  assert.deepEqual(Object.fromEntries(calls), Object.fromEntries(CHECK_IDS.slice(0, 2).map((id) => [id, 1])));
  assert.equal(result[1].status, "FAIL");
  assert(result.slice(2).every(({ status, error }) => status === "FAIL" && error === "Not executed after earlier failure."));
  assert.throws(() => validateReport(validReport(CHECK_IDS.slice(0, -1).map(resultFor))));
  await assert.rejects(executeChecks({ checks: exactScenarios().reverse() }));
});

test("owned setup failure still writes one safe result for every exact check", async () => {
  const repositoryRoot = await fakeRepository();
  const result = await runIntegration({
    repositoryRoot, keep: false,
    setup: async () => { throw new Error("fixture setup raw failure"); },
  });
  assert.equal(result.exitCode, 1);
  const report = JSON.parse(await readFile(join(result.runRoot, "report.json"), "utf8"));
  assert.deepEqual(report.checks.map(({ id }) => id), CHECK_IDS);
  assert.equal(report.checks[0].error, "Acceptance setup failed safely.");
  assert(report.checks.slice(1).every(({ error }) => error === "Not executed after earlier failure."));
});

test("scalar fixture secret files contain no harness-invalid whitespace", () => {
  const bytes = scalarSecretBytes("CANARY-secret-value-123456");
  assert.equal(bytes.toString("utf8"), "CANARY-secret-value-123456");
  assert.equal([...bytes].some((byte) => /\s/.test(String.fromCharCode(byte))), false);
  assert.throws(() => scalarSecretBytes("bad secret"));
});

test("secret scanner excludes only the direct fixture-secrets subtree", async () => {
  const repositoryRoot = await fakeRepository();
  const run = await createOwnedRun({ repositoryRoot });
  const canary = "CANARY-secret-value-123456";
  await mkdir(join(run.root, "fixture-secrets"));
  await writeFile(join(run.root, "fixture-secrets", "allowed"), canary);
  const paths = [
    "logs/a.log", "responses/a.json", "rendered/a.yaml", "exports/raw/a.zip",
    "exports/extracted/a.md", "requests/a.json", "report-preview.md", "nested/fixture-secrets/not-excluded",
  ];
  for (const path of paths) {
    await mkdir(dirname(join(run.root, path)), { recursive: true });
    await writeFile(join(run.root, path), `prefix ${canary} suffix`);
  }
  const findings = await scanSecrets({ runRoot: run.root, forbiddenValues: [canary] });
  assert.deepEqual(new Set(findings.map((finding) => finding.path)), new Set(paths));
});

test("secret scanner examines reachable Git blobs, not just loose file bytes", async () => {
  const repositoryRoot = await fakeRepository();
  const run = await createOwnedRun({ repositoryRoot });
  const canary = "GIT-CANARY-secret-value-987654";
  const gitRoot = join(run.root, "author");
  await mkdir(gitRoot);
  await execFileAsync("git", ["init", "--initial-branch=main"], { cwd: gitRoot });
  await execFileAsync("git", ["config", "user.name", "Scanner Test"], { cwd: gitRoot });
  await execFileAsync("git", ["config", "user.email", "scanner@example.invalid"], { cwd: gitRoot });
  await writeFile(join(gitRoot, "secret.txt"), canary);
  await execFileAsync("git", ["add", "secret.txt"], { cwd: gitRoot });
  await execFileAsync("git", ["commit", "-m", "secret blob"], { cwd: gitRoot });
  await execFileAsync("git", ["rm", "secret.txt"], { cwd: gitRoot });
  await execFileAsync("git", ["commit", "-m", "remove worktree copy"], { cwd: gitRoot });
  const findings = await scanSecrets({ runRoot: run.root, forbiddenValues: [canary] });
  assert.equal(findings.some((finding) => finding.path.startsWith("git-object:")), true);
});

test("successful lifecycle honors keep and cleanup", async () => {
  const repositoryRoot = await fakeRepository();
  const checks = exactScenarios();
  const kept = await runIntegration({ repositoryRoot, keep: true, checks });
  assert.equal(kept.exitCode, 0);
  assert.equal((await lstat(kept.runRoot)).isDirectory(), true);
  const cleaned = await runIntegration({ repositoryRoot, keep: false, checks });
  assert.equal(cleaned.exitCode, 0);
  await assert.rejects(lstat(cleaned.runRoot));
});

test("command helper accepts only executable plus separate argv", async () => {
  await assert.rejects(runCommand("git status"));
  await assert.rejects(runCommand({ executable: "/bin/echo", argv: "hello" }));
  await assert.rejects(runCommand({ executable: "/bin/echo", argv: [], shell: true }));
  await assert.rejects(runCommand({ executable: "git status; rm -rf /", argv: [] }));
  const repositoryRoot = await fakeRepository();
  const executable = join(repositoryRoot, "executable with spaces");
  await writeFile(executable, "#!/bin/sh\nprintf '%s' \"$1\"\n", { mode: 0o700 });
  await chmod(executable, 0o700);
  const result = await runCommand({ executable, argv: ["literal;not-a-shell"] });
  assert.equal(result.stdout, "literal;not-a-shell");
  assert.equal(result.code, 0);
});


test("safe environment rejects ambient THT and keeps only strict process allowlist plus fixture values", () => {
  const safe = buildSafeEnvironment({
    ambient: { PATH: "/safe/bin", HOME: "/home/test", LANG: "C", THT_SECRETS_FILE: "/real/secrets", AWS_SECRET_ACCESS_KEY: "real" },
    fixture: { THT_BIN: "/fixture/tht", THT_WORKSPACE_SECRET_ROOTS: "/fixture/secrets" },
  });
  assert.deepEqual(safe, {
    PATH: "/safe/bin", HOME: "/home/test", LANG: "C",
    THT_BIN: "/fixture/tht", THT_WORKSPACE_SECRET_ROOTS: "/fixture/secrets",
  });
});

test("secret scan fails closed when Git enumeration fails", async () => {
  const repositoryRoot = await fakeRepository();
  const run = await createOwnedRun({ repositoryRoot });
  await mkdir(join(run.root, "remote.git"));
  await assert.rejects(scanSecrets({ runRoot: run.root, forbiddenValues: ["CANARY-value-123"] }), /Git secret scan failed closed/);
});


test("negative request evidence persists only case label and expected input field", () => {
  const value = negativeRequestEvidence("credential-field", "evidence.source.password");
  assert.deepEqual(value, { case: "credential-field", expectedInputField: "evidence.source.password" });
  assert.equal(JSON.stringify(value).includes("body"), false);
});

test("external fetch guard permits only the owned loopback API and records external attempts", async () => {
  const called = [];
  const guard = installExternalFetchGuard("http://127.0.0.1:12345", async (url) => { called.push(String(url)); return { ok: true }; });
  await guard.fetch("http://127.0.0.1:12345/workspaces");
  await assert.rejects(guard.fetch("https://evidence.example.test/guide.md"), /external fetch prohibited/);
  await assert.rejects(guard.fetch("http://127.0.0.1:9999/health"), /external fetch prohibited/);
  assert.deepEqual(called, ["http://127.0.0.1:12345/workspaces"]);
  assert.equal(guard.externalAttempts.length, 2);
});


test("export archive evidence path matches the persisted binary request id", () => {
  assert.equal(exportArchiveEvidencePath("export-p1-filesystem"), "exports/raw/export-p1-filesystem.zip");
});


test("announce callback observes PASS and manual pending before non-keep cleanup", async () => {
  const repositoryRoot = await fakeRepository();
  let observed;
  const result = await runIntegration({
    repositoryRoot, keep: false, checks: exactScenarios(),
    announce: async ({ report, runRoot }) => {
      observed = { overall: report.overall, manual: "PENDING", rootExists: (await lstat(runRoot)).isDirectory() };
    },
  });
  assert.deepEqual(observed, { overall: "PASS", manual: "PENDING", rootExists: true });
  assert.equal(result.retained, false);
});

test("public wrapper replaces ambient environment before invoking the runner", async () => {
  const wrapper = await readFile(join(dirname(fileURLToPath(import.meta.url)), "..", "..", "scripts", "p1-acceptance.sh"), "utf8");
  assert.match(wrapper, /safe_env=\(env -i/);
  assert.match(wrapper, /P1_ACCEPTANCE_FAIL_AT/);
  assert.doesNotMatch(wrapper, /export THT_BIN/);
});