reporting-governance-plugin/scripts/test_long_task_governor_wrapper.mjs

#!/usr/bin/env node
import assert from 'node:assert/strict';
import fs from 'node:fs';
import os from 'node:os';
import { execFileSync, spawnSync } from 'node:child_process';
import path from 'node:path';
import { fileURLToPath } from 'node:url';

const __filename = fileURLToPath(import.meta.url);
const __dirname = path.dirname(__filename);
const repoRoot = path.resolve(__dirname, '..');
const wrapperPath = path.join(repoRoot, 'scripts', 'long_task_governor_wrapper.mjs');

const fixtures = [
  {
    name: 'example',
    file: path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_example.json'),
    assert(output) {
      assert.equal(output.classification, 'long_task');
    },
  },
  {
    name: 'borderline wrapper inference',
    input: {
      requestText: 'Inspect the current hook and compare it to the wrapper outputs before replying.',
      canReplyNow: false,
    },
    assert(output) {
      assert.equal(output.classification, 'long_task');
      assert.equal(output.needsCheckpoint, true);
    },
  },
  {
    name: 'invalid silent',
    file: path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_invalid_silent_example.json'),
    assert(output) {
      assert.equal(output.silentLaunchOk, false);
    },
  },
  {
    name: 'general chat',
    file: path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_general_chat_example.json'),
    assert(output) {
      assert.equal(output.classification, 'general_chat');
    },
  },
  {
    name: 'non-silent long task',
    file: path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_non_silent_long_task_example.json'),
    assert(output) {
      assert.equal(output.classification, 'long_task');
      assert.equal(output.silentCandidate, false);
    },
  },
  {
    name: 'owner decision',
    file: path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_silent_owner_decision_example.json'),
    assert(output) {
      assert.equal(output.handoff.mode, 'button_path');
    },
  },
  {
    name: 'subagent wait',
    file: path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_subagent_wait_example.json'),
    assert(output) {
      assert.equal(output.silentCandidate, true);
      assert.ok(output.progressEvidence && typeof output.progressEvidence === 'object', 'subagent wait: missing progressEvidence');
      assert.equal(typeof output.progressEvidence.sessionKey, 'string', 'subagent wait: missing progressEvidence.sessionKey');
      assert.ok(output.progressEvidence.sessionKey.length > 0, 'subagent wait: empty progressEvidence.sessionKey');
      assert.equal(typeof output.externalizedCheckpointPath, 'string', 'subagent wait: missing externalizedCheckpointPath');
      assert.ok(output.externalizedCheckpointPath.length > 0, 'subagent wait: empty externalizedCheckpointPath');
      assert.equal('task_name' in (output.progressEvidence ?? {}), false, 'subagent wait: progressEvidence must not backfill taskRecord.task_name');
      assert.equal(JSON.stringify(output.progressEvidence).includes('Wait for delegated log survey'), false, 'subagent wait: progressEvidence must not derive from taskRecord.task_name');
    },
  },
];

function runFixture(fixture) {
  const args = fixture.file
    ? [wrapperPath, '--compact', '--input', fixture.file]
    : [wrapperPath, '--compact'];
  const options = {
    cwd: repoRoot,
    encoding: 'utf8',
  };

  if (fixture.input) {
    options.input = `${JSON.stringify(fixture.input)}\n`;
  }

  const stdout = execFileSync(process.execPath, args, options);

  let output;
  try {
    output = JSON.parse(stdout);
  } catch (error) {
    throw new Error(`Fixture \"${fixture.name}\" did not produce valid JSON: ${error.message}\nOutput: ${stdout}`);
  }

  assert.ok(output.classification !== undefined, `${fixture.name}: missing classification`);
  assert.ok(output.silentCandidate !== undefined, `${fixture.name}: missing silentCandidate`);
  assert.ok(output.silentLaunchOk !== undefined, `${fixture.name}: missing silentLaunchOk`);
  assert.ok(output.requiredNextAction !== undefined, `${fixture.name}: missing requiredNextAction`);
  assert.ok(output.handoff && output.handoff.mode !== undefined, `${fixture.name}: missing handoff.mode`);

  fixture.assert(output);

  return {
    name: fixture.name,
    output,
  };
}

function assertErrorCase(name, args, expectedStderr, input) {
  const result = spawnSync(process.execPath, [wrapperPath, ...args], {
    cwd: repoRoot,
    encoding: 'utf8',
    input,
  });

  assert.notEqual(result.status, 0, `${name}: expected non-zero exit`);
  assert.equal(result.stdout, '', `${name}: expected empty stdout`);
  assert.equal(result.stderr.trim(), expectedStderr, `${name}: unexpected stderr`);
}

function main() {
  const results = fixtures.map(runFixture);

  const realismWorkspace = fs.mkdtempSync(path.join(os.tmpdir(), 'wrapper-realism-'));
  try {
    const realismInput = path.join(repoRoot, 'docs', '_artifacts', 'long_task_governor_wrapper_subagent_wait_example.json');
    const stdout = execFileSync(process.execPath, [wrapperPath, '--compact', '--input', realismInput], {
      cwd: realismWorkspace,
      encoding: 'utf8',
    });
    const output = JSON.parse(stdout);
    assert.equal(typeof output.externalizedCheckpointPath, 'string', 'realism: missing externalizedCheckpointPath');
    assert.ok(output.externalizedCheckpointPath.length > 0, 'realism: empty externalizedCheckpointPath');
    const artifactPath = path.join(realismWorkspace, output.externalizedCheckpointPath);
    assert.ok(fs.existsSync(artifactPath), `realism: checkpoint artifact missing at ${artifactPath}`);
    const artifactBody = fs.readFileSync(artifactPath, 'utf8');
    assert.ok(artifactBody.trim().length > 0, 'realism: checkpoint artifact should be readable and non-empty');
    assert.equal('task_name' in (output.progressEvidence ?? {}), false, 'realism: progressEvidence must not include task_name fallback');
    assert.equal(artifactBody.includes('Wait for delegated log survey'), false, 'realism: checkpoint artifact must not fall back to taskRecord.task_name');
    results.push({
      name: 'real checkpoint artifact',
      output: {
        classification: output.classification,
        silentCandidate: output.silentCandidate,
        silentLaunchOk: output.silentLaunchOk,
        requiredNextAction: output.requiredNextAction,
        handoff: output.handoff,
      },
    });
  } finally {
    fs.rmSync(realismWorkspace, { recursive: true, force: true });
  }

  assertErrorCase('invalid json', ['--compact'], 'INVALID_JSON: input must be valid JSON', 'not-json\n');
  assertErrorCase('missing input value', ['--input'], 'CLI_ERROR: --input requires a value');
  assertErrorCase('unknown argument', ['--bogus'], 'CLI_ERROR: unknown argument: --bogus');

  const summary = {
    passed: results.length,
    fixtures: results.map(({ name, output }) => ({
      name,
      classification: output.classification,
      silentCandidate: output.silentCandidate,
      silentLaunchOk: output.silentLaunchOk,
      requiredNextAction: output.requiredNextAction,
      handoffMode: output.handoff.mode,
    })),
    errorCases: 3,
  };

  process.stdout.write(JSON.stringify(summary, null, 2) + '\n');
}

main();