reporting-governance-plugin/scripts/test_force_recall_long_task_preflight.mjs

#!/usr/bin/env node
import assert from 'node:assert/strict';
import fs from 'node:fs/promises';
import os from 'node:os';
import path from 'node:path';
import { pathToFileURL } from 'node:url';
import { stripTypeScriptTypes } from 'node:module';

const __dirname = path.dirname(new URL(import.meta.url).pathname);
const repoRoot = path.resolve(__dirname, '..');
const handlerPath = path.join(repoRoot, 'hooks', 'force-recall', 'handler.ts');
const wrapperPath = path.join(repoRoot, 'scripts', 'long_task_governor_wrapper.mjs');
const gateLockPath = path.join(repoRoot, 'scripts', 'long_task_gate_lock.mjs');
const plannerPath = path.join(repoRoot, 'scripts', 'plan_long_task_auto_chain.mjs');

async function importTsModule(tsPath) {
  const source = await fs.readFile(tsPath, 'utf8');
  const jsSource = stripTypeScriptTypes(source, { mode: 'strip' });
  const dataUrl = `data:text/javascript;charset=utf-8,${encodeURIComponent(jsSource)}\n//# sourceURL=${encodeURIComponent(pathToFileURL(tsPath).href)}`;
  return import(dataUrl);
}

function escapeRegex(snippet) {
  return snippet.replace(/[.*+?^${}()|[\]\\]/g, '\\$&');
}

async function runScenario(forceRecall, requestText) {
  const event = {
    type: 'message',
    action: 'preprocessed',
    context: {
      workspaceDir: repoRoot,
      body: requestText,
      bodyForAgent: requestText,
    },
  };

  await forceRecall(event);
  const injected = event.context?.bodyForAgent;
  assert.equal(typeof injected, 'string', 'event.context.bodyForAgent should be a string after handler runs');
  return injected;
}

async function withPatchedWrapper(tempContent, callback) {
  const originalWrapper = await fs.readFile(wrapperPath, 'utf8');
  await fs.writeFile(wrapperPath, tempContent, 'utf8');
  try {
    return await callback();
  } finally {
    await fs.writeFile(wrapperPath, originalWrapper, 'utf8');
  }
}

function buildWrapperScript(wrapperResult) {
  return `#!/usr/bin/env node\nprocess.stdout.write(JSON.stringify(${JSON.stringify(wrapperResult)}, null, 0) + "\\n");\n`;
}

async function main() {
  await Promise.all([fs.access(wrapperPath), fs.access(gateLockPath), fs.access(plannerPath)]);
  const { default: forceRecall } = await importTsModule(handlerPath);
  assert.equal(typeof forceRecall, 'function', 'force-recall handler should export default function');

  const requestText = [
    'Please inspect the workspace files and verify the hook injection path.',
    'I need you to review the behavior, choose the final accept/reject decision,',
    'and continue in background with a follow-up later.',
  ].join(' ');

  const injected = await runScenario(forceRecall, requestText);

  const expectedSnippets = [
    '[LONG_TASK_GOVERNOR_PREFLIGHT]',
    'classification=long_task',
    'silentLaunchOk=false',
    'handoff.mode=button_path',
    '[LONG_TASK_GATE_LOCK]',
    'gateStatus=fail',
    '[LONG_TASK_AUTO_CHAIN_PLAN]',
    'plannerStatus=blocked_by_gate',
    'derivedAction=none',
    'dispatchMode=no_dispatch',
    'autoChainAllowed=false',
    'reason=gateStatus must pass before auto-chain planning can proceed',
    'requiredEvidence=gateStatus=pass',
    'requiredEvidence=externalizedCheckpoint',
    'requiredEvidence=concreteNextAction',
    'requiredEvidence=buttonPathMode',
    'reason=silent long-task cannot continue without externalized checkpoint path',
    'reason=claimed execution requires evidence of a concrete next action',
    'reason=owner decision flow must end in button-path, not plain text',
    'ENFORCEMENT: Hook input should include progressEvidence (or equivalent concrete fields) whenever a progression claim is present.',
    'HARD_GATE: Block any plain-text handoff or silent-continuation claim when externalized checkpoint evidence is missing.',
    'HARD_GATE: If owner decision is involved, do not replace button-path closure with plain-text handoff.',
    'ENFORCEMENT: Forbidden path: plain-text handoff that pretends the long task is already continuing without an externalized checkpoint.',
    'ENFORCEMENT: Forbidden path: stating you have already entered the next task/step when the record only contains planning language and no concrete execution evidence.',
  ];

  const unexpectedSnippets = [
    'reason=claimed progression without concrete progress evidence is forbidden',
    'requiredEvidence=progressEvidence',
  ];

  for (const snippet of expectedSnippets) {
    assert.match(injected, new RegExp(escapeRegex(snippet)), `missing snippet: ${snippet}`);
  }
  for (const snippet of unexpectedSnippets) {
    assert.doesNotMatch(injected, new RegExp(escapeRegex(snippet)), `unexpected snippet present: ${snippet}`);
  }

  const { evaluateGate } = await import(pathToFileURL(gateLockPath).href + `?t=${Date.now()}`);
  assert.equal(typeof evaluateGate, 'function', 'long_task_gate_lock should export evaluateGate for direct tests');

  const passResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'dispatch_follow_up_subagent',
    autoChainNextAction: 'dispatch_follow_up_subagent',
    autoChainDispatchEvidence: {
      action: 'dispatch_follow_up_subagent',
      dispatched: true,
      event: 'dispatch',
    },
    progressionClaim: 'already progressing to the next step in background',
    progressEvidence: { sessionKey: 'task-123' },
  });
  assert.equal(passResult.gateStatus, 'pass', 'pass-path should pass with concrete progressEvidence');

  const failResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'dispatch_follow_up_subagent',
    autoChainNextAction: 'dispatch_follow_up_subagent',
    progressionClaim: 'already progressing to the next step in background',
    executionEvidence: { concreteNextAction: 'dispatch_follow_up_subagent' },
  });
  assert.equal(failResult.gateStatus, 'fail', 'fail-path should fail when explicit auto-chain action lacks dispatch evidence');
  assert.match(JSON.stringify(failResult), /autoChainDispatchEvidence/, 'fail-path should require autoChainDispatchEvidence');

  const neutralResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'summarize findings for reply',
    executionEvidence: { concreteNextAction: 'summarize findings for reply' },
  });
  assert.equal(neutralResult.gateStatus, 'pass', 'neutral-path should pass when there is no explicit auto-chain next action');
  assert.doesNotMatch(JSON.stringify(neutralResult), /autoChainDispatchEvidence/, 'neutral-path should not require auto-chain dispatch evidence');

  const directAutoChainFailResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'dispatch_follow_up_subagent',
    autoChainNextAction: 'dispatch_follow_up_subagent',
  });
  assert.equal(directAutoChainFailResult.gateStatus, 'fail', 'direct evaluator should fail when explicit auto-chain action has no dispatch evidence');
  assert.match(JSON.stringify(directAutoChainFailResult), /explicit auto-chain next action requires dispatched-action evidence/, 'direct evaluator fail-path should mention missing dispatched-action evidence');

  const mismatchedDispatchEvidenceResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'dispatch_follow_up_subagent',
    autoChainNextAction: 'dispatch_follow_up_subagent',
    autoChainDispatchEvidence: {
      action: 'dispatch_other_subagent',
      dispatched: true,
      event: 'dispatch',
    },
  });
  assert.equal(mismatchedDispatchEvidenceResult.gateStatus, 'fail', 'mismatched dispatch evidence should fail');
  assert.match(JSON.stringify(mismatchedDispatchEvidenceResult), /autoChainDispatchEvidence/, 'mismatched dispatch evidence should still require matching autoChainDispatchEvidence');

  const fakeCheckpointDispatchEvidenceResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'dispatch_follow_up_subagent',
    autoChainNextAction: 'dispatch_follow_up_subagent',
    autoChainDispatchEvidence: {
      sessionKey: 'task-123',
      checkpointPath: 'checkpoints/task-123.json',
    },
  });
  assert.equal(fakeCheckpointDispatchEvidenceResult.gateStatus, 'fail', 'checkpoint/session-only dispatch evidence should fail');
  assert.match(JSON.stringify(fakeCheckpointDispatchEvidenceResult), /explicit auto-chain next action requires dispatched-action evidence/, 'checkpoint/session-only dispatch evidence should be rejected as fake dispatch evidence');

  const neutralSnakeCaseResult = evaluateGate({
    classification: 'long_task',
    claimedExecution: true,
    concreteNextAction: 'summarize findings for reply',
    autoChainNextAction: 'checkpoint_session_metadata_only',
    executionEvidence: { concreteNextAction: 'summarize findings for reply' },
  });
  assert.equal(neutralSnakeCaseResult.gateStatus, 'pass', 'neutral snake_case non-dispatch action should not trigger dispatch-evidence requirement');
  assert.doesNotMatch(JSON.stringify(neutralSnakeCaseResult), /autoChainDispatchEvidence/, 'neutral snake_case non-dispatch action should not mention dispatch-evidence requirement');

  const passInjected = await withPatchedWrapper(buildWrapperScript({
    classification: 'long_task',
    silentCandidate: true,
    needsCheckpoint: true,
    needsSubagent: false,
    needsOwnerDecision: false,
    silentLaunchOk: true,
    silentLaunchReason: 'checkpoint established',
    requiredNextAction: 'dispatch_follow_up_subagent',
    autoChainDispatchEvidence: {
      action: 'dispatch_follow_up_subagent',
      dispatched: true,
      event: 'dispatch',
    },
    taskRecord: { task_name: 'task-123' },
    handoff: { mode: 'direct_reply' },
  }), async () => runScenario(forceRecall, requestText));
  assert.match(passInjected, /gateStatus=pass/, 'hook pass-path should pass when wrapper provides concrete progressEvidence');
  assert.match(passInjected, /\[LONG_TASK_AUTO_CHAIN_PLAN\]/, 'hook pass-path should emit auto-chain plan block');
  assert.match(passInjected, /plannerStatus=pass/, 'hook pass-path should expose planner pass result');
  assert.match(passInjected, /derivedAction=dispatch_spec_review/, 'hook pass-path should derive dry-run spec review dispatch');
  assert.match(passInjected, /dispatchMode=dry_run_dispatch/, 'hook pass-path should stay in dry-run dispatch mode');
  assert.match(passInjected, /autoChainAllowed=true/, 'hook pass-path should allow auto-chain in dry-run planner output');

  const failInjected = await withPatchedWrapper(buildWrapperScript({
    classification: 'long_task',
    silentCandidate: false,
    needsCheckpoint: false,
    needsSubagent: false,
    needsOwnerDecision: false,
    silentLaunchOk: false,
    requiredNextAction: 'dispatch_follow_up_subagent',
    handoff: { mode: 'direct_reply' },
  }), async () => runScenario(forceRecall, requestText));
  assert.match(failInjected, /gateStatus=fail/, 'hook fail-path should fail when wrapper exposes explicit auto-chain action without dispatch evidence');
  assert.match(failInjected, /\[LONG_TASK_AUTO_CHAIN_PLAN\]/, 'hook fail-path should emit auto-chain plan block');
  assert.match(failInjected, /plannerStatus=blocked_by_gate/, 'hook fail-path should report planner blocked by gate');
  assert.match(failInjected, /derivedAction=none/, 'hook fail-path should not derive a dry-run action');
  assert.match(failInjected, /dispatchMode=no_dispatch/, 'hook fail-path should remain no-dispatch');
  assert.match(failInjected, /autoChainAllowed=false/, 'hook fail-path should not allow auto-chain');
  assert.match(failInjected, /reason=explicit auto-chain next action requires dispatched-action evidence/, 'hook fail-path should mention missing dispatched-action evidence');
  assert.match(failInjected, /requiredEvidence=autoChainDispatchEvidence/, 'hook fail-path should require autoChainDispatchEvidence');

  const neutralInjected = await withPatchedWrapper(buildWrapperScript({
    classification: 'long_task',
    silentCandidate: false,
    needsCheckpoint: false,
    needsSubagent: false,
    needsOwnerDecision: false,
    silentLaunchOk: false,
    requiredNextAction: 'summarize findings for reply',
    handoff: { mode: 'direct_reply' },
  }), async () => runScenario(forceRecall, requestText));
  assert.match(neutralInjected, /gateStatus=pass/, 'hook neutral-path should pass when wrapper does not expose an explicit auto-chain action');
  assert.match(neutralInjected, /\[LONG_TASK_AUTO_CHAIN_PLAN\]/, 'hook neutral-path should emit auto-chain plan block');
  assert.match(neutralInjected, /plannerStatus=none/, 'hook neutral-path should report no derived auto-chain action');
  assert.match(neutralInjected, /derivedAction=none/, 'hook neutral-path should keep derivedAction as none');
  assert.match(neutralInjected, /dispatchMode=no_dispatch/, 'hook neutral-path should remain no-dispatch');
  assert.match(neutralInjected, /autoChainAllowed=false/, 'hook neutral-path should keep auto-chain disabled');
  assert.doesNotMatch(neutralInjected, /reason=explicit auto-chain next action requires dispatched-action evidence/, 'hook neutral-path should not fail on auto-chain evidence when no explicit tool action exists');

  const originalGateLock = await fs.readFile(gateLockPath, 'utf8');
  const tempDir = await fs.mkdtemp(path.join(os.tmpdir(), 'force-recall-gate-lock-'));
  const backupPath = path.join(tempDir, path.basename(gateLockPath));
  await fs.writeFile(backupPath, originalGateLock, 'utf8');
  await fs.writeFile(gateLockPath, '#!/usr/bin/env node\nprocess.exit(1);\n', 'utf8');

  let degradedInjected;
  try {
    degradedInjected = await runScenario(forceRecall, requestText);
  } finally {
    const backup = await fs.readFile(backupPath, 'utf8');
    await fs.writeFile(gateLockPath, backup, 'utf8');
    await fs.rm(tempDir, { recursive: true, force: true });
  }

  const degradedExpectedSnippets = [
    '[LONG_TASK_GATE_LOCK]',
    'gateStatus=degraded',
    'gateRequired=unknown',
    'HARD_GATE: Evaluator unavailable is not permission to claim silent continuation or next-task progression without verifiable progress evidence.',
    'HARD_GATE: Fall back to a non-silent, evidence-preserving follow-up if you cannot prove checkpoint state or concrete execution.',
  ];

  for (const snippet of degradedExpectedSnippets) {
    assert.match(degradedInjected, new RegExp(escapeRegex(snippet)), `missing degraded snippet: ${snippet}`);
  }

  process.stdout.write(JSON.stringify({
    ok: true,
    gatePaths: {
      pass: passResult.gateStatus,
      fail: failResult.gateStatus,
      neutral: neutralResult.gateStatus,
    },
    bodyPreview: injected.split('\n').slice(0, 35),
  }, null, 2) + '\n');
}

main().catch((error) => {
  console.error(error);
  process.exitCode = 1;
});