diff --git a/package.json b/package.json
index 8d2b07d..0ea5957 100644
--- a/package.json
+++ b/package.json
@@ -15,6 +15,9 @@
     "demo:setup": "node scripts/demo-setup.js",
     "demo": "node examples/awf-demo/src/run-demo.js",
     "demo:cross-runtime": "cat examples/cross-runtime/terminal-demo.md",
+    "authorize:blocked":    "node tools/authorize-task/authorize-task.js --task-class db_migration --risk-lane high --runtime codex --workspace awf-demo",
+    "authorize:authorized": "node tools/authorize-task/authorize-task.js --task-class ui_refactor --risk-lane low --runtime cursor --workspace awf-demo",
+    "authorize:supervised": "node tools/authorize-task/authorize-task.js --task-class api_change --risk-lane high --runtime claude_code --workspace awf-demo",
     "test": "npm run test --workspaces --if-present"
   },
   "dependencies": {
diff --git a/tools/authorize-task/authorize-task.js b/tools/authorize-task/authorize-task.js
new file mode 100644
index 0000000..6d923b0
--- /dev/null
+++ b/tools/authorize-task/authorize-task.js
@@ -0,0 +1,297 @@
+#!/usr/bin/env node
+// AWF Runtime Authorization — Sprint 3 demo CLI.
+//
+// Simulates the Trust-Based Runtime Authorization algorithm from
+// AWF Sprint Plan v4.4.2. Hard-codes a small library of task risk
+// profiles and per-runtime trust capability profiles, then prints
+// the AUTHORIZED | SUPERVISED | BLOCKED decision with the same
+// fields the production decision record carries:
+//
+//   - task risk profile (5 dimensions, composite, risk lane)
+//   - trust capability profile (tier, sessions, evidence strength)
+//   - decision + required vs current tier
+//   - reason text and D1-D4 evidence summary
+//   - required controls
+//   - recommended alternative runtime (when blocked)
+//   - audit event type the auth service would emit
+//
+// Usage:
+//   node tools/authorize-task/authorize-task.js \
+//     --task-class db_migration \
+//     --risk-lane high \
+//     --runtime codex \
+//     --workspace awf-demo
+
+const TIER_ORDER = ['PROVISIONAL', 'RESTRICTED', 'STANDARD', 'HIGH'];
+
+// Per-task-class risk dimensions (each 1-5). Composite = sum.
+const TASK_RISK_PROFILES = {
+  ui_refactor:  { complexity: 1, blast: 2, security: 1, concurrency: 1, business: 2 },
+  api_change:   { complexity: 4, blast: 4, security: 3, concurrency: 3, business: 3 },
+  db_migration: { complexity: 4, blast: 5, security: 3, concurrency: 3, business: 5 },
+  security_fix: { complexity: 3, blast: 3, security: 5, concurrency: 2, business: 4 }
+};
+
+// Per-runtime trust capability, keyed by task_class.
+// Profiles with no D1-D4 fields render as "no history" (E0, n=0).
+const TRUST_PROFILES = {
+  cursor: {
+    ui_refactor: {
+      tier: 'HIGH', sessions: 47, evidence: 'E3',
+      d1: '92 (n=47)', d2: '88', d3: '100 (zero violations)', d4: '100 (no recurrence)'
+    },
+    api_change: {
+      tier: 'STANDARD', sessions: 12, evidence: 'E2',
+      d1: '78 (n=12)', d2: '70', d3: '95', d4: '90'
+    },
+    db_migration: { tier: 'PROVISIONAL', sessions: 0, evidence: 'E0' },
+    security_fix: { tier: 'PROVISIONAL', sessions: 0, evidence: 'E0' }
+  },
+  codex: {
+    ui_refactor: {
+      tier: 'STANDARD', sessions: 18, evidence: 'E2',
+      d1: '81 (n=18)', d2: '72', d3: '98', d4: '94'
+    },
+    api_change: {
+      tier: 'RESTRICTED', sessions: 4, evidence: 'E1',
+      d1: '74 (n=4)', d2: '60', d3: '95', d4: 'insufficient (n<5)'
+    },
+    db_migration: { tier: 'PROVISIONAL', sessions: 0, evidence: 'E0' },
+    security_fix: { tier: 'PROVISIONAL', sessions: 0, evidence: 'E0' }
+  },
+  claude_code: {
+    ui_refactor: {
+      tier: 'HIGH', sessions: 62, evidence: 'E3',
+      d1: '94 (n=62)', d2: '90', d3: '100', d4: '100'
+    },
+    api_change: {
+      tier: 'STANDARD', sessions: 22, evidence: 'E3',
+      d1: '86 (n=22)', d2: '81', d3: '98', d4: '95',
+      recent_failure: 'breaking_change_without_version_bump'
+    },
+    db_migration: {
+      tier: 'STANDARD', sessions: 14, evidence: 'E3',
+      d1: '82 (n=14)', d2: '78', d3: '100', d4: '95'
+    },
+    security_fix: {
+      tier: 'STANDARD', sessions: 15, evidence: 'E3',
+      d1: '85 (n=15)', d2: '80', d3: '100', d4: '95'
+    }
+  }
+};
+
+// (task_class, lane) → controls that must accompany a SUPERVISED
+// decision, or that would have been required for a BLOCKED one.
+const CONTROLS = {
+  db_migration: {
+    critical: [
+      'Human pre-approval required',
+      'Migration plan review',
+      'Schema validation',
+      'Rollback plan attached'
+    ]
+  },
+  api_change: {
+    high: [
+      'pre-tool-call hook on schema files',
+      'tech-lead review before PR',
+      'contract tests must pass'
+    ]
+  }
+};
+
+// Display-only names for runtime (sentence-start in reason text)
+// and task-class plural (for the BLOCKED critical reason sentence).
+const RUNTIME_DISPLAY = {
+  cursor: 'Cursor',
+  codex: 'Codex',
+  claude_code: 'Claude Code'
+};
+const TASK_DISPLAY_PLURAL = {
+  ui_refactor:  'UI refactors',
+  api_change:   'API changes',
+  db_migration: 'database migrations',
+  security_fix: 'security fixes'
+};
+
+function parseArgs(argv) {
+  const args = {};
+  for (let i = 0; i < argv.length; i++) {
+    if (argv[i].startsWith('--')) {
+      args[argv[i].slice(2)] = argv[i + 1];
+      i++;
+    }
+  }
+  return args;
+}
+
+function laneFor(composite) {
+  if (composite <= 9)  return 'low';
+  if (composite <= 14) return 'medium';
+  if (composite <= 19) return 'high';
+  return 'critical';
+}
+
+function laneLabel(lane) {
+  return { low: 'Low', medium: 'Medium', high: 'High', critical: 'Critical' }[lane];
+}
+
+function requiredTierFor(lane) {
+  return { low: 'RESTRICTED', medium: 'STANDARD', high: 'STANDARD', critical: 'HIGH' }[lane];
+}
+
+const tierIdx = t => TIER_ORDER.indexOf(t);
+
+function decide(lane, currentTier, requiredTier) {
+  // Critical lane requires human approval, which runtime auth alone cannot grant.
+  if (lane === 'critical') return 'BLOCKED';
+  if (tierIdx(currentTier) < tierIdx(requiredTier)) return 'BLOCKED';
+  if (lane === 'high' && tierIdx(currentTier) < tierIdx('HIGH')) return 'SUPERVISED';
+  return 'AUTHORIZED';
+}
+
+function recommendRuntime(taskClass, excluding) {
+  let best = null, bestIdx = -1;
+  for (const [runtime, profiles] of Object.entries(TRUST_PROFILES)) {
+    if (runtime === excluding) continue;
+    const p = profiles[taskClass];
+    if (!p) continue;
+    const idx = tierIdx(p.tier);
+    if (idx > bestIdx) { bestIdx = idx; best = { runtime, tier: p.tier }; }
+  }
+  return best;
+}
+
+function auditEvent(decision) {
+  return {
+    AUTHORIZED: 'RUNTIME_AUTHORIZATION_GRANTED',
+    SUPERVISED: 'RUNTIME_AUTHORIZATION_GRANTED_WITH_CONTROLS',
+    BLOCKED:    'RUNTIME_AUTHORIZATION_BLOCKED'
+  }[decision];
+}
+
+function reasonLines({ decision, runtime, taskClass, lane, currentTier, requiredTier }) {
+  const name = RUNTIME_DISPLAY[runtime] ?? runtime;
+  if (decision === 'BLOCKED') {
+    if (lane === 'critical') {
+      return [
+        `${name} has not earned sufficient trust for ${laneLabel(lane)} lane`,
+        `${TASK_DISPLAY_PLURAL[taskClass] ?? taskClass} in this workspace.`
+      ];
+    }
+    return [
+      `${name} holds ${currentTier} tier for ${taskClass},`,
+      `below the ${requiredTier} requirement for ${laneLabel(lane)} lane execution.`
+    ];
+  }
+  if (decision === 'SUPERVISED') {
+    return [
+      `${name} holds ${currentTier} tier for ${taskClass}, sufficient`,
+      `for ${laneLabel(lane)} lane execution under control wrappers.`
+    ];
+  }
+  return [
+    `${name} holds ${currentTier} tier for ${taskClass}, exceeding`,
+    `the ${requiredTier} requirement for ${laneLabel(lane)} lane execution.`
+  ];
+}
+
+// Column alignment helpers. Values land at column 28 in all sections.
+const head = (label, value) => `${label.padEnd(16)}${value}`;
+const row  = (label, value) => `  ${label.padEnd(26)}${value}`;
+const ev   = (label, value) => `    ${label.padEnd(24)}${value}`;
+const DIV  = '─────────────────────────────────────────────';
+
+function main() {
+  const args = parseArgs(process.argv.slice(2));
+  const taskClass = args['task-class'];
+  const riskLaneArg = args['risk-lane'];
+  const runtime = args.runtime;
+  const workspace = args.workspace;
+
+  const risk = TASK_RISK_PROFILES[taskClass];
+  if (!risk) { console.error(`unknown task class: ${taskClass}`); process.exit(1); }
+  const trust = TRUST_PROFILES[runtime]?.[taskClass];
+  if (!trust) { console.error(`unknown runtime/task pair: ${runtime}/${taskClass}`); process.exit(1); }
+
+  const composite = risk.complexity + risk.blast + risk.security
+                  + risk.concurrency + risk.business;
+  const lane = laneFor(composite);
+  const requiredTier = requiredTierFor(lane);
+  const decision = decide(lane, trust.tier, requiredTier);
+
+  const out = [];
+  out.push('AWF Runtime Authorization');
+  out.push(DIV);
+  out.push(head('Task class:', taskClass));
+  out.push(head('Risk lane:', riskLaneArg));
+  out.push(head('Candidate:', runtime));
+  out.push(head('Workspace:', workspace));
+  out.push('');
+  out.push('Task Risk Profile:');
+  out.push(row('Code complexity:',      risk.complexity));
+  out.push(row('Blast radius:',         risk.blast));
+  out.push(row('Security sensitivity:', risk.security));
+  out.push(row('Concurrency risk:',     risk.concurrency));
+  out.push(row('Business criticality:', risk.business));
+  out.push(row('Composite score:',      `${composite}  →  ${laneLabel(lane)} lane`));
+  out.push('');
+  out.push('Trust Capability Profile:');
+  out.push(row('Runtime:',            runtime));
+  out.push(row('Task class:',         taskClass));
+  out.push(row('Current trust tier:', trust.tier));
+  out.push(row('Sessions:',           trust.sessions));
+  out.push(row('Evidence strength:',  trust.evidence));
+  out.push('');
+  out.push('Authorization Decision:');
+  out.push(row('Decision:',            decision));
+  out.push(row('Required trust tier:',
+    decision === 'SUPERVISED' ? `${requiredTier} + controls` : requiredTier));
+  out.push(row('Current trust tier:',  trust.tier));
+  out.push('  ');
+  out.push('  Reason:');
+  reasonLines({ decision, runtime, taskClass, lane, currentTier: trust.tier, requiredTier })
+    .forEach(l => out.push(`  ${l}`));
+  out.push('  ');
+  out.push('  Evidence:');
+  out.push(ev('D1 Correctness:',   trust.d1 ?? 'no history'));
+  out.push(ev('D2 Observability:', trust.d2 ?? 'no history'));
+  out.push(ev('D3 Policy:',        trust.d3 ?? 'no history'));
+  out.push(ev('D4 Recurrence:',    trust.d4 ?? 'no history'));
+  if (trust.recent_failure) {
+    out.push('');
+    out.push(row('Recent failure pattern:', trust.recent_failure));
+  }
+  out.push('');
+
+  const controls = CONTROLS[taskClass]?.[lane];
+  if (decision === 'BLOCKED') {
+    if (controls) {
+      out.push('  Required controls:');
+      controls.forEach(c => out.push(`    - ${c}`));
+      out.push('');
+    }
+    const rec = recommendRuntime(taskClass, runtime);
+    if (rec) {
+      out.push(row('Recommended runtime:', rec.runtime));
+      out.push(`  Note: ${rec.runtime} holds ${rec.tier} tier for ${taskClass}`);
+      out.push(`  but ${laneLabel(lane)} lane requires HIGH. No runtime is authorized`);
+      out.push(`  for ${laneLabel(lane)} lane without human approval regardless of tier.`);
+      out.push('');
+    }
+  } else if (decision === 'SUPERVISED' && controls) {
+    out.push('  Required controls:');
+    controls.forEach(c => out.push(`    - ${c}`));
+    out.push('');
+  } else if (decision === 'AUTHORIZED') {
+    out.push('  Required controls:');
+    out.push('    None beyond standard merge approval');
+    out.push('');
+  }
+
+  out.push(row('Audit event:', auditEvent(decision)));
+  out.push(DIV);
+  console.log(out.join('\n'));
+}
+
+main();