feat: add ThoughtProof verification plugin (Sentinel + RV)

examples/thoughtproof-verification/index.ts

@@ -0,0 +1,256 @@
+/**
+ * ThoughtProof Verification Agent — Demo for GOAT AgentKit
+ *
+ * Demonstrates the full Sentinel → RV escalation flow:
+ *
+ *   Agent decides → Sentinel (fast triage, every decision)
+ *     ├── ALLOW  → Execute action
+ *     ├── BLOCK  → Stop, log reason
+ *     └── UNCERTAIN → Escalate to RV (deep verification)
+ *                       ├── ALLOW → Execute
+ *                       └── BLOCK → Stop
+ *
+ * After verification, optionally attest the result on-chain.
+ *
+ * Run:  npx tsx examples/thoughtproof-verification/index.ts
+ * Env:  THOUGHTPROOF_API_KEY (optional — public endpoints work without key)
+ */
+
+// NOTE: These imports use monorepo-relative paths. If you copy this example
+// into your own project, replace with package imports:
+//   import { ActionProvider } from '@goatnetwork/agentkit/providers';
+//   import { PolicyEngine } from '@goatnetwork/agentkit/core';
+//   import { ... } from '@goatnetwork/agentkit/plugins';
+import { ActionProvider } from '../../providers/action-provider';
+import { PolicyEngine } from '../../core/policy/policy-engine';
+import { ExecutionRuntime } from '../../core/runtime/execution-runtime';
+import {
+  thoughtproofSentinelAction,
+  thoughtproofVerifyAction,
+  thoughtproofAttestAction,
+  thoughtproofStatusAction,
+  HttpThoughtProofAdapter,
+} from '../../plugins/thoughtproof/index';
+import type {
+  SentinelVerifyOutput,
+  RVVerifyOutput,
+  StatusOutput,
+} from '../../plugins/thoughtproof/adapters/types';
+
+// ── Configuration ───────────────────────────────────────────
+
+const adapter = new HttpThoughtProofAdapter({
+  apiKey: process.env.THOUGHTPROOF_API_KEY,
+});
+
+const provider = new ActionProvider();
+provider.register(thoughtproofSentinelAction(adapter));
+provider.register(thoughtproofVerifyAction(adapter));
+provider.register(thoughtproofAttestAction(adapter));
+provider.register(thoughtproofStatusAction(adapter));
+
+const policy = new PolicyEngine({
+  allowedNetworks: ['goat-mainnet', 'goat-testnet'],
+  maxRiskWithoutConfirm: 'medium',
+  writeEnabled: true,
+});
+
+const runtime = new ExecutionRuntime(policy, {
+  maxRetries: 1,
+  retryDelayMs: 500,
+});
+
+// ── Helpers ─────────────────────────────────────────────────
+
+function makeContext(traceId: string) {
+  return {
+    traceId,
+    network: 'goat-mainnet',
+    now: Date.now(),
+    caller: 'thoughtproof-demo-agent',
+  };
+}
+
+function logSection(title: string) {
+  console.log(`\n${'═'.repeat(60)}`);
+  console.log(`  ${title}`);
+  console.log('═'.repeat(60));
+}
+
+function logVerdict(source: string, verdict: string, confidence: number, latencyMs: number) {
+  const icon = verdict === 'ALLOW' ? '✅' : verdict === 'BLOCK' ? '🛑' : '⚠️';
+  console.log(`  ${icon} [${source}] ${verdict} (confidence: ${confidence}, ${latencyMs}ms)`);
+}
+
+// ── Core: Sentinel → RV Escalation ─────────────────────────
+
+interface VerificationResult {
+  finalVerdict: 'ALLOW' | 'BLOCK';
+  source: 'sentinel' | 'rv';
+  requestId: string;
+  details: SentinelVerifyOutput | RVVerifyOutput;
+}
+
+async function verifyAgentDecision(
+  claim: string,
+  context: string,
+  traceId: string,
+): Promise<VerificationResult> {
+  const ctx = makeContext(traceId);
+
+  // Step 1: Sentinel pre-check (~$0.003, ~2s)
+  console.log('\n  → Running Sentinel pre-check...');
+  const sentinelResult = await runtime.run(
+    provider.get('thoughtproof.sentinel'),
+    ctx,
+    { claim, context },
+  );
+
+  if (!sentinelResult.ok || !sentinelResult.output) {
+    console.log(`  ❌ Sentinel error: ${sentinelResult.error}`);
+    return { finalVerdict: 'BLOCK', source: 'sentinel', requestId: '', details: {} as SentinelVerifyOutput };
+  }
+
+  const sentinel = sentinelResult.output as SentinelVerifyOutput;
+  logVerdict('Sentinel', sentinel.verdict, sentinel.confidence, sentinel.latencyMs);
+
+  if (sentinel.reasons.length) {
+    sentinel.reasons.forEach((r) => console.log(`     └─ ${r}`));
+  }
+
+  if (sentinel.verdict === 'ALLOW') {
+    return { finalVerdict: 'ALLOW', source: 'sentinel', requestId: sentinel.requestId, details: sentinel };
+  }
+
+  if (sentinel.verdict === 'BLOCK') {
+    return { finalVerdict: 'BLOCK', source: 'sentinel', requestId: sentinel.requestId, details: sentinel };
+  }
+
+  // Step 2: UNCERTAIN → escalate to RV (~$0.02, ~5-15s)
+  console.log('\n  → Sentinel UNCERTAIN — escalating to RV deep verification...');
+  const rvResult = await runtime.run(
+    provider.get('thoughtproof.verify'),
+    ctx,
+    { claim, context, tier: 'standard' as const },
+  );
+
+  if (!rvResult.ok || !rvResult.output) {
+    console.log(`  ❌ RV error: ${rvResult.error}`);
+    return { finalVerdict: 'BLOCK', source: 'rv', requestId: '', details: {} as RVVerifyOutput };
+  }
+
+  const rv = rvResult.output as RVVerifyOutput;
+  logVerdict('RV', rv.verdict, rv.confidence, rv.latencyMs);
+
+  if (rv.synthesis) {
+    console.log(`     └─ Synthesis: ${rv.synthesis.slice(0, 200)}${rv.synthesis.length > 200 ? '...' : ''}`);
+  }
+
+  return {
+    finalVerdict: rv.verdict === 'ALLOW' ? 'ALLOW' : 'BLOCK',
+    source: 'rv',
+    requestId: rv.requestId,
+    details: rv,
+  };
+}
+
+// ── Demo Scenarios ──────────────────────────────────────────
+
+interface Scenario {
+  name: string;
+  claim: string;
+  context: string;
+  expectedBehavior: string;
+}
+
+const SCENARIOS: Scenario[] = [
+  {
+    name: 'Safe trade — clear signal',
+    claim: 'Based on the 4h MACD crossover confirmed by volume increase of 340%, and price breaking above the 200-day moving average, execute a limit buy of 0.3 ETH at $3,450. This is within our 2% position size limit.',
+    context: 'Conservative trading agent. Max position: 1 ETH. Max drawdown: 5%. Only trades on confirmed multi-indicator signals.',
+    expectedBehavior: 'Should ALLOW — clear reasoning, within risk parameters, multiple confirming indicators.',
+  },
+  {
+    name: 'Risky trade — reasoning gaps',
+    claim: 'ETH will pump because of the upcoming merge. All in. Buy 50 ETH at market price immediately.',
+    context: 'Conservative trading agent. Max position: 1 ETH. Max drawdown: 5%. Only trades on confirmed multi-indicator signals.',
+    expectedBehavior: 'Should BLOCK — violates position limits, no technical analysis, uses "will pump" (speculative), "all in" (emotional).',
+  },
+  {
+    name: 'Ambiguous — needs deeper analysis',
+    claim: 'Correlation between BTC dominance declining and ETH outperformance suggests a rotation. Place a 0.5 ETH buy order 2% below current price as a swing trade.',
+    context: 'Moderate risk trading agent. Position limits respected. However the correlation analysis is based on a 30-day window which may be insufficient.',
+    expectedBehavior: 'May trigger UNCERTAIN → RV escalation — claim is plausible but methodology could be questioned.',
+  },
+];
+
+// ── Main ────────────────────────────────────────────────────
+
+async function main() {
+  console.log('🐐 ThoughtProof × GOAT AgentKit — Verification Demo');
+  console.log('   Sentinel (triage) → RV (deep verification) escalation flow\n');
+
+  // Step 0: Health check
+  logSection('Health Check');
+  const statusResult = await runtime.run(
+    provider.get('thoughtproof.status'),
+    makeContext('health-check'),
+    {},
+  );
+
+  if (statusResult.ok && statusResult.output) {
+    const status = statusResult.output as StatusOutput;
+    console.log(`  Sentinel: ${status.sentinel.healthy ? '✅ healthy' : '❌ down'} (${status.sentinel.latencyMs}ms)`);
+    console.log(`  RV:       ${status.rv.healthy ? '✅ healthy' : '❌ down'} (${status.rv.latencyMs}ms)`);
+
+    if (!status.sentinel.healthy && !status.rv.healthy) {
+      console.log('\n  ⚠️  Both APIs unreachable. Running in demo mode (expect errors).');
+    }
+  } else {
+    console.log(`  ⚠️  Status check failed: ${statusResult.error}`);
+    console.log('  Running scenarios anyway...');
+  }
+
+  // Run scenarios
+  for (let i = 0; i < SCENARIOS.length; i++) {
+    const scenario = SCENARIOS[i];
+    logSection(`Scenario ${i + 1}: ${scenario.name}`);
+    console.log(`  Expected: ${scenario.expectedBehavior}`);
+    console.log(`  Claim: "${scenario.claim.slice(0, 100)}..."`);
+
+    try {
+      const result = await verifyAgentDecision(
+        scenario.claim,
+        scenario.context,
+        `demo-scenario-${i + 1}`,
+      );
+
+      console.log(`\n  📋 Final Decision:`);
+      console.log(`     Verdict: ${result.finalVerdict}`);
+      console.log(`     Decided by: ${result.source}`);
+      console.log(`     Request ID: ${result.requestId}`);
+
+      if (result.finalVerdict === 'ALLOW') {
+        console.log('     → Agent would EXECUTE the trade');
+      } else {
+        console.log('     → Agent would SKIP the trade');
+      }
+    } catch (err) {
+      console.log(`  ❌ Scenario failed: ${err instanceof Error ? err.message : String(err)}`);
+    }
+  }
+
+  // Summary
+  logSection('Demo Complete');
+  console.log('  The ThoughtProof plugin provides:');
+  console.log('  • Sentinel: Fast pre-execution triage (~$0.003, ~2s)');
+  console.log('  • RV: Adversarial deep verification (~$0.02-0.08, 5-45s)');
+  console.log('  • Attest: On-chain attestation (EAS / TP-VC)');
+  console.log('  • Automatic ERC-8004 reputation feedback (internal hook)');
+  console.log('');
+  console.log('  Sentinel catches obvious issues fast and cheap.');
+  console.log('  RV handles the edge cases with multi-model adversarial analysis.');
+  console.log('  Together: cost-efficient verification for every agent decision.\n');
+}
+
+main().catch(console.error);

plugins/index.ts

@@ -27,3 +27,4 @@ export * from './erc8004/index';
 export * from './x402-merchant/index';
 export * from './giftcard/index';
 export * from './gns/index';
+export * from './thoughtproof/index';