Skip to content

demo-script.md

Latest commit

 

History

History
153 lines (110 loc) · 4.92 KB

demo-script.md

File metadata and controls

153 lines (110 loc) · 4.92 KB

AgentSafe Demo Script

Step-by-step walkthrough for ETHDenver judges (5–7 minutes).

Pre-Demo Setup

# Terminal 1: Backend
cd apps/backend && pnpm dev       # http://localhost:4000

# Terminal 2: Frontend
cd apps/web && pnpm dev           # http://localhost:3000

Verify both are running:

curl http://localhost:4000/health | jq

Demo Flow (5–7 minutes)

Step 1: Dashboard Overview (30s)

Open http://localhost:3000/dashboard

What to say:

"This is AgentSafe — an ERC-4337 smart wallet on Base protected by SwarmGuard, a multi-agent AI defense system. The dashboard shows real-time swarm status, governance proposals, and integration health."

Point out:

  • Swarm status (alive/agents online)
  • Proposals count
  • Sponsor integration summary

Step 2: SwarmGuard Defense Demo (2 min) ⭐

Navigate to http://localhost:3000/defense

What to say:

"Let's see SwarmGuard in action. I'll submit a suspicious transaction — an unlimited ERC-20 approval to an unknown contract."

  1. Fill in the form (defaults are pre-set for a suspicious approval):
    • Chain ID: 8453 (Base)
    • To: 0xdead000000000000000000000000000000000000
    • Data: 0x095ea7b3 (approve selector)
    • Kind: APPROVAL
  2. Click "Evaluate Transaction"
  3. Wait for results (~1s)

What to show:

  • Agent report timeline: Each agent (Sentinel, Scam Detector, MEV Watcher, Liquidation) produces an independent risk assessment
  • Point out severity, riskScore, confidence, and reasons for each agent
  • Click "Evidence" to expand raw evidence JSON
  • Show the Consensus Card: ALLOW/BLOCK/REVIEW_REQUIRED with final severity
  • Show the Intent Card: The system produces an ActionIntent (EXECUTE_TX or BLOCK_TX)
  • Click "Execute on Base" → Shows "Simulated (MVP)" confirmation

What to say about provenance:

"Each agent signs its risk report independently. The consensus decision is recorded as a provenance receipt — this maps directly to our ProvenanceRegistry on Base. In production, the agent approval hash is stored on-chain before any UserOp executes."

Step 3: Governance Demo (1.5 min) ⭐

Navigate to http://localhost:3000/governance

What to say:

"GovernanceSafe monitors DAO proposals and provides AI-powered risk analysis with policy checks."

  1. Point out the loaded proposals (fetched from backend)
  2. Click "Get AI Recommendation" on any proposal
  3. Show the VoteIntent result:
    • Recommendation (FOR/AGAINST/ABSTAIN)
    • Confidence percentage
    • Reasons
    • Policy checks (TREASURY_RISK, GOV_POWER_SHIFT, URGENCY_FLAG)
  4. Toggle "Auto-vote enabled" → Show the warning
  5. Click "Human Veto" → Shows VETOED state

What to say:

"The system uses Kite AI for summarisation and a keyword policy engine for risk checks. Auto-voting has a mandatory human veto window — no on-chain vote executes without final human approval."

Step 4: Sponsor Proof Panel (1.5 min) ⭐⭐⭐

Navigate to http://localhost:3000/integrations

What to say:

"Here's our integration proof panel. Each sponsor technology has verifiable evidence."

Walk through each section:

  1. Base — Show chain ID 8453, deployed contract addresses (AgentSafeAccount, PolicyEngine, GovernanceModule, ProvenanceRegistry)
  2. QuickNode — Show health check status, mode (live/disabled), blockNumber if available
  3. Kite AI — Show mode (live/stub), click "Run Kite Summary Test" to verify pipeline
  4. Nouns / Proposals — Show proposals loaded count + preview
  5. 0g — Mention as stretch goal
  6. Expand "Raw Proof Data" to show raw /health and /status JSON

Step 5: Policy Engine (30s)

Navigate to http://localhost:3000/policy

What to say:

"Policies are deterministic guardrails that AI agents cannot override. They're enforced at three layers: server-side consensus, PolicyEngine.sol on-chain, and ERC-4337 UserOp validation."

Show:

  • Consensus rules (approvals required, critical block)
  • Policy simulator with mock agent reports

Fallback: Backend Down

If the backend is unreachable during the demo:

  1. All pages show friendly error banners (not crashes)
  2. Navigate to /integrations — shows ❌ Missing badges with clear explanations
  3. Point out the architecture: "The UI gracefully degrades — all API calls have timeout handling and error states"
  4. Use curl to show static responses if backend recovers

Key Technical Talking Points

  • ERC-4337 Account Abstraction: Smart wallet on Base with bundler-compatible UserOps
  • Multi-Agent Consensus: 4 specialist agents + coordinator, deterministic voting threshold
  • Provenance Registry: On-chain record of every swarm decision (agent hashes + final verdict)
  • Policy Engine: Immutable on-chain guardrails (spending limits, approval blocking, allowlists)
  • GovernanceSafe: Proposal parsing → AI risk analysis → VoteIntent → human veto → execution