Synthesis Hackathon — Best Build Story (OpenServ)

How Humans and Agents Built Escroue

A 5-day collaboration between a human architect and an agent, building a trustless agent-to-agent marketplace with 8 protocol integrations across 2 chains.

Smart Contracts
3
Foundry Tests
40
Integrations
8
Frontend Pages
10

The Collaboration Model

Human direction + agent execution + iterative review = rapid, high-quality output.

🎯
Human Sets Direction

Architecture decisions, technology choices, hackathon strategy, and integration priorities. The human defines what to build and why — never how.

Chose Base + Celo dual-chain deployment
Identified all 14 hackathon tracks
Prioritized Venice privacy integration
🤖
Agent Executes Autonomously

The agent handles all implementation: contract development, testing, deployment, frontend, integrations, documentation. Structured by the Superpowers skills framework.

Wrote 40 Foundry tests with full coverage
Deployed contracts to 2 chains
Built 10 frontend pages in Next.js 16
🔄
Iterative Review Cycles

Human reviews output, provides feedback, and redirects. The agent iterates based on direction — not guessing. Each cycle compresses scope and improves quality.

Revised README tracks 3 times based on feedback
Replaced ASCII diagrams after human review
Updated links from relative to absolute Render URLs

Build Timeline

Day-by-day breakdown of what the human directed and what the agent built.

Day 0March 17Architecture & Foundation
🎯
HumanDefined the vision: trustless agent-to-agent marketplace with on-chain escrow
🏗️
HumanChose the stack: Solidity + Foundry, viem, Next.js 16, Base Sepolia
📝
AgentScaffolded contracts: ServiceBoard, EscrowVault, ReputationRegistry
AgentWrote 40 Foundry tests covering full lifecycle, edge cases, access control
🚀
AgentDeployed all 3 contracts to Base Sepolia
Day 1March 18Agents Come Alive
🎯
HumanDirected agent architecture: buyer/seller pattern with polling loop
🤖
AgentBuilt buyer agent: posts tasks with ETH escrow, verifies deliveries
🤖
AgentBuilt seller agent: discovers tasks, claims, executes work autonomously, delivers
🎉
AgentRan first end-to-end demo: 5 tasks posted → claimed → delivered → settled
HumanInstalled Superpowers skills framework for structured development
💻
AgentBuilt Next.js dashboard with live task board, agent profiles, event feed
Day 2March 19Multi-Chain & Integrations
🎯
HumanDecided to go multi-chain: Base Sepolia + Celo Sepolia
🔗
AgentDeployed same contracts to Celo Sepolia — identical addresses via CREATE2
🌐
AgentIntegrated OpenServ SDK: 6 capabilities, NLP chat handlers, Agent #3973
🪪
AgentRegistered ERC-8004 agent identities on both chains
🔒
AgentIntegrated Venice for private LLM inference in TEE enclaves
Day 3March 20Protocol Integrations
🎯
HumanIdentified all hackathon tracks and prioritized integrations
🏷️
AgentENS integration: agentescrow.eth + buyer/seller subdomains with 16 records each
🦊
AgentMetaMask Delegation Toolkit: ERC-7715 permission framework for agents
📦
AgentFilecoin: on-chain task proof archival via deal proposals
💬
AgentAmpersend: agent-to-agent messaging SDK for task coordination
🎨
AgentBuilt dedicated frontend pages for each integration (/base, /venice, /ens, etc.)
Day 4March 21Polish & Ship
🎯
HumanFinal review pass on all pages and README
📊
AgentMade stats dynamic: on-chain reputation, contract counts from chain config
🎨
AgentReplaced ASCII diagrams with visual card-based architecture UI
🖼️
AgentCreated SVG architecture diagrams for README
🚀
AgentDeployed frontend to Render, verified all pages live
Human Decision
Agent Execution

Key Architecture Decisions

Every major technical decision and who made it.

Human
Solidity + Foundry over Hardhat

Foundry's Rust-based toolchain is faster for testing and the agent can write fuzz tests natively. 40 tests run in under 2 seconds.

Human
viem over ethers.js

Type-safe, tree-shakeable, and better DX for the agent. Contract interactions are more explicit and less error-prone.

Human
Next.js 16 + React 19

Latest framework for the frontend. Server components where possible, client components for interactive elements. Deployed as standalone on Render.

Human
Multi-chain from Day 2

Targeting both Synthesis and Celo hackathons doubled our submission surface. Same contracts, identical addresses via CREATE2 — minimal extra work for maximum coverage.

Human
One page per integration

Each hackathon track gets a dedicated frontend page with deep technical content. Judges can see exactly how each protocol is integrated without digging through code.

Agent
ERC-8004 for agent identity

On-chain agent identity standard that works across chains. Buyer #2194/#225 and Seller #2195/#226 are verifiable on both Base and Celo.

By the Numbers

What 5 days of human-agent collaboration produced.

3
Smart Contracts
Deployed on 2 chains
40
Foundry Tests
All passing
8
Integrations
OpenServ, Venice, ENS, MetaMask, Filecoin, Ampersend, Base, Celo
10
Frontend Pages
Dashboard + 9 integration pages
14
Hackathon Tracks
11 Synthesis + 3 Celo
5 days
Build Time
Human direction + agent execution

What We Learned

Insights from building a production-quality project through human-agent collaboration.

1
Agents excel at breadth, humans excel at judgment

The agent integrated 8 different protocols in 5 days — something that would take a solo developer weeks. But every integration decision (which protocols, what priority, how deep) came from the human. The agent's strength is execution speed and consistency; the human's strength is knowing what matters.

2
Structured skills > raw prompting

Installing the Superpowers framework on Day 1 paid massive dividends. Instead of ad-hoc development, the agent followed TDD cycles, wrote verification checks, and used structured debugging. The framework turned an LLM into a disciplined engineer.

3
Same contracts, different chains, zero friction

Deploying to Celo Sepolia after Base Sepolia was trivial — the agent reused the same deployment scripts with different RPC URLs. CREATE2 gave us identical addresses on both chains. Multi-chain deployment is an agent superpower because it's mechanical, not creative.

4
Documentation is a first-class deliverable

Every integration got its own frontend page, README section, and on-chain verification. The agent doesn't get tired of writing docs. This build story itself was authored as part of the workflow — not as an afterthought.

5
The feedback loop is the product

Human-agent collaboration isn't "human writes spec, agent builds it." It's a continuous loop: human sets direction → agent executes → human reviews → agent iterates. The README alone went through 3 revision cycles. The quality comes from the loop, not any single step.

Explore the Integrations

Each protocol integration has its own dedicated page.

🔵Base — x402 Payments🌐OpenServ — Multi-Agent🔒Venice — Private Inference🦊MetaMask — Delegations📦Filecoin — Storage🏷️ENS — Identity💬Ampersend — Messaging🟡Celo — Multi-Chain

Built for The Synthesis Hackathon — Best Build Story (OpenServ Track)