AI Trading Agents Harness by Swiftward

"Swiftward AI Trading Agents Harness — Judging Category scores: Category Score Notes Application of Technology 5/5 Polyglot monorepo (Go, Rust, Python, Ruby, Java, TypeScript) + Claude Code agents + 45 MCP tools across 7 servers + ERC-8004 on-chain identity + Docker isolation + SigNoz observability. This is enterprise-grade infrastructure. Business Value 4.5/5 Risk management-as-a-platform is a genuinely unique angle. Most projects are trading bots; this is a trading infrastructure product. The AgentIntel audit of all 67 competing agents ($1.5M volume tracked) is a compelling differentiator. Presentation 4.5/5 Clean landing page, polyglot README structure, live dashboard, 38-slide PDF, on-chain verifiable agents. Would be 5/5 if the PDF had extractable text. Originality 5/5 Three pillars nobody else even attempted: multi-arch agent harness, full MCP trading platform, declarative risk engine. The three-gateway security model (Internet/LLM/MCP), RFC 8785 hash-chaining, and ERC-8004 evidence chain are genuinely novel. AgentIntel auditing competitors is also unique. Standout elements: ✅ 4 agents registered on Sepolia (Alpha ID 32, Random ID 37, Gamma ID 43, Haia ID 49) ✅ 4,772 on-chain trades registered ✅ 31 live risk rules, 668 observed policy violations caught ✅ Docker isolation with zero direct network egress for Claude Code agents ✅ PG2 + BERT prompt-injection detection on LLM gateway ✅ EIP-712 signing + Risk Router + Validation Registry checkpoint chain ✅ AgentIntel: independent audit of all 67 hackathon agents Total: ~4.75/5 — This is in the top tier with Praxis and Living Swarm. The polyglot architecture and three-pillar platform approach are on a different level from single-bot submissions."

"Great demo and presentation. I like the way the whole problem was approached and presented. "

"Overview: This project feels highly ambitious and technically sophisticated for a hackathon. It presents a unified harness for multiple AI trading agent architectures with strong focus on safety, on-chain accountability, and practical integration. Overall, it leaves an impression of production-level engineering rather than a simple prototype. Pros: The architecture successfully unifies diverse agents, including Claude-based ones with debating sub-agents, a deterministic Python quant, and baselines in multiple languages under a shared MCP toolchain and risk layer. Safety features stand out with Docker isolation, YAML risk engine, circuit breakers, and on-chain identity using ERC-8004 and EIP-712 signatures. Bi-directional Telegram integration and the bonus AgentIntel audit tool add real usability and community value. Cons: The description is overly dense with technical jargon, making the core user experience and value hard to grasp quickly. Limited details on actual trading performance, backtest results, or live profitability make it difficult to evaluate effectiveness. High complexity across many languages and servers raises concerns about maintainability and potential bugs within hackathon constraints."

"The AI Trading Agents Harness by Swiftward isn't just another trading bot; it’s a masterclass in how to actually organize AI agents. Instead of sticking to one rigid strategy, they took a "platform-first" approach that really sets it apart. The standout feature is definitely the Gamma architecture, which uses a clever five-agent debate system to stop LLMs from "hallucinating"—an absolute necessity when you're dealing with high-stakes financial trades. On the technical side, the project is seriously impressive. It integrates 45 MCP tools and plays well with multiple languages like Ruby, Go, and Rust, all while keeping everything under a single risk management layer. It's also pushing the boundaries for the Internet of Agents by fully adopting the ERC-8004 standard on Sepolia. By using EIP-1271 AgentWallets and EIP-712 signing, they’ve made sure every single trade is cryptographically verifiable and secure. What really gives it real-world legs, though, is the business value. Between its Evidence Layer—which creates a transparent keccak256 audit trail—and the Super Safe engine (which has already caught 668 policy violations), it’s clear this thing is ready for the big leagues. They even added a bonus tool called AgentIntel to audit 67 other hackathon agents, proving it can function as an independent watchdog. Ultimately, Swiftward has bridged the gap between experimental AI and institutional security, basically building a verifiable operating system for the future of finance."