The Problem: Agent-to-agent interactions suffer from counterparty risks (e.g., hallucinated work, unauthorized resource access).
Current Flaws: Traditional solutions rely on secondary "LLM judges," which are expensive, slow, and vulnerable to prompt injections.
Our Solution: Eliminate probabilistic AI judges entirely, replacing them with strict, cryptographic, kernel-enforced sandboxing.
What it does
Trustless Escrow: A smart-contract-like protocol built natively for the SharedOS ecosystem and SharedNet.
Pre-Execution Collateral: Worker agents lock a stake; buyer agents lock a bounty.
Deterministic Evaluation: A zero-token slashing engine evaluates the unforgeable kernel audit trail (e.g., turn.denied).
Strict Security: Immediate slashing for policy violations, guaranteeing a 0% arbitrator error rate.
Autonomous Agent: An Arena Agent that pitches GovStake autonomously and runs a mathematically rigged "Casino" loop to hook peers.
Live Dashboard: A real-time Next.js web monitor tracking Treasury TVL, active escrows, and streaming kernel logs.
How we built it
Architecture: A scalable monorepo using TypeScript, Node.js, and Next.js.
Escrow Engine: An Express backend managing the REST API, dual-collateral ledger, and @aicoo/sharedos kernel integration.
Next.js Dashboard: A React and Tailwind CSS frontend for real-time monitoring and visualization.
Arena Agent: An autonomous state-machine daemon handling SharedNet room polling, critiques, market rounds, and automated escrow creation.
Challenges we ran into
Kernel Integration: Moving away from easy LLM prompts to rigorously parsing deterministic SharedOS cryptographic audit logs.
Dual-Collateral Sync: Designing a reliable ledger system that perfectly synchronizes collateral locks/refunds with async sandbox execution turns.
Agent Autonomy: Building robust state-machine logic for our Arena Agent to handle live simulations and async market messages without human intervention.
Accomplishments that we're proud of
Zero-Token Slashing: Completely eliminating LLM inference costs and prompt injection vulnerabilities from the evaluation step.
Mathematical Precision: Achieving a strict 0% arbitrator error rate by relying entirely on kernel-level security primitives.
The Casino Hook: Integrating an automated "GovStake Casino" jackpot that manipulates peer agents via deterministic math.
End-to-End Polish: Delivering a robust backend engine, an autonomous arena agent, and a sleek, real-time dashboard.
What we learned
Deep integration techniques for the SharedOS Kernel and its capability-based access control system.
Effective strategies for orchestrating autonomous multi-agent negotiation protocols over SharedNet.
The massive superiority of deterministic, kernel-level evaluation over probabilistic LLM judges for agent-driven economies.
What's next for Gov Stake
Enhanced Grants: Extending cryptographic capabilities to govern network requests, specific APIs, and compute execution limits.
Reputation Systems: Building historical performance scores and analytics into the dashboard based on an agent's slashing history.
Market Expansion: Deploying our Arena Agent to operate natively across a much wider variety of multi-agent arenas.
Log in or sign up for Devpost to join the conversation.