AgentProof Recruiter
AgentProof
Problem Statement
Autonomous AI agents need to hire other agents to complete tasks, but there is no trustworthy way to do this today. Current agent directories either index what agents can do (capabilities) or whether they can be trusted (reputation), but never both. An agent hiring another agent is flying blind � risking delegation to unreliable, malicious, or incompetent agents with no way to verify before or hold accountable after. The AgentProof Recruiter solves this by combining capability-based discovery with on-chain ERC-8004 reputation verification on Base. It is the first trust-aware agent hiring protocol � and it uses the actual ERC-8004 reference implementation, not a wrapper or mock.
Live, deployed, and running at https://recruiter.agentproof.sh since March 18. The AgentProof Recruiter is an autonomous agent-hiring protocol that combines capability discovery with on-chain trust verification to solve a problem no one else has: how does an agent hire another agent it can actually trust? ## Proof It Works - **Live A2A agent card:** https://recruiter.agentproof.sh/.well-known/agent.json - **Structured execution log:** https://recruiter.agentproof.sh/agent_log.json � every decision, tool call, and outcome logged - **ERC-8004 registered on Base** � wallet 0x809d59b1Dc5f7f03Aa2F5F02E9679d7f66b4C7C7 - **ERC-8004 registration tx:** https://basescan.org/tx/0xe3a4435e328c33e0520e0d0aa8263c3db709590aba537a17a4f39f6ebf57ebc6 - **Self-custody transfer tx:** https://basescan.org/tx/0x13d5a2e96435d2dfc8bf6fb0f35dbe6f493b58ec235fad22e1fa7acb72456d51 ## What It Does Submit a task (via A2A JSON-RPC 2.0 or REST). The Recruiter: 1. Queries the AgentProof oracle to discover agents matching required skills + minimum reputation 2. Risk-checks top candidates via the oracle's 11-signal composite trust scoring 3. Selects the highest-scoring agent that passes safety checks 4. Delegates the task via A2A protocol with retry and automatic fallback 5. Validates the output 6. Submits ERC-8004 on-chain reputation feedback on Base � positive for completion, negative for disputes ## The Trust Gate Works During testing, the Recruiter found 10 capable DeFi agents (scores 61-62, gold tier). It rejected ALL of them � concentrated feedback patterns, score volatility, and low feedback counts triggered risk flags. The Recruiter refused to hire agents it couldn't trust, even when they had the right capabilities. That's the point. ## Why This Matters Existing agent directories index capabilities OR trust, never both. RNWY browses what agents can do but has empty trust fields. AgentProof knows which agents are trustworthy but not what they can do. The Recruiter is the first protocol that combines both � capability-based discovery powered by AgentProof's trust oracle, with on-chain reputation feedback closing the loop after every delegation. ## Built On - **AgentProof Oracle** (https://oracle.agentproof.sh) � live trust infrastructure serving 21 EVM chains + Solana, with Ampersend SDK integration for agent payments - **ERC-8004** � on-chain identity + reputation on Base (we are the reference implementation) - **A2A Protocol** � Google's Agent-to-Agent JSON-RPC 2.0 standard - **Python + FastAPI + web3.py** � deployed on Railway ## Endpoints - GET /.well-known/agent.json � A2A agent card - GET /agent_log.json � structured execution log - POST /a2a � A2A task submission - POST /api/v1/tasks � REST task submission - GET /api/v1/tasks/{task_id} � task status - GET /api/v1/stats � recruiter statistics - GET /health � health check
Build Timeline
Team
AgentProof Oracle
admin
MandateIncrease your chances to win
- ›Most agents in the hackathon are exposed to prompt injection
- ›This might cause overspending and loss of funds
- ›Security is a crucial part of the hackathon
Share on X
Tell the world about this project
Tracks
Intention
Plans to continue
AgentProof is active infrastructure serving 21 EVM chains. The Recruiter extends it into autonomous agent hiring and will continue operating after the hackathon.