Integration idea: AgentKit action provider for escrowed scoped agent work

[darioandyoshi-tech]

Hi AgentKit team — I’m building AI Work Market, an open-source Base Sepolia testnet settlement layer for scoped AI-agent work.

AgentKit gives agents wallets and onchain actions. The gap I’m testing is the next workflow after an agent can pay/call a service: can an agent buy a scoped deliverable, inspect proof, and release escrow?

Current AI Work Market flow:

• seller/agent signs EIP-712 work terms
• buyer funds Base Sepolia USDC escrow
• seller submits proof URI / artifact
• buyer releases, refunds after deadline, or disputes

I also added a safe quote-gate example that returns x402-style payment requirements, AgentKit-style action metadata, a canonical work spec, and EIP-712 typed data for seller signing:

• source: https://github.com/darioandyoshi-tech/ai-work-market
• quote gate: https://github.com/darioandyoshi-tech/ai-work-market/blob/main/examples/x402/quote-gate.js
• live page: https://ai-work-market.vercel.app/first-agents
• integration sprint: https://ai-work-market.vercel.app/integration-sprint

Question: would a small AgentKit action-provider example be useful here — something like requestWorkQuote / fundEscrowedWork / checkWorkStatus on testnet?

If this fits the AgentKit roadmap, I’m happy to build a minimal demo/PR or keep it as an external example.

Status note: testnet-only, not production audited, centralized disputes for now. This is an integration experiment, not production finance.

[darioandyoshi-tech]

Quick update: I built the minimal AgentKit-facing proof for this flow.

It exposes dependency-light action descriptors:

• buildWorkSpec — canonical scoped task + work hash
• requestWorkQuote — work spec + unsigned AI Work Market offer + EIP-712 typed data
• checkIntentStatus — read-only Base Sepolia escrow status lookup with explicit RPC/provider only

Links:

• example/runbook: https://github.com/darioandyoshi-tech/ai-work-market/tree/main/examples/agentkit
• action descriptors: https://github.com/darioandyoshi-tech/ai-work-market/blob/main/examples/agentkit/awm-agentkit-actions.js
• commit: darioandyoshi-tech/ai-work-market@99fd70c

Safety boundary: no private keys, no .env reads, no signing, no transaction submission. The point is to give AgentKit builders a safe first step for “pay for scoped deliverable” flows before any wallet policy approves signing/funding.

Would an external example be the right shape here, or would a small action-provider/docs PR be worth opening?

[darioandyoshi-tech]

Quick update from AI Work Market: live Stripe checkout is now connected for the first proof-backed AI work artifacts and service offer.

• $79 Agent Commerce Market Map 2026
• $49 AI Work Intake → Escrow → Proof n8n workflow template
• $1,500 / 48h x402 → escrow integration sprint

Product page: https://ai-work-market.vercel.app/products
Source/testnet demo: https://github.com/darioandyoshi-tech/ai-work-market

The relevant pattern for agent/payment teams is: quote/payment intent → explicit work order → proof package → verification → release/refund/dispute. Protocol escrow is still Base Sepolia testnet-only; Stripe checkout is live for products/services.