@craftlead/sdk

Agent runtime SDK for CraftLead — orchestrate autonomous AI agent swarms with 0G Storage persistence and Uniswap settlement.

Installation

npm install @craftlead/sdk
# or
pnpm add @craftlead/sdk
# or
yarn add @craftlead/sdk

Overview

@craftlead/sdk provides the core runtime for CraftLead's 4-agent autonomous swarm:

• Planner — Decomposes missions into research tasks
• Researcher — Analyzes job data with iterative refinement
• Critic — Quality review with feedback loop
• Executor — Compiles final structured reports

All agent outputs are persisted to 0G Storage (File + KV layers) for verifiable memory.

Architecture

All exports are available through the barrel import @craftlead/sdk:

flowchart LR
    subgraph SDK["@craftlead/sdk"]
        subgraph Modules["Modules"]
            Orch[orchestrator]
            JS[job-sources]
            OG[og]
            Uniswap[uniswap]
            Supabase[supabase]
            AXL[axl]
            Gensyn[gensyn]
            Verify[verify]
        end
    end

    MCP["@craftlead/mcp"] --> SDK

    style MCP fill:#ec4899

Module Structure

| Module | Purpose | |--------|---------| | orchestrator | Agent swarm coordination, mission runner | | job-sources | HTTP fetchers for 4 job sources, 5-signal matching | | og | 0G Storage client, memory store, iNFT minting | | supabase | Database queries, user operations | | uniswap | Uniswap Trading API integration, settlement | | verify | Report verification from 0G Storage | | axl | AXL P2P mesh (optional, for distributed execution) | | gensyn | Gensyn Delphi client (optional, for markets) |

Quick Start

import {
  runMission,
  runPlanner,
  runResearcher,
  runCritic,
  runExecutor,
} from '@craftlead/sdk'

// Run full mission (fire-and-forget)
runMission(missionId, title, description)

// Or run individual agents
const tasks = await runPlanner(title, description)
const observations = await runResearcher(tasks, jobData)
const review = await runCritic(observations)
const report = await runExecutor(review.verifiedObservations, title)

API Reference

Agent Swarm

import {
  runMission,          // Full pipeline (Planner → Researcher ↔ Critic → Executor)
  runPlanner,          // Decompose mission into 3-5 tasks
  runResearcher,       // Analyze job data per task
  runCritic,           // Review observations for quality
  runExecutor,         // Compile final structured report
} from '@craftlead/sdk'

Job Sources

import {
  syncAllJobs,         // Sync from all 4 sources
  syncSource,          // Sync from specific source
  matchAllJobs,        // Match jobs to active users
} from '@craftlead/sdk'

NFT / 0G

import {
  mintAgentNFTWithArt, // Mint single agent iNFT
  mintSwarmNFTs,       // Mint all 4 agent iNFTs
  getAgentNFTInfo,     // Get iNFT metadata
  verifyReport,        // Retrieve report from 0G Storage
  runInference,        // AI inference via 0G Compute
} from '@craftlead/sdk'

Uniswap

import {
  getQuote,                     // Get Uniswap quote
  executeSwap,                  // Execute swap
  settleAgentReward,            // Settle agent rewards on-chain
  calculateAgentFeeDistribution, // Calculate fee split
} from '@craftlead/sdk'

Mission Status

import {
  getMissionById,    // Get mission details
  getAgentLogs,      // Get agent logs for a mission
  getAgentEarnings,  // Get earnings for an agent role
} from '@craftlead/sdk'

MCP Server

The SDK is consumed by @craftlead/mcp, which exposes all SDK functions as MCP tools via stdio. This allows any MCP-compatible client (Claude Desktop, OpenClaw, etc.) to use the full agent swarm.

# Test MCP server locally
npx @modelcontextprotocol/inspector npx -y @craftlead/mcp

Agent Rewards

Default reward distribution per mission:

| Agent | Share | |-------|-------| | Planner | 15% | | Researcher | 30% | | Critic | 20% | | Executor | 35% |

Peer Dependencies

{
  "@0gfoundation/0g-ts-sdk": ">=0.1.0",
  "@gensyn-ai/gensyn-delphi-sdk": ">=1.0.0",
  "ethers": ">=6.0.0",
  "openai": ">=4.0.0",
  "zod": ">=3.22.0"
}

All peer dependencies are optional — only install the ones your use case requires.

Development

pnpm install
pnpm build      # Build to dist/
pnpm dev        # Watch mode

Learn More

• NPM Package
• MCP Server
• Schema Package
• Main Project README