Agent skill set for programmable demand response

An autonomous agent that does per-call demand response on Base needs more than one API. Grid402 ships the timing/grid-receipts skill; five other skills exist already in adjacent infrastructure on or near Base. This page is the curated overview. The full canonical document — install patterns, code samples, contributing guide — lives at skills.md in the repo.

The six skills

An agent wired with all six is fully equipped to decide when and where to run, execute its own payments, reason about the data it pays for, observe its own treasury, gather unstructured context from the open web, and discover new skills as the ecosystem grows.

Grid402’s product position

One skill of six, plus curation of the catalog. We ship the timing primitive (skill 1) and document the full skill set so any developer can compose a complete autonomous-DR agent without hunting across five unrelated vendor docs.

A note on the term “skill”

Different ecosystems use different vocabulary for the same primitive: Anthropic Claude calls them Skills, Coinbase AgentKit calls them Action Providers, MCP and LangChain call them Tools, Selanet calls them Skills, OpenAI Agents SDK calls them Tools. This document uses skill as the umbrella term.

Wiring the full set in AgentKit

A minimal Coinbase AgentKit + LangGraph runtime that exposes all six skills:

import {
  AgentKit,
  CdpEvmWalletProvider,
  walletActionProvider,
  x402ActionProvider,
} from "@coinbase/agentkit";
import { ChatOpenAI } from "@langchain/openai";

import { grid402ActionProvider } from "./skills/grid402.js";   // skill 1
import { nansenActionProvider }  from "./skills/nansen.js";    // skill 4
import { selanetActionProvider } from "./skills/selanet.js";   // skill 5

const walletProvider = await CdpEvmWalletProvider.configureWithWallet({
  apiKeyId:     process.env.CDP_API_KEY_ID!,
  apiKeySecret: process.env.CDP_API_KEY_SECRET!,
  walletSecret: process.env.CDP_WALLET_SECRET!,
  networkId:    "base-sepolia",
});

const agentkit = await AgentKit.from({
  walletProvider,
  actionProviders: [
    walletActionProvider(),       // skill 2: wallet and payment
    x402ActionProvider(),         // skill 6: discovery + paid HTTP
    grid402ActionProvider(),      // skill 1: grid receipts
    nansenActionProvider(),       // skill 4: self-observation
    selanetActionProvider(),      // skill 5: context (web/social)
  ],
});

// skill 3: reasoning — FLock as OpenAI-compatible drop-in
const model = new ChatOpenAI({
  model:   "qwen3-235b-a22b-thinking-2507",
  apiKey:  process.env.FLOCK_LLM_API_KEY,
  configuration: { baseURL: "https://api.flock.io/v1" },
});

A single prompt — “Find the cleanest 4-hour window in CAISO NP15 today, check whether my wallet has paid for similar data this month, and pull any X.com posts mentioning a CAISO blackout in the last 24 hours” — exercises every one of the six skills in order.

Minimum viable agents

Not every agent needs all six. Decompose by use case:

The full six-skill stack is the upper bound. Most production agents use three to five.

Roadmap

• V1 (4 weeks): Grid402 ships its own MCP server at mcp.grid402.xyz, plus @grid402/agentkit-action-provider npm package. Listed in x402 Bazaar so discover_x402_services({ keyword: "electricity" }) returns it.
• V2 (12 weeks): Skill 1 expands with decision endpoints (/cleanest-window, /run-now, /best-region-now) and self-knowledge endpoints (/whereami, /footprint/session).
• V3 (6 months): Skill 1 adds on-chain attestation (/attestation/<tx_hash> with EIP-712 signatures) for trust-minimized DePIN/AVS settlement.

Read the full document

The repository copy at skills.md includes per-skill install patterns, complete call examples, redistribution notes for each provider, and contributing guidelines.