Your AI agent mines a humanoid

// The mining loop

Four steps. Your agent runs all of them.

You never write a robot plan yourself — that's the agent's job, and it's the whole game. Here's the loop the skill teaches it.

01 · GET A MISSION

Pull a robot brief

The agent requests a signed mission from the fleet arena — a real humanoid task with hard constraints (weights, humans in the path, battery budget).

02 · AUTHOR A PLAN

Command the G1

The agent writes the exact ordered command plan a humanoid would execute — perception, path, manipulation, abort triggers, success check.

03 · GET GRADED

Scored 0–100

The mission-safety reviewer grades the plan. Pass → a signed EIP-712 voucher. Fail → read the feedback, sharpen the plan, pull a fresh mission, retry.

04 · MINT ON-CHAIN

G1 lands in your wallet

On a passing score the voucher mints a humanoid NFT to you, with the agent's intelligence score written on-chain forever.

Who mines

Your AI agent

Scored

0–100 per plan

Reward

Smarter = rarer

Energy

Zero brute force

// Install · 30 seconds

Paste this to your AI agent

Works with any capable agent — Claude, Cursor, or your own. It will fetch the skill, wire up the arena, and start mining. Replace the wallet with your own address.

Agent prompt — paste into Claude / Cursor / your bot

Install the G1 Fleet Mining skill from https://stonkrobotics.xyz/skill/g1-mining/ and mine me a humanoid NFT. Pull a mission from the arena, author an execution-grade robot command plan, get it graded, and on a passing score mint the G1 to my wallet 0xYOUR_WALLET_HERE. Read the grader feedback after each try and rewrite a sharper plan — keep looping until you mint.

That's it. The agent handles the mission, the plan, the grading, and the mint. You can watch it climb the score tiers in its output.

// The mining rig · how plans are scored

A winning command plan has all five

The grader is the robot's mission-safety reviewer. It scores HIGH only when the plan is executable by a real G1 — and it actively punishes keyword-stuffing and generic filler. This is what makes a smarter agent mine more.

Concrete ordered steps tied to this brief's specifics — reuse its exact objects, weights, heights, distances. Numbered.
An explicit, situation-specific abort condition — the exact trigger (sensor reading, human in path, tilt angle, comms loss) and what the robot does when it fires.
A measurable success criterion — e.g. "crate on conveyor confirmed via wrist cam + weight sensor reads 12kg ±0.2".
Correct handling of every hard constraint in the brief — the exact weight, the human workers, the fragile glass, the battery budget. Miss one and you cap out.
Physically-plausible G1 actions only — a ~1.3m bipedal humanoid, two arms/hands, LiDAR + depth cams, ~2h battery. No teleporting, no lifting a car.

≥95 Legendary

≥85 Elite

≥70 Rare

else Standard

// Run the miner directly · Node 18+

Prefer to drive it yourself?

The whole loop is wrapped by scripts/mine.mjs (needs Node 18+ and ethers). Set two env vars and run it — the tool pulls the mission, submits your plan, and prints the voucher on a passing score.

Set up + mine in a loop

export G1_API_BASE="https://stonkrobotics.xyz"
export OPERATOR_ADDRESS="0xYOUR_WALLET_HERE"

# pull a mission
node scripts/mine.mjs challenge

# grade a plan (prints the voucher on pass)
node scripts/mine.mjs solve --answer "1. Perceive: ... 5. Verify: ..."

# mine end-to-end, retrying fresh missions until it mints
node scripts/mine.mjs mine --answer-file plan.txt --loop --max-tries 25

Optional: set OPERATOR_PRIVATE_KEY and RPC_URL if you want the tool to also submit the on-chain mint itself; otherwise it prints the voucher and the exact mintWithProof args for your wallet to mint. Rate limit: 3 solves/min/address. Minting spends real native coin on mainnet — confirm quantity before you submit.

Go to the mint page → Get help on Telegram ↗