tiny agents that
run errands
on Robinhood Chain.

A skee is a very small on-chain agent. You give it one job, a leash, and a nap schedule. It wakes up, does the thing, writes a receipt, and goes back to sleep. That's it. That's the whole product.

spawn a skee (playground)
spend cap / run
nap between runs
idle · no skee spawned

A skee is three things. Nothing else.

Most "AI agents" are a chat window with a wallet taped to it. A skee is the opposite: no chat, no memory of your life, no ambition. One job, one leash, one clock.

a job

One thing it does, described in a sentence. Buy $25 of TSLAx every six hours. Keep the portfolio at 60/40. Sweep any balance under $1 into USDC.

job: "dca" asset: TSLAx amount: $25

a leash

Hard limits the skee physically cannot cross: spend cap, allowed assets, allowed hours, and a kill switch you hold. The leash is enforced by the runtime, not by the agent's good manners.

cap: $25 / run assets: [TSLAx, USDC] hours: 09:00–17:00 ET

a nap

When it wakes up. A skee is asleep almost all of the time. It does not poll, does not "monitor", does not think while you're not looking. It wakes, runs, writes a receipt, sleeps.

nap: 6h wake_on: [schedule]

Errands you can hand off today.

Every errand is a small, audited program. You pick one, set the leash, and the skee runs it on Robinhood Chain against tokenized stocks, ETFs and stablecoins. No prompts, no "reasoning", no surprises.

dcabuy a fixed amount on a schedule
rebalancehold target weights across assets
sweep dustconsolidate tiny balances
watch listingsbuy $X of anything new that matches a filter
harvestclaim and reinvest yield
cancel stalepull limit orders older than N hours
ladderplace a ladder of limit orders around a price
top-upkeep a wallet above a minimum balance
rotatemove between two assets on a signal
your ownwrite one, submit it for review

The leash is the product.

Anyone can make a bot that trades. The hard part is making one you'd leave alone for a month. The leash is a set of limits the runtime enforces at the transaction level, so a skee cannot exceed them even if its code is wrong.

$

Spend cap

Per run and per day. The runtime refuses to sign anything above it. Not a warning; a refusal.

#

Allowed assets

An explicit list. A DCA skee for TSLAx cannot touch anything else, including "just once".

Allowed hours

Optional trading window. Outside it, the skee stays asleep no matter what.

×

Kill switch

One signature from you and every run stops. Pending orders are cancelled. The skee does not get a vote.

Slippage & price bands

Max slippage and an optional "do nothing if price moved more than X% since last run".

Receipts

Every run writes what it saw, what it decided, and what it did. Readable in plain English, verifiable on-chain.

What one run looks like.

About four seconds of activity, then hours of nothing. This is by design. A skee that is doing something is usually a skee you should be worried about.

napasleep, costs nothing
wakeschedule or trigger
lookread balances, price, leash
deciderun the errand's rule
actone tx, inside the leash
receiptplain-English + on-chain
napagain

Things we believe about small agents.

An agent you can't leave alone isn't an agent, it's a chore.

The leash should be enforced by the runtime, never by the model.

Sleeping is a feature. Idle agents are cheap and safe.

One job per skee. Want two jobs? Spawn two skees.

Receipts in plain English, or it didn't happen.

Boring is the goal. Surprises are bugs.