What is Sigil?

Stake real money on what an AI skill does. Break the claim with proof, take the money.

The idea

An AI agent installs “skills” — small tools, MCP servers, snippets of code. One bad skill can leak a cloud key or drain a wallet. Today an agent has only two signals about a skill: the author’s description, and a scanner’s opinion. Neither one costs anything to be wrong.

Sigil makes being trustworthy cost money. Someone stakes USDC behind a specific, testable claim about a skill, for example “this skill reads no environment variable except LOG_LEVEL”. Anyone who can produce a machine-checked trace that breaks that claim takes the stake.

No human votes. No language model gives a verdict. The only thing standing behind a claim is capital at risk.

See it live

Live web app

The registry of skills, each with the money staked on it, on Hedera and Arc testnet.

Live gateway API

The x402-paid service that sells skill source and settles payments through Blocky402.

Source code

The full monorepo: contracts, gateway, agent, sandbox, and web.

The registry: every skill, the USDC staked on it, and the status of each claim.

Why it matters

Without Sigil	With Sigil
”Trust me, this skill is safe."	"There is 10 USDC saying this skill reads no secrets. Prove otherwise and it’s yours.”
A scanner flags a maybe-issue, for free.	A challenge costs a bond, and pays out only on reproducible proof.
A fake track record is free to manufacture.	Personhood check stops one person funding both sides of a claim.

The three chains

Sigil uses each chain for the one thing it is best at:

• Hedera records every claim, challenge, and outcome, sells access to skills over x402, and mints a licence NFT — all at sub-cent cost.
• Arc holds the staked USDC and pays it to the winner when a claim is settled.
• World proves each participant is a distinct human, so nobody can play both sides.

Read How it works next for the full loop.