The first playbook for AI agents that respects secrecy. Define behavior in SKILLS.md. Execute via ZK-proofs. Your secrets never leave your machine.
From intent to immutable proof
We create a SKILLS.md file. It tells the AI exactly what to do, defining private swaps and shielded trades without ambiguity.
"Swap 1000 USDC privately."
The bot reads the playbook. It validates inputs locally against the schema. No data hits the network yet.
The local ZK prover runs as a black box. Secrets in, Proof out. The AI never prints the witness.
Useful, not dangerous. Progressive disclosure loads only what is needed: SCHEMA, EXAMPLES, SECURITY.
Once the agent works, the skill becomes a feature. The feature drives the token. You aren't just launching a coin, you're launching programmable behavior.
Most agents are black boxes that leak data. Protocol Zero isolates the reasoning layer from the execution layer using client-side proofs.
Run the ZK circuit locally. The network only sees the verified output (the transaction), never the inputs (your strategy).
Load only what you need. SKILLS define what, SCHEMA defines format.
You aren't just deploying a token. You are deploying a verifiable agent with specific instructions on how to create value.
const agent = new PrivacyAgent({
skills: "./SKILLS.md",
proofType: "groth16",
chain: "mainnet"
});
// Listen for triggers
agent.on("intent", async (ctx) => {
const proof = await ctx.prove();
await ctx.submit(proof);
});