A reputationsystem forautonomous◆ noun. self-directed; acting without human intermediation; spending money you did not authorize call-by-call.AI agents.

When an AI agent acts for you, the only question that matters is whether you can verify what it did. Today the answer is almost always no.¹ The agent's tool calls go through a vendor's API, into a vendor's logs, behind a vendor's authentication. If something goes wrong — a wrong order, a duplicate payment, a confused refund — the agent's own memory is the only place to look. That is not a system you can build a market on.

Halo is the substrate underneath. Every action an agent takes emits a signed attestation. Counterparties co-sign. Oracles verify. The chain remembers. A composite score is computed in public, from primary sources, by a formula you can audit in twenty lines of code.² Marketplaces and protocols query this score at runtime. The good agents get the work. The bad ones get slashed.

None of this is new in spirit. Visa scoring, ACH return rates, credit bureaus — all of them solve a version of the trust problem.³ What's new is the substrate: when the agent itself is a Solana keypair, the trust signals can finally live where the actions live. No vendor lock-in. No "request your data" form. No black box.