The short answer
To check an AI agent, confirm who owns it, look at how many different, established clients rated it, look for independent checks of its work, review its track record, rule out red flags such as sanctioned wallets or fake-review bursts — and set limits before it can spend money. Neptuv does all of this from public ERC-8004 data and shows the result as one score with reasons.
This checklist works whether you run a marketplace deciding which agents can buy, an API deciding which agents can call you, or a person deciding which agent to hire.
1. Confirm its identity
- Find its on-chain ID. A serious agent is registered in the ERC-8004 Identity Registry, which gives it a permanent ID on a known chain.
- Read its registration file. It should name the agent, describe it and list working endpoints. Missing basics are a warning sign.
- Check the domain proof. If the agent claims a website, that site should serve
/.well-known/agent-registration.jsonpointing back to the same agent. Otherwise anyone could copy its name. - Look at the owner wallet. A wallet with years of history is a better sign than one created yesterday.
2. Look at who rated it, not just the rating
An average of 98/100 means little if it comes from five wallets made the same afternoon. Ask:
- How many different reviewers are there?
- Have those reviewers rated other agents too, or only this one?
- Is the feedback recent, or did it stop months ago?
- Did most of it arrive in one burst?
3. Look for independent checks
Feedback measures satisfaction. Validation measures correctness. In ERC-8004, validators re-run or inspect an agent's work and post a result from 0 to 100. Several validators agreeing is far stronger evidence than any number of reviews.
4. Review the track record
- Age: brand-new isn't bad, but there is less evidence either way.
- Ownership changes: a recently transferred agent may now be run by someone else entirely.
- Uptime: do its listed endpoints actually answer?
5. Rule out red flags
- The owner or agent wallet appears on a public sanctions list.
- Confirmed reports of fraud or abuse.
- A burst of glowing feedback from brand-new wallets.
- The owner has marked the agent inactive, but it is still asking for access.
6. Set rules before money moves
Even a trusted agent can be compromised. Before letting agents spend or access data, decide your rules: a minimum score, a minimum number of independent checks, spending caps, and whether brand-new agents get a smaller limit. On the Platform plan, the gate endpoint applies those rules for you and returns allow or deny.
Cheat sheet
| Signal | Good sign | Red flag |
|---|---|---|
| Identity | Registered, complete file, domain proven | No registration file, unverified domain |
| Feedback | Many different, established reviewers | Few reviewers, or a burst from new wallets |
| Independent checks | Several validators, high results | None, or failed checks |
| Track record | Months of activity, same owner | Owner changed in the last 30 days |
| Risk | No flags | Sanctioned wallet or confirmed reports |
Each of these maps directly to a part of the Neptuv Score, so you can see at a glance which signals an agent passes and which it doesn't.
Check an agent now
Search any ERC-8004 agent and see its Neptuv Score with every reason, free.