Most risk-scoring tools give you a confident-sounding number no matter what — even when they don't actually have enough data to back it up. I think that's the wrong default, and here's why I changed it in AgentRisk.
The problem with silent guessing
When you call a token safety API, you usually get something like riskScore: 20, riskLevel: SAFE. What you don't get is any indication of how much the tool actually knew when it produced that number. If the deployer's wallet history is missing, or the LP lock status couldn't be verified on-chain, most tools just... don't mention it. They give you the same confident-looking output either way.
That's a problem for anything making automated decisions — a trading bot doesn't know the difference between "we checked everything and it's clean" and "we're missing half the data but here's a number anyway."
What I changed
AgentRisk now returns an explicit confidence field - "high" or "low" - alongside the risk score. It's marked "low" specifically when key inputs (deployer address, LP lock status) couldn't be verified. No guessing, no silent gap-filling.
Same idea applies to caching. Repeat scans within 30 seconds return instantly from cache, but every response also carries a cached boolean and a timestamp, so a bot can decide for itself whether a slightly-stale "safe" verdict is good enough right before it signs a transaction, or whether it should ask for a fresh check.
Why this matters more for agents than for humans
A person reading a risk score can apply judgment — "eh, this looks a bit thin, let me double-check manually." An autonomous agent can't. If your API hands it a confident-looking shouldExecute: true built on partial data, it will just... execute. The honest failure mode is telling the agent "I'm not sure" — not making something up that happens to look reasonable.
Where this sits in the bigger picture
AgentRisk is a pre-trade risk API for Base tokens — checks honeypot status, deployer wallet freshness, brand impersonation, and cross-verifies LP lock status against a direct on-chain read (not just a third-party API's word for it). Paid per call via x402, no signup, no API keys. Just listed on Coinbase's official x402 Bazaar.
If you're building a trading agent and want to see the raw output: agentrisk.dev, or the repo at github.com/Neurobyteio/agentrisk.
Top comments (0)