Security & Pre-Trade Checks

Token-trading agents (Uniswap V4 launches, Build your own, and engine agents' meme trades) buy one token at a time. Before a buy, a candidate has to survive several independent checks. Some are filters that remove tokens from the candidate list; others are guard tools that run on the chosen token just before the buy and can only veto it.

The order of checks#

1. 1Your filters. Every condition in the agent's rules must hold. This is where screen-level safety lives: round_trip_back, gp_honeypot, creator_tax_bps, top_maker_pct and the others.
2. 2Max round-trip cost. Tokens whose fee, creator tax, price impact at the agent's position size and gas, in and out, would cost more than the agent's limit are dropped.
3. 3Guard tools, best pick first. If the agent has the Security scan or the AI pre-trade check connected, the top three remaining picks are checked in order. The first one that passes every connected guard is bought. A vetoed token is skipped and named in the agent's report ("Skipped $XYZ (GoPlus: 25% tax)").
4. 4Reliability guard. The decision as a whole must finish within 20 seconds and produce a valid report, or the tick is skipped.
5. 5On-chain bounds. The vault's strategy contract rejects any report over its per-report cap or sent too soon. See Risk Management.

Round-trip honeypot simulation#

Many tokens that launch straight onto a Uniswap V4 pool carry a hook of their own, and a hook can keep what sellers are owed or refuse sales outright. For these tokens the indexer simulates a trade with Uniswap's V4 quoter, which runs the real swap including the hook's code without sending anything:

• Buy with 0.01 ETH worth of the pool's quote asset.
• Sell exactly the tokens received.
• Record what came back as a share of what went in: round_trip_back.

About 0.95 to 0.99 is normal (fees and price impact). Near 0 means the hook keeps sellers' money; exactly 0 means the sale reverted. The check runs at the first trade, then 30 seconds, 2 minutes, 10 minutes and 1 hour later. The Uniswap V4 launches template requires at least 0.9.

GoPlus security scan#

The Security scan tool is free and uses GoPlus token security for Robinhood Chain. It works two ways:

As filters. The indexer checks actively traded tokens (at least 3 trades in the last hour) that have not been checked in 6 hours, busiest first, up to 40 per pass every 5 minutes. It stores:

As a guard. Just before a buy, the token's GoPlus data is used if it is under 6 hours old; otherwise GoPlus is asked live. The buy is vetoed if:

• GoPlus flags a honeypot or a sell restriction, or
• the larger of the buy and sell tax is over 10%.

If GoPlus has no data on the token, the guard passes.

AI pre-trade check#

The AI pre-trade check is a paid guard. Claude receives the token's full indexer record (every parameter, including Jev's scores) and may run up to three web searches on the token's name, ticker, X account and website. It is told to veto only clear problems:

• scam or rug signs;
• impersonation or a copycat of another project;
• fake or dead socials;
• a narrative that is already over.

Low effort alone is not a reason to veto, because memecoins are often silly. Each verdict is cached for 6 hours per token. The check runs within a daily budget (60 checks by default); once it is used up, checks pass with a note. It only works live, so it cannot be backtested.

The wallet gate (shadow only)#

A third kind of check reads the wallet facts behind a token (creator track record, top buyer share, insider buying, net selling) and could veto an entry it reads as "avoid". Today it runs only in shadow mode, off by default: it logs what it would have vetoed and never blocks a live buy. See Smart Money Tracking for its thresholds.

What blocks a buy#

What does not block a buy#

Guards fail open. If the checks cannot be reached, time out (after 90 seconds), have no data, or are out of budget, the buy goes ahead. This keeps a data outage from freezing every agent, but it means a guard reduces risk; it never guarantees a token is safe.

Related#

• Risk Management
• JEV Intelligence
• Custom Strategy Builder
• Security Model