wtf is solana-safe-trade-skill?

This project is a safety check that runs before an automated trading agent buys a token on the Solana blockchain. It is built as a skill for the Solana AI Kit, a framework that lets AI agents hold a wallet and trade tokens on their own. The problem it solves is that most Solana trading skills will happily try to buy any token an agent points at, including scam tokens set up to look real but that trap your money once you buy in, sometimes called honeypots. The tool runs four checks, called gates, in order from cheapest to most expensive, and stops as soon as one fails. First it checks whether the token is on Jupiter's list of known, verified tokens, which catches fake tokens pretending to be something like USDC. Second it compares the token's price against a trusted price oracle called Pyth, with a service called Birdeye as a backup, to catch tokens with wildly wrong prices. Third it checks whether the token actually has real trading activity and enough pooled money behind it, using data from DexScreener, to catch honeypots that look tradeable but are not. Fourth it asks Jupiter's own trading router whether a real trade route exists for the token, to catch tokens that pass every other check but still cannot actually be swapped. A token has to pass all four gates before the agent is allowed to trade it. The author reports this same four gate approach already running in production on a different blockchain called BSC, where it caught a real token that looked safe by every individual measure but had almost no trading volume. The repository includes a working TypeScript version of the checks, a test suite, and an install script that adds this project as a submodule into an existing Solana AI Kit setup. It also documents how to tune the safety thresholds differently depending on whether the agent is a cautious institutional bot or a more aggressive memecoin trader. The project is released under the MIT license.