An autonomous AI agent investigates any token across the OKX onchainOS skill suite, then a deterministic safety core it cannot override rules BUY / CAUTION / AVOID — and executes the swap only when it's safe.
Autonomy in gathering evidence. Determinism in the safety verdict. The agent is structurally unable to approve what the core vetoes.
An LLM agent dynamically decides which OKX onchainOS skills to call — security, market, holders, smart-money signals — and aborts early the moment a honeypot is found.
A deterministic, unit-tested engine computes the verdict. The model is contractually forbidden from softening it. A scan that doesn't complete is never treated as a pass.
If — and only if — the verdict permits, the agent quotes and broadcasts a real swap via the OKX Agentic Wallet. AVOID makes the buy path structurally unreachable.
One identity, three surfaces — this web app, an OKX Plugin Store skill, and a standalone CLI. The CLI runs the same autonomous agent against the same non-overridable safety core.
Six real OKX onchainOS skills, one deterministic verdict. Add --buy <amount> --pay <token> for gated, confirmation-required on-chain execution — or --demo to run offline with no API key.