llm-trading-agent-security

Layered security patterns for autonomous trading agents that sign and send transactions — prompt-injection filters on external data, independent per-tx/daily hard spend limits, mandatory pre-send simulation with a required `min_amount_out`, drawdown-based circuit breakers, a dedicated session hot wallet (never the treasury), and private mempool routing (e.g. Flashbots) with per-strategy slippage and deadline. Pre-deploy checklist treats no single check as sufficient.