Nine Wickets, Nine Checks: The Pre-Deployment Gauntlet Every Agent Owner Should Run
Photo: Acabashi, CC BY-SA 4.0, via Wikimedia Commons
In cricket, a wicket can fall in an instant—one unplayable delivery, one lapse in concentration, and the innings is over. Deploying an autonomous trading agent onto live markets works the same way. The difference is that in crypto, you get to inspect the pitch beforehand.
What follows are nine wickets—nine checks every agent owner should require their system to clear before a dollar of real capital is at risk. Treat each as a stage-gate: if the agent fails one, it does not deploy.
Wicket One: The Sandbox Gauntlet
Run the agent in a simulated environment against historical data that includes the ugliest periods you can find—the March 2020 crash, the LUNA unwind, the FTX collapse. What to test: behavior under volatility spikes and cascading liquidations. Red flag: the agent performs well only in trending markets.
Wicket Two: Mainnet Dress Rehearsal
Simulations lie about gas costs, block times, and mempool dynamics. Execute the full strategy on mainnet with trivially small amounts. What to test: actual transaction costs, failed transactions, and slippage. Red flag: execution economics that only work at scale.
Wicket Three: The Permission Audit
Enumerate every approval, allowance, and administrative key the agent holds. What to test: whether permissions exceed what the strategy requires. Red flag: unlimited token approvals or an owner key that can drain funds without time locks.
Wicket Four: Circuit Breaker Verification
Manually trigger every kill switch. What to test: whether halts actually halt—positions closed, withdrawals blocked, exposure frozen. Red flag: a circuit breaker that pauses new trades but leaves open positions swinging.
Wicket Five: Oracle Independence
Feed the agent deliberately corrupted price data. What to test: whether it detects deviation from consensus sources or blindly executes. Red flag: single-source price feeds with no staleness checks.
Wicket Six: Failure Mode Documentation
For every external dependency—RPC providers, liquidity venues, bridge protocols—write down what happens when it fails. What to test: graceful degradation, not graceful hope. Red flag: any dependency whose failure has no documented answer.
Wicket Seven: The Adversary Drill
Hire or simulate a hostile party: flash loan attacks, sandwich bots, front-running. What to test: whether the agent's execution path leaks value to MEV extraction. Red flag: transactions submitted without private routing or slippage bounds.
Wicket Eight: Recovery and Reconciliation
Force a restart mid-strategy. What to test: whether the agent reconstructs accurate state from on-chain data alone. Red flag: any reliance on in-memory state that cannot be rebuilt.
Wicket Nine: The Human Handoff
Give the documentation to someone who did not build the system and ask them to take control. What to test: whether emergency procedures are actually executable by a stressed human at 3 a.m. Red flag: tribal knowledge that lives only in the founder's head.
Playing the Full Innings
None of these checks is exotic. What separates resilient deployments from the graveyard of exploited agents is not sophistication—it is the discipline to refuse deployment until every wicket stands. In cricket, nine wickets down still means you are batting. In agent deployment, nine wickets cleared means you are finally ready to play.
