Coinbase's Auditing Standards with Shashank Agrawal

Coinbase's security process protecting over $7 billion in TVL rejects the single-audit model common in DeFi. Shashank Agrawal, Senior Engineering Manager, Protocol Security at Coinbase, explains their multi-round validation approach: internal security teams (separated from product engineering) audit first, then external firms audit, and rounds continue until external auditors surface only lows and informationals—never highs or criticals.

This stopping rule creates a quality bar where internal audits must catch everything significant before external validation. For the Base bridge specifically, this meant independent OP Stack security validation despite Optimism's existing audit work, driven by the "absolutely zero room for error" standard when contracts hold substantial user funds. Their approach treats external auditors as verification layers rather than primary discovery mechanisms.

 

Topics discussed:

• Multi-round audit methodology continuing until external firms find zero high-severity or critical issues
• Internal security team structure operating independently from product engineering before external validation
• Base bridge security requiring custom OP Stack validation independent of Optimism's audit coverage
• In-house MPC library development using professor-reviewed specs bridging research papers to production implementation
• Tabletop war gaming exercises simulating worst-case chain scenarios with security, engineering, legal, and compliance teams
• Free Hexagate monitoring partnership providing base-layer protocol coverage for Base ecosystem builders
• Security hiring process using live code audits at different complexity levels for senior (level 5) versus staff (level 6) positions
• Off-chain infrastructure security: key management and transaction signing treated as equal priority to smart contract auditing
• AI smart contract auditing tools showing current production limitations in determinism and false positive rates
• Incident response planning where monitoring systems and alert workflows prioritize minute-by-minute decision speed