SuperAPI AI Directory

skill-08-security-audit

Security & auditing

View repository

Updated 10-16 06:43

About this skill

Security Audit

When to Use

  • Evaluating unknown smart contracts before interaction
  • Detecting rug pull and honeypot patterns
  • Monitoring for active exploits and protocol incidents
  • Reviewing contract code for common vulnerabilities
  • Pre-deployment security checklist

Prerequisites

  • MCP: blockchain-rpc-server, defi-protocol-server
  • Knowledge base: knowledge-base/security/

Workflow

1. Automated Rug Pull Checks

Run before any token swap or liquidity add:

Check Red Flag
Ownership owner() can mint unlimited tokens
Liquidity LP not locked or unlock imminent
Honeypot Sell simulation reverts
Hidden fees Transfer tax > 10% or modifiable
Proxy Unverified implementation, no timelock
Blacklist isBlacklisted or pause on transfers

Use eth_call simulation: attempt buy → approve → sell in single trace.

2. Vulnerability Patterns

Scan for OWASP Smart Contract Top 10:

  1. Reentrancy — external calls before state updates
  2. Access control — missing onlyOwner on critical functions
  3. Oracle manipulation — spot price without TWAP
  4. Flash loan attacks — single-block price dependency
  5. Integer issues — unchecked math (pre-0.8.0)
  6. Delegatecall — untrusted implementation in proxy
  7. Front-running — no slippage/deadline protection

See knowledge-base/security/common-vulnerabilities/ for patterns.

3. Exploit Monitoring

  • Monitor Rekt News, DeFiLlama hacks dashboard
  • Check contract against known exploit DB
  • Verify protocol TVL hasn't dropped > 20% in 24h

4. Code Review Checklist

  • Uses Solidity ≥ 0.8.0 with overflow checks
  • External calls use checks-effects-interactions
  • No selfdestruct or delegatecall to user input
  • Timelock on admin functions (> 24h for DeFi)
  • Events emitted for state changes
  • Formal verification or audit report available

Verdict Format

RISK: LOW | MEDIUM | HIGH | CRITICAL
FINDINGS: [list with severity]
RECOMMENDATION: PROCEED | CAUTION | BLOCK

Block interaction when RISK = CRITICAL or honeypot detected.

References

  • knowledge-base/security/audit-patterns/
  • knowledge-base/security/exploit-db/