iktok90-design/ai-smart-contract-auditor/known-good-comparison

Known-good comparison (meta-skill)

Most production Solidity is forks of well-known references. The fastest path to high-precision audit is to know what "normal" looks like and treat deviations as suspect.

Reference catalog

The skill maintains awareness of canonical shapes:

Tokens

• OpenZeppelin ERC20, ERC721, ERC1155 (versions 4.x and 5.x)
• Solady ERC20, ERC721, ERC1155, ERC4626
• ERC-2612 permit canonical
• Permit2 Uniswap canonical

Access / governance

• OZ Ownable, Ownable2Step
• OZ AccessControl, AccessControlEnumerable, AccessControlDefaultAdminRules
• OZ Governor, GovernorVotes, GovernorTimelockControl
• Compound Governor Bravo
• OZ TimelockController

Proxies / upgrades

• OZ ERC1967, UUPS, TransparentUpgradeable
• OZ BeaconProxy
• Diamond standard EIP-2535 reference
• Solady LibClone

Math / utilities

• OZ Math.mulDiv
• Solady FixedPointMathLib
• PRBMath UD60x18, SD59x18

AMMs / DEX

• Uniswap V2 Pair, Router, Factory
• Uniswap V3 Pool, NonfungiblePositionManager
• Uniswap V4 PoolManager + hook templates
• Curve StableSwap, CryptoSwap
• Balancer V2, V3 vaults

Lending

• Aave V3 Pool, AToken, VariableDebtToken
• Compound V3 Comet
• Morpho Blue

Vaults

• OZ ERC4626 (v4.9+ for inflation defense)
• Yearn V3 strategies
• MetaMorpho

How to use

Step 1 — Detect reference

If the contract:

• Imports from @openzeppelin/contracts or solady/
• Inherits from a recognizable parent
• Implements a recognizable interface (IERC20, IPoolManager, etc.)
• Has function signatures matching a reference

→ Identify the reference and version.

Step 2 — Diff the deviations

For each function inherited or overridden from the reference:

• Has its body been modified?
• Has its access control been changed?
• Has its return value been altered?
• Has its modifier set been changed (e.g. nonReentrant removed)?

For each storage variable added beyond the reference:

• Is the storage layout still compatible with upgrade safety?

Step 3 — Treat deviations as suspect

Each deviation gets a justification check:

• Was this change intentional and beneficial?
• Or was it a typo / mistake / regression?

Most exploits in forks live in the deviation, not the reference.

Output integration

When finding-writing, include a reference comparison if applicable:

[REENT-001 | High] Reentrancy in Vault.withdraw
 
  Reference: OpenZeppelin v5.0.2 ERC4626.withdraw
  Deviation: caller-side nonReentrant guard removed (commit 0xabc)
  Impact:    classic reentrancy pattern reintroduced
  
  → Confidence: HIGH (deviation from known-good adds the bug)

When NOT to use

• Pure novel implementations with no canonical parent.
• Contracts so heavily modified that the reference comparison is misleading.

Related

• /diff-audit command — exposes this skill as a user-facing operation
• [[confidence-scoring]] — known-good match → confidence MEDIUM-to-HIGH
• [[multi-pass-self-critique]] — known-good is a third independent pass