review

name	review
description	Review code against CLAUDE.md standards with focus on config separation, BigInt precision, and strategy patterns.

Review Skill

Review code changes against the root CLAUDE.md standards.

Usage

/review              # Review changed files (default)
/review <path>       # Review specific file or directory

Workflow

1. Gather Standards

Read the root CLAUDE.md to understand the code standards.

2. Identify Files to Review

Run:

git diff --name-only origin/main...HEAD

Filter to .ts files, exclude node_modules, dist, and generated files.

For a specific path: review only files under the provided path.

3. Review Each File

For each file, check for violations of CLAUDE.md standards. Focus areas:

Config Separation (CRITICAL):

• Client utility code (e.g. utils/maths.ts) must NOT import from config
• Client code must NOT import from process.env or dotenv
• Client code must NOT hardcode addresses, chain IDs, or parameters that belong in config
• New parameters must be added to config types and passed through, not hardcoded
• Strategy selection must only be defined in config, never hardcoded in client

BigInt Precision:

• No number type for on-chain values (amounts, shares, caps)
• No manual 10 ** n — use parseUnits/formatUnits or percentToWad
• Rounding direction is correct
• No floating-point arithmetic on token amounts

Multi-Chain Safety:

• No hardcoded chain-specific addresses in client code (they belong in config)
• Chain ID assumptions are documented
• New chain additions include all required config fields

Strategy Patterns:

• New strategies implement the Strategy interface
• Registered in the factory switch statement
• Type name added to the StrategyName union type in config
• Config constants exported from config package
• Tests added

viem Usage:

• Uses viem types (Address, Hex, Chain) consistently
• Uses viem/actions for chain interactions
• Proper error handling around on-chain calls

Error Handling:

• On-chain calls wrapped in try/catch
• Individual vault reallocation failures don't crash the bot
• Errors logged with vault address for debugging

4. Output Format

For each issue found:

[PRIORITY] TITLE
  Standard: SECTION > RULE
  FILE:LINE
  DESCRIPTION

Where:

• PRIORITY: P0 (critical — config/secret leak, BigInt misuse), P1 (high), P2 (medium), P3 (low)
• Standard: Reference to the CLAUDE.md section violated

5. No Issues

If no violations found:

No issues found (reviewed N files)