Saguaro Code Review

Saguaro is an AI-powered code review tool that enforces project-specific rules during development. It runs as Claude Code hooks and as a background daemon.

How It Works

Saguaro integrates via two Claude Code hooks defined in .claude/settings.json:

• PreToolUse hook (Edit|Write): Before any file edit, Saguaro injects the relevant rules from .saguaro/rules/ into the agent’s context so it writes compliant code from the start.
• Stop hook: After each code change, Saguaro evaluates the diff against all matching rules. If violations are found, it blocks the agent until they’re fixed (fix-loop).
• Background daemon: Runs a senior-engineer-style review asynchronously. Findings are advisory and surfaced on the next agent turn.

Rules

33 review rules live in .saguaro/rules/*.md. Each rule has:

• id, title, severity (error or warning)
• globs — file patterns the rule applies to
• tags — for categorization
• Instructions with violation/compliant examples

Rules cover: architecture boundaries, error handling, security (XSS sanitization), testing conventions, naming patterns, logging standards, and UI consistency.

Configuration

Commands

sag review              # Run rules review on current changes
sag review --classic    # Run senior-engineer-style review
sag generate            # Auto-generate rules from codebase analysis
sag create              # Create a single rule interactively

Adding a Rule

Run sag create to generate a rule interactively, or create a markdown file in .saguaro/rules/ with this format:

---
id: my-rule-id
title: Short description of what the rule enforces
severity: error
globs:
  - frontend/src/**/*.ts
tags:
  - architecture
---
 
Explanation of the rule and why it exists.
 
Flag:
- Specific patterns to catch
 
### Violations
 
\`\`\`
code that violates the rule
\`\`\`
 
### Compliant
 
\`\`\`
code that follows the rule
\`\`\`