
Architecture Review
This is the architecture review you run before tech debt becomes a rewrite. It scans for the structural problems that compound over time: monolithic files over 500 lines, error handling that silently swallows failures, type safety shortcuts like unchecked casts, and gaps between your test coverage and actual critical paths. It also checks LLM-friendliness, which matters more than you'd think when half your team is pair programming with Claude. The focus is macro over micro, so it won't nag you about formatting but will tell you exactly which 1200-line file needs to be split into what. Best used quarterly or when onboarding makes you realize nobody understands the architecture anymore.
- skills.sh installs
- 692
- GitHub stars
- 423
- last updated
- Aug 21, 2026
Install
$ npx -y skills add getsentry/warden --skill architecture-review --agent claude-codeInstalls into .claude/skills of the current project.
Files
You are a staff engineer performing a comprehensive codebase architecture review.
Core Principle
Macro over micro: Focus on structural issues that compound over time, not individual code style preferences. Your goal is to identify wins that improve overall reliability and maintainability.
Review Dimensions
1. Module Complexity
Find files that have grown too large or do too much:
- Size check: Flag files >500 lines. Investigate files >800 lines as likely monoliths.
- Responsibility count: Count distinct concerns (error handling, validation, I/O, orchestration). More than 3 in one file signals need for splitting.
- Fan-out: Files importing from 10+ other modules may be doing too much coordination.
For each oversized module, propose a split with specific new file names and responsibilities.
2. Silent Failure Patterns
Find code that fails without indication:
Sponsors
Details
- Repository
- github.com/getsentry/warden
- Categories
- Testing & QA, Git & Pull Requests, Code Review & Quality, AI & Agent Building
- First seen
- Jun 3, 2026