CodeWarden — Security skill for Claude Code

🛡️ CodeWarden

**An agentic AI reviewer that watches every pull request for security risk and standards drift.**

CodeWarden is an autonomous code review agent that integrates directly with GitHub to review pull requests in real time. It performs multi-layered static security scans (Semgrep + OWASP Top 10), pairs findings with an LLM reasoning engine (Claude), and enforces team-specific coding standards using a Retrieval-Augmented Generation (RAG) pipeline powered by ChromaDB.

---

🌟 Key Capabilities

1. **Security Scanner Agent**

   • Combines deterministic AST/SAST scanning (Semgrep OWASP Top 10 rulesets) with LLM deep reasoning.
   • Categorizes issues with precise severity (critical, high, medium, low) and explains why the vulnerability is risky and how to remediate it.
   • In-code suppression support via // codewarden-ignore comments.

2. **RAG Standards Agent**

   • Ingests and vectorizes your team's internal engineering guidelines, conventions, and style guides.
   • Dynamically retrieves standard chunks matching changed files in PR diffs.
   • Enforces team conventions without generic or noisy lint warnings.

3. **Intelligent Aggregation Layer**

   • Merges findings from both agents, eliminates duplicates, and ranks them by severity.
   • Posts line-specific GitHub comments and a formatted review summary with clear emojis and action recommendations (approve, comment, or request-changes).

4. **Interactive Developer Dashboard**

   • Dark-mode-first developer UI built with React 19, Tailwind CSS v4, and Recharts.
   • Overview metrics: Total PRs, Critical issues caught, Average response time, and Team health score.
   • Detailed repository settings, standards document management, and review inspection with syntax-highlighted diffs and inline finding markers.

---

🏗️ Architecture

                       ┌────────────────────┐
                       │   GitHub Repo       │
                       │  (PR opened/synced) │