warden-core 1.8.4

Warden - AI Code Guardian for comprehensive code validation

Warden Core - The AI Code Guardian

"AI writes the code. Warden guards the production."

Warden is an AI-native security and quality gate designed to validate code generated by LLMs (Claude, GPT-4, etc.) before it enters your codebase. It acts as a strict, impartial judge that enforcing architectural rules, preventing security vulnerabilities, and maintaining code hygiene.

---

🤖 Why Warden?

AI coding assistants are powerful but prone to:

• Hallucinations: Inventing non-existent APIs or libraries.
• Subtle Bugs: Introducing edge cases that human reviewers miss.
• Security Flaws: Hardcoding secrets or using unsafe patterns.
• Drift: Deviating from established project architecture.

Warden provides the "Verify-Loop" mechanism to ensure every AI-generated change is validated against a rigorous set of rules.

---

🚀 Core Features

1. 🧠 AI-Native Integration (MCP & Hooks)

Warden is built to integrate directly with your AI Agent:

• Context Loading: Automatically injects .warden/AI_RULES.md into every session via Claude SessionStart hooks.
• Protocol Enforcement: Instructs AI agents to follow the Plan -> Execute -> Verify loop.
• Feedback Loop: Provides JSON/SARIF reports that AI agents can read to self-correct.

2. 🛡️ The 6-Frame Validation Pipeline

Warden runs your code through 6 specialized "Frames":

1. SecurityFrame (Critical): Detects SQLi, XSS, and hardcoded secrets.
2. ChaosEngineeringFrame (High): Validates error handling and resilience patterns.
3. FuzzTestingFrame (High): Checks for type safety and edge case handling.
4. PropertyTestingFrame (Medium): Verifies idempotency and invariants.
5. ArchitecturalConsistencyFrame (Medium): Enforces SOLID principles and file limits.
6. StressTestingFrame (Low): Identifies performance bottlenecks (N+1 queries).

3. 🚦 False Positive Management

Warden gives you granular control over what to check:

• Inline Suppression: Use # warden-ignore: rule-id to suppress specific issues.
• Global Config: Define suppressions in .warden/suppressions.yaml.
• File Exclusion: Respects .gitignore and supports .warden/ignore.yaml.

---

🏁 Quick Start

Installation

pip install warden-core

Initialization (The Critical Step)

To setup Warden for your project and configure your AI Agent:

warden init --agent

This command:

1. Analyzes your project structure.
2. Creates .warden/AI_RULES.md.
3. Configures Claude Code hooks (.claude/settings.json) to auto-load rules.
4. Sets up MCP configuration.
5. Creates .env and .env.example for your API keys.

The "Verify-Loop" Protocol

AI Agents working in a Warden project follow this strict protocol:

1. PLAN: Design the change (Task or Phase).
2. EXECUTE: Implement the code.
3. VERIFY: Run warden scan at the end of the task/phase.
   • IF PASS: Display Quality Score (X/10) 🚀
   • IF FAIL: Fix issues and Repeat.

---

🛠️ Command Reference

Command	Description
warden scan	Runs the full validation pipeline on the project.
warden validate <file>	Scans a single file for immediate feedback.
warden serve	Starts the MCP Server for AI integration.
warden doctor	Checks project health and configuration status.
warden install	Installs/Updates validation frames.
warden search <query>	Searches Warden Hub or local codebase.

---

📂 Project Structure

.
├── .warden/
│   ├── AI_RULES.md          # Protocol for AI Agents
│   ├── config.yaml          # Pipeline configuration
│   ├── ignore.yaml          # File exclusions
│   ├── suppressions.yaml    # False positive rules
│   └── reports/             # Scan results (JSON, SARIF)
├── .claude/
│   └── settings.json        # Hooks for auto-loading context
└── src/                     # Your source code

---

🧩 Architecture

Warden is designed with modular "Frames" and extends easily:

• Pipes & Filters: Sequential execution pipeline.
• Priority Groups: Critical security checks run first (Fail-Fast).
• LSP Synergy: Uses Language Server Protocol for deep code understanding.
• Vector Search: Semantic search for context-aware fixes.

---

📄 License

[License Info]