This project provides specialized "guard skills" designed as second-pass quality gates for AI-generated code, tests, and documentation. It targets developers and AI coding agents, offering a mechanism to catch systematic failure modes common in AI outputs before integration, thereby improving artifact quality and reliability.

How It Works

The core approach involves a collection of distinct "guard skills" (e.g., clean-code-guard, test-guard) that act as review passes on AI-generated work. Each skill is structured with a lightweight SKILL.md entrypoint and progressive-disclosure references, ensuring efficient loading. This design allows deeper guidance to be fetched only when needed, optimizing performance while providing targeted validation against common AI pitfalls like hallucinated APIs, error swallowing, or documentation drift.

Quick Start & Requirements

Installation is managed via the Skills CLI:

• Install all guards: npx skills add amElnagdy/guard-skills
• Install specific guards: npx skills add amElnagdy/guard-skills --skill <guard-name>
• Install for specific agents: npx skills add amElnagdy/guard-skills --skill <guard-name> --agent <agent-name>

Requires the Skills CLI and compatible AI agents such as Claude Code, Codex, Cursor, or OpenCode.

Highlighted Details

• clean-code-guard: Catches LLM code smells, over-abstraction, broad error swallowing, bad naming, SOLID/DRY/KISS/YAGNI violations, and AI-specific issues like hallucinated APIs or premature abstraction.
• test-guard: Enforces quality for generated tests, preventing mock abuse, duplicate tests, and implementation-detail assertions, while promoting rules like mocking only at system boundaries.
• docs-guard: Verifies documentation accuracy by treating it as a list of claims, ensuring symbols exist, samples run, and documentation aligns with the codebase.
• wp-guard: A specialized guard for WordPress code, focusing on escaping, sanitization, nonces, prepared queries, i18n, and caching discipline.
• woo-guard: Extends wp-guard for WooCommerce, addressing HPOS compatibility, CRUD operations over direct meta, checkout validation, and money-handling accuracy.

Maintenance & Community

No specific details regarding maintainers, community channels (e.g., Discord/Slack), or roadmaps are provided in the README.

Licensing & Compatibility

The project is licensed under the MIT license, permitting broad use and compatibility, including for commercial purposes, without copyleft restrictions. It integrates with various AI coding agents via the Skills CLI.

Limitations & Caveats

This package functions as a set of focused review gates rather than a comprehensive development framework. Its primary strength lies in post-generation artifact review, although it can be invoked proactively during writing. It specifically targets systematic AI failure modes rather than providing exhaustive platform coverage.