What it does

guard-skills is a package of five review prompts—“skills” in the Skills CLI ecosystem—that you run after an AI agent produces code, tests, or docs. Each guard checks for failure modes that are common to LLM output: hallucinated APIs, mock-heavy tests that test nothing, docs that drift from the code, WordPress plugins that echo raw $_POST data, WooCommerce extensions that break HPOS.

The interesting bit

The author treats AI-generated code as a distinct risk category with its own smell profile. The clean-code-guard explicitly references published research on package hallucination and agents declaring success despite failed tests—problems most linters weren’t designed for. The skills load cheaply (small SKILL.md entrypoints) and pull deeper references only when relevant, so you’re not paying context-window tax for WooCommerce rules on a Go project.

Key highlights

• Five guards: clean-code-guard, test-guard, docs-guard, wp-guard, woo-guard
• Installs per-skill or globally via npx skills add; works with Claude Code, Codex, Cursor, OpenCode
• No executable code, network calls, or MCP dependencies—just Markdown + lightweight agent metadata
• Progressive-disclosure references keep the hot path small
• Maintainer validates with skillspector scan and quick_validate.py before release

Caveats

• Requires the Skills CLI ecosystem; not a standalone tool
• WordPress/WooCommerce guards are clearly the author’s depth area; general-language coverage is “any language” in principle but light on specifics beyond the AI-specific layer

Verdict

Worth a look if you’re already using Claude Code, Codex, or Cursor and want a structured second opinion before you git add. Skip it if you want a traditional linter or don’t work in the supported agent ecosystem.