This repository provides a single Markdown file (CLAUDE.md) containing Linus Torvalds-inspired guidelines aimed at improving the behavior of AI coding assistants. It addresses common AI pitfalls such as making assumptions, overcomplicating code, and shipping unverified changes. The goal is to instill a pragmatic, data-structure-first, and bloat-averse approach in AI-generated code, benefiting developers seeking more robust and maintainable AI-assisted development.

How It Works

The core approach is structured around four principles derived from Torvalds' coding philosophy: Data First, Simplicity First, Surgical Changes, and Show Me the Code. These principles directly counter AI tendencies by prioritizing data modeling to prevent hidden edge cases, enforcing minimal code for problem-solving, advocating for precise modifications without unrelated churn, and demanding verifiable code through tests or benchmarks. This methodology aims to produce less speculative, more focused, and demonstrably correct code.

Quick Start & Requirements

• Primary install:
  • Recommended: npx skills add leopiney/linus-torvalds-skills
  • Per-project: curl -o CLAUDE.md https://raw.githubusercontent.com/leopiney/linus-torvalds-skills/main/CLAUDE.md
• Prerequisites: Node.js/npm (for npx), curl. Includes a pre-configured rule file (.cursor/rules/torvalds-doctrine.mdc) for integration with the Cursor IDE. See CURSOR.md for setup instructions.
• Resource Footprint: Minimal; primarily involves downloading a file or executing a command-line tool.

Highlighted Details

• Bogus Shit Detector: Explicitly instructs AI to identify and flag categories of poor engineering practices, such as "abstraction with no payoff," "brain-damaged API," and "random churn."
• Linux-Style Review Comments: Provides AI with direct, critical phrases mimicking kernel maintainer reviews (e.g., "This is bogus shit. Fix the data structure instead of piling on conditionals.").
• Expanded LKML-Style Phrasebook: Offers a more extensive set of blunt critiques for AI to employ on code, diffs, or commit messages (e.g., "too ugly to live," "pure and utter garbage").
• Cursor IDE Integration: Comes with a committed project rule for automatic application of the doctrine within the Cursor editor.

Maintenance & Community

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

Licensing & Compatibility

• License: MIT.
• Restrictions: Explicitly designated as a "parody skill" and stated "should not actually be used." This implies it is not intended for integration into production environments or serious AI model training.

Limitations & Caveats

The project is explicitly a parody and is not intended for actual implementation or use in real-world development workflows. It serves as a set of guidelines or a conceptual tool rather than a functional software component.