My Favorite Skill is rick-rubin
AI wants to help. Sometimes it helps too much.
You ask for a bug fix. The agent fixes the bug, then refactors the surrounding code, adds error handling you did not request, and suggests three “improvements” to unrelated functions. The original task is done. So are five things you never asked for.
Scope creep in agentic development is not malicious. It is the model trying to be useful. But usefulness without discipline creates chaos. Projects die from accumulated “improvements” that nobody requested.
The Philosophy Behind the Skill
Rick Rubin is one of the most influential music producers alive. He produced the Beastie Boys, Johnny Cash, Jay-Z, Eminem, System of a Down, Shakira, and Mick Jagger. His approach is reduction. Strip away everything that does not serve the song.
From his book The Creative Act:
“With the extra layers removed, you may stand back and notice that the work is successful as it is, in its simplest form.”
The quality comes from what remains after you cut everything unnecessary. Not from what you add. Coding works the same way. Simple foundations support complex systems. Complexity without discipline collapses under its own weight.
Announcing Skills and Hooks
I have been building tooling to enforce discipline in agentic development. Two repositories:
4meta5/skills is a curated collection of Claude Code skills. Each skill is a markdown file that teaches Claude how to approach specific tasks with specific constraints.
4meta5/hooks is TypeScript tooling that scans projects and activates relevant skills automatically. It examines your codebase, identifies what kind of work you are doing, and loads the appropriate skills.
The Skills Available
The collection covers common development patterns:
Code Review: code-review-rust, code-review-ts, differential-review for focused feedback on changes.
Testing: tdd enforces red-green-refactor discipline.
Development Enforcement: dogfood ensures you use your own tools. rick-rubin enforces scope discipline.
Repository Maintenance: repo-hygiene keeps things clean. refactor-suggestions identifies improvement opportunities without acting on them.
Creation: skill-maker helps you build new skills for your own workflows.
Why rick-rubin is My Favorite
The rick-rubin skill encodes a simple philosophy: prefer simplicity, treat deletion as progress, defend scope explicitly.
The problem it solves is fundamental. AI models expand scope naturally. They see adjacent code and want to improve it. They notice patterns and want to abstract them. They find edge cases and want to handle them. All of this feels helpful. Most of it is not.
The solution is structured prompts that enforce discipline. The skill provides ten prompts for different situations, each calibrated to a specific level of scope tolerance.
Situations and Prompts
Starting a new feature: Use Prompt A to audit your planning documents before writing code. You feed it your requirements and it outputs a scope-trim proposal, a list of unclear requirements, and high-leverage clarification questions. The goal is to tighten scope before implementation begins.
Example: “Review this feature spec for a user dashboard. Identify anything speculative or unused. Give me a minimal execution set for the first milestone.”
Implementing from a plan: Use Prompt D when the plan is the contract. No deviations. No extra features. No cleanup unless explicitly required. The agent follows steps in order and makes the smallest possible diff.
Example: “Implement this plan exactly. If you encounter ambiguity, report it and proceed with the least-scope option. Do not add tests unless the plan specifies them.”
Reviewing a PR: Use Prompt F for line-by-line review assuming all additions are unnecessary until proven otherwise. No new features. No future-proofing. No stylistic refactors. For each violation, the agent provides the exact code location, why it violates scope, and minimal rollback steps.
Example: “Review this diff. Flag any code that does not trace directly to an explicit requirement. Assume everything is unnecessary until justified.”
After fixing bugs: Use Prompt H to identify root causes without proposing redesigns. The agent describes architectural decisions that enabled the bug and concrete failure patterns. It does not propose fixes unless strictly necessary.
Example: “The bug is fixed. Now tell me what design decision made this bug possible. Do not propose a refactor. Just describe the pattern.”
Vibe Coding as Art
Cutting scope is the art. The simplest solution that works is the beautiful one.
This is what Rick Rubin understands about music that applies equally to code. The song does not get better by adding more instruments. The feature does not get better by handling more edge cases. The codebase does not get better by abstracting every pattern.
Complex systems survive by being grounded in simple foundations. Every unnecessary addition is weight. Weight slows you down. Eventually it stops you entirely.
The rick-rubin skill is my favorite because it encodes this truth in a form Claude can follow. Prefer simplicity. Treat deletion as progress. Defend scope explicitly.
The best code is the code you did not write.