# AI use cases for Andrej Karpathy Skills (multica-ai/andrej-karpathy-skills) Repo: [multica-ai/andrej-karpathy-skills](https://github.com/multica-ai/andrej-karpathy-skills) | Use cases: 3 | Page: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/ A single CLAUDE.md file that changes how Claude Code writes code, based on Andrej Karpathy's observations about LLM coding pitfalls. It encodes four principles: Think Before Coding, Simplicity First, Surgical Changes and Goal-Driven Execution, so the agent states assumptions, writes less code, touches only what the task needs and works toward verifiable success criteria. This file holds every use case of this repo. Each one starts with a "## " heading and ends with its URL. - Stop Claude Code from overengineering your code: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/stop-claude-code-from-overengineering/ - Add the Karpathy coding rules to a single repo's CLAUDE.md: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/add-karpathy-rules-to-one-project/ - Make your agent work toward tests instead of vague tasks: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/turn-tasks-into-verifiable-goals/ ## How to stop Claude Code from overengineering your code with Andrej Karpathy Skills Repo: [multica-ai/andrej-karpathy-skills](https://github.com/multica-ai/andrej-karpathy-skills) | Difficulty: Beginner | Time: 5 minutes | Basis: From the README Install the Karpathy guidelines as a plugin so every project gets the Simplicity First and Surgical Changes rules. **Who it is for:** A developer whose agent turns a 50 line fix into a 400 line abstraction and rewrites comments it did not need to touch. **The problem:** LLM agents bloat abstractions, add unrequested configurability and make orthogonal edits as side effects. **The outcome:** An agent that writes the minimum code for the task, matches existing style and leaves unrelated code alone. ### Steps 1. Open Claude Code in any project 2. Run `/plugin marketplace add forrestchang/andrej-karpathy-skills` to add the marketplace 3. Run `/plugin install andrej-karpathy-skills@karpathy-skills` to install the guidelines across all your projects 4. Give the agent a small change and check that every changed line traces to your request 5. If it notices unrelated dead code, expect it to mention it rather than delete it ### Prompt to paste into your agent ```text Install the andrej-karpathy-skills plugin in Claude Code, then make this change: . Follow Simplicity First and Surgical Changes: touch only what the task needs, match existing style, and list any unrelated issues you notice instead of fixing them. ``` ### Caveat The guidelines bias toward caution over speed; the README says trivial tasks do not need the full rigor. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/stop-claude-code-from-overengineering/ ## How to add the Karpathy coding rules to a single repo's CLAUDE.md with Andrej Karpathy Skills Repo: [multica-ai/andrej-karpathy-skills](https://github.com/multica-ai/andrej-karpathy-skills) | Difficulty: Beginner | Time: 5 minutes | Basis: From the README Drop the guidelines into one project's CLAUDE.md so the whole team's agent sessions follow them. **Who it is for:** A tech lead who wants consistent agent behavior in one shared repository without changing everyone's global setup. **The problem:** Each teammate's agent behaves differently, and global plugins do not travel with the repo. **The outcome:** A committed CLAUDE.md that makes every agent session in the repo state assumptions, keep changes small and verify against goals. ### Steps 1. For a new project run `curl -o CLAUDE.md https://raw.githubusercontent.com/forrestchang/andrej-karpathy-skills/main/CLAUDE.md` 2. For a project that already has a CLAUDE.md, append instead with `curl https://raw.githubusercontent.com/forrestchang/andrej-karpathy-skills/main/CLAUDE.md >> CLAUDE.md` 3. Add your project specific rules below the guidelines, such as test or style requirements 4. Commit CLAUDE.md so every teammate's agent reads the same rules 5. Cursor users can use the committed rule in .cursor/rules/karpathy-guidelines.mdc ### Prompt to paste into your agent ```text Append the andrej-karpathy-skills CLAUDE.md to this repo's existing CLAUDE.md without removing anything already there, add a short project section with , and show me the final diff before committing. ``` ### Caveat Risk, remote script: The curl commands download a file from GitHub into your repo; read it before committing since every agent session will follow it. Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/add-karpathy-rules-to-one-project/ ## How to make your agent work toward tests instead of vague tasks with Andrej Karpathy Skills Repo: [multica-ai/andrej-karpathy-skills](https://github.com/multica-ai/andrej-karpathy-skills) | Difficulty: Beginner | Time: 15 minutes | Basis: From the README Use the Goal-Driven Execution principle to turn requests like fix the bug into a test that must pass. **Who it is for:** A developer who keeps re-prompting because the agent declares success without proof. **The problem:** Weak criteria like make it work need constant clarification, and the agent stops before the job is really done. **The outcome:** Tasks phrased as verifiable goals, with a short plan where each step has a check, so the agent can loop on its own until they pass. ### Steps 1. Install the guidelines with the plugin or the CLAUDE.md option 2. Rewrite your request as a success criterion, for example write a test that reproduces the bug, then make it pass 3. Ask for a brief plan where each step ends with verify: and a check 4. Let the agent loop until the checks pass instead of approving each step by hand 5. Review the final diff against the success criteria you set ### Prompt to paste into your agent ```text Using the Goal-Driven Execution principle from andrej-karpathy-skills, turn this task into verifiable success criteria: . Write the failing test first, give me a numbered plan where each step has a verify check, then loop until all checks pass. ``` ### Caveat Disclaimer: Third-party repos we don't own or vouch for. AI-generated use cases can be wrong: review everything and run at your own risk. Full text: https://dedupedai.com/disclaimer/ URL: https://dedupedai.com/use-cases/multica-ai-andrej-karpathy-skills/turn-tasks-into-verifiable-goals/