karpathy-coder
Use when writing, reviewing, or committing code to enforce Karpathy's 4 coding principles — surface assumptions before coding, keep it simple, make surgical changes, define verifiable goals. Triggers on "review my diff", "check complexity", "am I overcomplicating this", "karpathy check", "before I commit", or any code quality concern where the LLM might be overcoding.
- License 许可
- MIT
- 最近更新
- 2026-06-13
- Stars
- ★ 0
- Forks
- 0
01 前言 Front Matter
- name
- karpathy-coder
- tags
- [ "code-quality", "discipline", "karpathy", "simplicity", "surgical-changes", "anti-patterns", "review" ]
- author
- claude-code-skills
- context
- fork
- license
- MIT
- version
- 2.9.0
- description
- Use when writing, reviewing, or committing code to enforce Karpathy's 4 coding principles — surface assumptions before coding, keep it simple, make surgical changes, define verifiable goals. Triggers on "review my diff", "check complexity", "am I overcomplicating this", "karpathy check", "before I commit", or any code quality concern where the LLM might be overcoding.
- compatible_tools
- [ "claude-code", "codex-cli", "cursor", "antigravity", "opencode", "gemini-cli" ]
03 SKILL.md 正文
Karpathy Coder — Active Coding Discipline
Derived from Andrej Karpathy's observations on LLM coding pitfalls. This is not just guidelines — it ships Python tools that detect violations, a review agent, a slash command, and a pre-commit hook.
"The models make wrong assumptions on your behalf and just run along with them without checking. They don't manage their confusion, don't seek clarifications, don't surface inconsistencies, don't present tradeoffs, don't push back when they should."
"They really like to overcomplicate code and APIs, bloat abstractions, don't clean up dead code... implement a bloated construction over 1000 lines when 100 would do."
"LLMs are exceptionally good at looping until they meet specific goals... Don't tell it what to do, give it success criteria and watch it go."
— Andrej Karpathy
The four principles
1. Think Before Coding
Don't assume. Don't hide confusion. Surface tradeoffs.
- State assumptions explicitly. If uncertain, ask.
- If multiple interpretations exist, present them — don't pick silently.
- If a simpler approach exists, say so. Push back when warranted.
- If something is unclear, stop. Name what's confusing. Ask.
2. Simplicity First
Minimum code that solves the problem. Nothing speculative.
- No features beyond what was asked.
- No abstractions for single-use code.
- No "flexibility" or "configurability" that wasn't requested.
- No error handling for impossible scenarios.
- If you write 200 lines and it could be 50, rewrite it.
The test: Would a senior engineer say this is overcomplicated? If yes, simplify.
3. Surgical Changes
Touch only what you must. Clean up only your own mess.
- Don't "improve" adjacent code, comments, or formatting.
- Don't refactor things that aren't broken.
- Match existing style, even if you'd do it differently.
- If you notice unrelated dead code, mention it — don't delete it.
- Remove imports/variables/functions that YOUR changes made unused.
- Don't remove pre-existing dead code unless asked.
The test: Every changed line should trace directly to the user's request.
4. Goal-Driven Execution
Define success criteria. Loop until verified.
| Instead of... | Transform to... |
|---|---|
| "Add validation" | "Write tests for invalid inputs, then make them pass" |
| "Fix the bug" | "Write a test that reproduces it, then make it pass" |
| "Refactor X" | "Ensure tests pass before and after" |
For multi-step tasks, state a brief plan:
1. [Step] → verify: [check]
2. [Step] → verify: [check]
3. [Step] → verify: [check]
Slash command
/karpathy-check — Run the full 4-principle review on your staged changes.
Python tools (scripts/)
All tools are stdlib-only. Run with --help.
| Script | What it detects |
|---|---|
complexity_checker.py |
Over-engineering: too many classes, deep nesting, high cyclomatic complexity, unused params, premature abstractions |
diff_surgeon.py |
Diff noise: lines that don't trace to the stated goal — comment changes, style drift, drive-by refactors |
assumption_linter.py |
Hidden assumptions in a plan: unasked features, missing clarifications, silent interpretation choices |
goal_verifier.py |
Weak success criteria: vague plans without verifiable checks, missing test assertions |
Sub-agent
karpathy-reviewer — Runs all 4 principles against a diff. Dispatched by /karpathy-check or manually before committing.
Pre-commit hook
hooks/karpathy-gate.sh — runs complexity_checker.py and diff_surgeon.py on staged files. Warns (non-blocking) when violations are found. Wire it via .claude/settings.json or Husky.
References
references/karpathy-principles.md— the source quotes, deeper context, when to relax each principlereferences/anti-patterns.md— 10+ before/after examples across Python, TypeScript, and shellreferences/enforcement-patterns.md— how to wire hooks, CI integration, team adoption
When to relax
These principles bias toward caution over speed. For trivial tasks (typo fixes, obvious one-liners), use judgment. The principles matter most on:
- Non-trivial implementations (>20 lines changed)
- Code you don't fully understand
- Multi-step tasks with unclear requirements
- Anything that will be reviewed by humans
Cross-tool compatibility
Installs via plugin for Claude Code. For other tools, copy the principles into your schema file:
| Tool | Schema file |
|---|---|
| Claude Code | CLAUDE.md (auto-loaded by plugin) |
| Codex CLI | AGENTS.md |
| Cursor | AGENTS.md or .cursorrules |
| Antigravity / OpenCode / Gemini CLI | AGENTS.md |
Related skills (chains via context: fork)
self-eval— honest quality scoring after completing workcode-reviewer— broader code review; karpathy-coder focuses on the 4 LLM-specific pitfallsllm-wiki— compound knowledge; karpathy-coder ensures you don't overcomplicate while building it
04 同仓库更多 Skills
youtube-full
Use when the user needs YouTube transcripts, video search, channel browsing, playlist extraction, or content monitoring. Trigger phrases: 'get the transcript for', 'search YouTube for', 'what are the latest videos on', 'list this playlist', 'monitor this channel', or any request involving a YouTube URL, video ID, or @handle. Do NOT use for downloading video or audio files, YouTube engagement data (likes, comments), or private/age-restricted videos.
x-twitter-growth
X/Twitter growth engine for building audience, crafting viral content, and analyzing engagement. Use when the user wants to grow on X/Twitter, write tweets or threads, analyze their X profile, research competitors on X, plan a posting strategy, or optimize engagement. Complements social-content (generic multi-platform) with X-specific depth: algorithm mechanics, thread engineering, reply strategy, profile optimization, and competitive intelligence via web search.
write-a-skill
Create new agent skills with proper structure, progressive disclosure, and bundled resources. Use when user wants to create, write, build, or author a new skill.
workflow-builder
Design and write deterministic multi-agent workflow scripts (.js files in .claude/workflows/) for Claude Code's Workflow tool. Use when a user wants to build, create, author, scaffold, or run a custom Claude Code workflow, orchestrate sub-agents (fan-out, pipeline, loop, judge-panel), or automate a repeatable multi-step task across fresh-context agents.
webinar-marketing
When the user wants to plan, promote, run, or improve a webinar or virtual event to generate and convert demand. Use when the user mentions 'webinar,' 'virtual event,' 'online event,' 'live demo,' 'virtual summit,' 'workshop,' 'masterclass,' 'fireside chat,' 'roundtable,' 'registration funnel,' 'show-up rate,' 'attendance rate,' 'webinar promotion,' 'webinar follow-up,' or 'on-demand webinar.' Also use when they have a webinar that isn't converting — low registrations, low show-up, or attendees who don't buy — and want to diagnose and fix it. Covers the full funnel: registration, promotion, show-up, live engagement, live-to-close, and post-event nurture. Distinct from launch-strategy (full product launches) and email-sequence (lifecycle nurture) — this is the end-to-end webinar/event motion. NOT for in-person field events logistics, and NOT for generic lifecycle email (use email-sequence).
vpe-advisor
VP of Engineering advisory for startups: delivery throughput (DORA 4 metrics + bottleneck identification), engineering hiring funnel (sourcing → screen → onsite → offer conversion + time-to-fill + pipeline gap), engineering team structure (squad/tribe/chapter design + tech-lead manager-trigger thresholds), and production discipline (on-call, deployment cadence, postmortem culture). Use when sprint velocity is dropping, eng hiring is broken, team structure is unclear, or deciding when to add a tech-lead manager. NOT a CTO skill (which owns architecture) — VPE owns delivery operations and how the team ships.