agentsclimarketplace
Sign in

Karpathy guidelines

Skill BAKUGOS1/SkilledAgents-Toolkit/skills/codex/karpathy-guidelines

Coding discipline guardrails for writing, reviewing, debugging, or refactoring code with fewer LLM-agent mistakes. Use when a task touches source files, tests, architecture, bug fixes, feature implementation, refactors, cleanup, or code review and needs explicit assumptions, simple solutions, surgical diffs, existing-style consistency, and verifiable success criteria.From its SKILL.md

Install
npx -y skills add BAKUGOS1/SkilledAgents-Toolkit --skill karpathy-guidelines

Assembled from the repository path, not quoted from the project. Check it against their README if it does not work.

One thing to look at

  • 0 stars0 stars. Stars are a popularity signal and not a quality one, but at this level it is likely that nobody has read this closely except its author, and you would be relying on your own review.

SKILL.md

2.7 KB, 490 tokens by cl100k_base, as published. Nobody here has run it

Karpathy Guidelines

Overview

Use this skill to keep coding work calm, small, and verifiable. It is adapted for Codex from the MIT-declared karpathy-guidelines skill in multica-ai/andrej-karpathy-skills, with the same core intent: think before editing, avoid speculative complexity, touch only the necessary code, and verify the result.

Core Behavior

Think First

  • State assumptions when they affect the implementation.
  • If the user request has multiple plausible meanings, choose the safest interpretation only when it is low risk; otherwise ask.
  • Surface tradeoffs before major edits.
  • Push back on unnecessary work, broad rewrites, or features that do not serve the goal.

Keep It Simple

  • Implement the minimum behavior that satisfies the request.
  • Do not add future-proofing, configuration, abstractions, or extra features unless the current code genuinely needs them.
  • Prefer clear local changes over clever generalization.
  • If a solution grows large, pause and look for a smaller design.

Make Surgical Changes

  • Touch only files and lines needed for the task.
  • Preserve unrelated user changes and existing working behavior.
  • Match the repository's current style, naming, and organization.
  • Clean up imports, variables, and helper code made unused by your own edits.
  • Mention unrelated dead code or design issues instead of fixing them without permission.

Verify The Goal

  • Convert the request into success criteria before or during implementation.
  • For bug fixes, prefer a reproducing test or concrete repro before editing when feasible.
  • For feature work, identify the expected behavior, edge cases, and regression checks.
  • Run the smallest meaningful tests/builds/checks that match the risk.
  • Report what was verified and what remains unverified.

Workflow

  1. Read the relevant code and project instructions.
  2. Define the smallest verifiable goal.
  3. Make a brief plan for multi-step work.
  4. Edit narrowly.
  5. Verify with tests, builds, lint, screenshots, or manual checks as appropriate.
  6. Finalize with changed files, verification, risks, and assumptions.

Response Rules

  • Do not over-explain simple tasks.
  • Do not claim success without evidence.
  • When refusing or pushing back, give the practical reason and a better path.
  • For trivial tasks, apply the principles silently and keep the response short.

What ships with it: 1 file

260 B alongside SKILL.md

agents/

Gives 0 of the 12 instructions most context ai engineering skills give in 490 tokens

Counted across 1,193 of the 1,976 authors here whose files we hold, read 2026-08-07

  • Dispatch a fresh implementer subagent per taskin 48 of 1193, across 19 files
  • Dispatch a final code reviewer after all tasksin 33 of 1193, across 8 files
  • Provide full task text to the subagentin 30 of 1193, across 9 files
  • Review spec compliance before code qualityin 27 of 1193, across 10 files
  • Make the hook script executablein 26 of 1193, across 8 files
  • Re-snapshot after navigation or DOM changesin 25 of 1193, across 19 files
  • Read files before editing themin 22 of 1193, across 11 files
  • Answer subagent questions before proceedingin 22 of 1193, across 7 files
  • Mark task complete in TodoWrite after approvalin 22 of 1193, across 6 files
  • Merge hook into existing settingsin 21 of 1193, across 3 files
  • Ask if installation is global or projectin 20 of 1193, across 2 files
  • Copy the hook script to target locationin 20 of 1193, across 2 files

Said here and by no other author read

  • state assumptions affecting the implementation
  • choose the safest low-risk interpretation
  • surface tradeoffs before major edits
  • implement the minimum satisfying behavior
  • touch only files and lines needed
  • match the current repository style

Grouped from the skills themselves: near-identical wordings counted once, and counted by distinct author, so one author publishing three of these counts once. Length counted with cl100k_base; the agent that loads this file may tokenize it differently.

Keep looking

Skills are one crate of 326,144. Ordering is by how many stacks a row turns up in, so the top of any crate is what has actually been picked rather than what has the most stars.