karpathy-extended
Skill0 starsChanged 10 days agoMITView on GitHub
Five behavioral guidelines that catch the most common LLM coding failure modes -- silently running with wrong assumptions, overcomplicating simple changes, making edits that drift outside the requested scope, chasing vague goals with no way to know when done, and stating unverified guesses with confident tone. Consult this proactively for any non-trivial coding task -- writing new code, reviewing a diff, refactoring, debugging, or planning multi-step work -- and whenever about to state a technical claim you haven't verified. Don't wait for the user to ask for "best practices" or "guidelines" explicitly; these apply by default to real coding work, not just when named.
1 other project uses a version of this.
Adopted by
Most recently taken up by
- sudeepgupta90/extended-andrej-karpathy-skillword for word, first seen Jul 30, 2026
Other versions1 project uses a version of this.
Word for word · 1
- sudeepgupta90/extended-andrej-karpathy-skill★ 0100% same
Edited · 0
None found.
Wrote their own version · 0
None found.
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository46 other files it gives its agents.
- AGENTS.mdAGENTS.md
- skills/caveman-learn/CLAUDE.mdCLAUDE.md
- brainstormingSkill
- cavecrewSkill
- caveman-commitSkill
- caveman-compressSkill
- caveman-discoverSkill
- caveman-evidence-reviewSkill
- caveman-exploreSkill
- caveman-helpSkill
- caveman-learnSkill
- caveman-manageSkill
- caveman-optimizeSkill
- caveman-reviewSkill
- caveman-setupSkill
- cavemanSkill
- caveman-statsSkill
- diagnosing-superpowersSkill
- dispatching-parallel-agentsSkill
- executing-plansSkill
- finishing-a-development-branchSkill
- handoffSkill
- investigate-firstSkill
- lean-buildSkill
- mcp-builderSkill
- megacaveSkill
- migrationSkill
- plugin-doctorSkill
- receiving-code-reviewSkill
- requesting-code-reviewSkill
- safe-refactorSkill
- scholastic-disputationSkill
- socratic-inquirySkill
- subagent-driven-developmentSkill
- surgical-patchSkill
- systematic-debuggingSkill
- terrasharkSkill
- test-driven-developmentSkill
- ultracaveSkill
- using-foundational-thinkingSkill
- using-git-worktreesSkill
- using-superpowersSkill
- verification-before-completionSkill
- verify-and-stopSkill
- writing-plansSkill
- writing-skillsSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
21 rules. About 0 are common, 20 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈20
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 19
If multiple interpretations exist, present them - don't pick silently.
- Distinctive: line 20
If a simpler approach exists, say so.
- Distinctive: line 21
If something is unclear, stop.
- Distinctive: line 27
No features beyond what was asked.
- Distinctive: line 28
No abstractions for single-use code.
- Distinctive: line 30
No error handling for impossible scenarios.
- Distinctive: line 31
If you write 200 lines and it could be 50, rewrite it.
- Distinctive: line 33
Ask yourself: "Would a senior engineer say this is overcomplicated?" If yes, simplify.
- Distinctive: line 39
When editing existing code:
- Distinctive: line 40
Don't "improve" adjacent code, comments, or formatting.
- Distinctive: line 41
Don't refactor things that aren't broken.
- Distinctive: line 43
If you notice unrelated dead code, mention it - don't delete it.
- Distinctive: line 45
When your changes create orphans:
- Very specific: line 46
Remove imports/variables/functions that YOUR changes made unused.
- Distinctive: line 47
Don't remove pre-existing dead code unless asked.
- Distinctive: line 73
When your knowledge is incomplete, inferred, or unverified:
- Distinctive: line 74
Use "possibly", "likely", "I'm not certain", or "you should verify this" — don't omit them.
- Distinctive: line 76
If a claim requires external verification before acting on it, flag that explicitly, *before* the developer would act on it — not buried at …
- Distinctive: line 77
Never let confident tone substitute for confident knowledge.
- Distinctive: line 79
When you notice you're filling a gap with an assumption:
- Distinctive: line 82
Don't bury uncertainty at the end of a long confident response.
≈ “Common” is an estimate: at least 1 in 5 of the 364,980 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Karpathy Extended Guidelines
Five behavioral principles for coding agents, each naming a specific way agents fail silently -- the fix is to make the failure visible instead of letting it slide by.
Tradeoff: these guidelines bias toward caution over speed. For trivial tasks (a typo fix, an obvious one-liner), use judgment rather than applying the full rigor below.
1. Think Before Coding
Don't assume. Don't hide confusion. Surface tradeoffs.
Before implementing:
- State your 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.
Ask yourself: "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.
When editing existing code:
- 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.
When your changes create orphans:
- 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.
Transform tasks into verifiable goals:
- "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]Strong success criteria let you loop independently. Weak criteria ("make it work") require constant clarification.
5. Signal Uncertainty
Don't state guesses as facts. When confidence is low, say so.
When your knowledge is incomplete, inferred, or unverified:
- Use "possibly", "likely", "I'm not certain", or "you should verify this" — don't omit them.
- Distinguish between what you know and what you're inferring.
- If a claim requires external verification before acting on it, flag that explicitly, before the developer would act on it — not buried at the end.
- Never let confident tone substitute for confident knowledge.
When you notice you're filling a gap with an assumption:
- Name the gap: "I don't have visibility into X, so I'm assuming Y."
- Offer to stop rather than guess: "I can proceed on that assumption, or you can verify first."
- Don't bury uncertainty at the end of a long confident response.
The test: Could a developer act on this response and only discover it was wrong after the damage is done? If yes, the uncertainty wasn't signalled clearly enough.
MIT. Attribution and license: https://github.com/sudeepgupta90/extended-andrej-karpathy-skill
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

