Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add LearnPrompt/andrej-karpathy-skills --skill karpathy-understanding-firstgit clone --depth 1 https://github.com/LearnPrompt/andrej-karpathy-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/learnprompt/andrej-karpathy-skills/karpathy-understanding-first)<a href="https://agentmods.dev/skills/learnprompt/andrej-karpathy-skills/karpathy-understanding-first"><img src="https://agentmods.dev/badge/skills/learnprompt/andrej-karpathy-skills/karpathy-understanding-first/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/learnprompt/andrej-karpathy-skills/karpathy-understanding-first"><img src="https://agentmods.dev/badge/skills/learnprompt/andrej-karpathy-skills/karpathy-understanding-first.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00114 | $0.01611 |
| Opus 5 | $0.00057 | $0.00805 |
| Sonnet 5 | $0.00023 | $0.00322 |
| Haiku 4.5 | $0.00011 | $0.00161 |
Grade A, and why
karpathy-understanding-first scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 12d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 168 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill 8: Understanding > Outsourcing（理解 > 外包）
Source: https://x.com/karpathy/status/2049907410303865030 "You can outsource your thinking but you cannot outsource your understanding." — 46k likes Also: Jagged Capabilities Modeling (Sequoia conversation)
Core Principle
Outsourcing is fine. Atrophy is not.
Use AI to go 10x faster. But every time you skip understanding something, you're borrowing from your own future. Understanding compounds. Blindly shipping agent output does not.
Karpathy's "jagged capabilities" insight: LLMs are wildly good at some things (verifiable, high-data domains) and silently wrong at others (off-distribution, novel combinations). You need a mental map of which is which.
The Understanding Audit
After every agent-assisted work session, run this:
I just had an AI help me [DESCRIBE WHAT WAS BUILT/WRITTEN/DECIDED].
Help me audit my own understanding:
1. List the 5 key decisions that were made (technical choices, assumptions, trade-offs)
2. For each decision: would I be able to explain WHY this choice was made to a colleague?
Mark each: UNDERSTOOD / SHALLOW / BLACK BOX
3. For the BLACK BOX items: give me a 3-sentence explanation I can verify
4. What would I need to read/learn to make the SHALLOW items UNDERSTOOD?
Be honest. If I'm shipping something I don't understand, I need to know.
Jagged Capabilities Map
LLMs are NOT uniformly capable. Use this mental model:
HIGH Reliability (on the rails)
These are well-defined, verifiable, data-rich domains:
- Code in popular languages (Python, JS, SQL)
- Text summarization and rewriting
- Structured data extraction
- Well-known algorithms
- Common patterns (REST APIs, SQL queries, regex)
MEDIUM Reliability (use carefully, verify)
- Less popular languages (Rust, Haskell, Elixir)
- Integrating multiple systems together
- Anything involving specific versions/APIs after training cutoff
- Multi-step reasoning chains
- Math beyond simple arithmetic
LOW Reliability (always verify independently)
- Cutting-edge research (may hallucinate papers)
- Novel combinations of technologies
- Domain-specific knowledge you can't easily verify
- Legal/medical/financial specifics
- Anything where being wrong is expensive
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 12d ago First seen · 168 lines · 114 tokens per session scan A 8d4fa98dd4ca
karpathy-understanding-first is a skill published in the GitHub repository LearnPrompt/andrej-karpathy-skills (97 stars, last pushed 2mo ago), licensed MIT. It adds 114 tokens to every session and 1,611 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
hr-onboarding
A new-hire onboarding plan as a single page — first week schedule, buddy + manager intro, learning track, equipment checklist, and "you're set when…" outcomes. Use when the brief mentions "onboarding", "new hire", "first week plan", or "入职".
book-mirror
Take any book (EPUB/PDF), produce a personalized chapter-by-chapter analysis. Each chapter is preserved in detail (The Chapter) and mirrored back to the reader's actual life (The Mirror) using brain context. The mirror observes and resonates — a friend pointing out parallels, NOT a consultant rearranging the reader's…
miniapp
Build a tiny interactive HTML playground only when someone asks to see, play with, or step through a mechanism.
eli5
Explain research, papers, or technical ideas in plain English with minimal jargon, concrete analogies, and clear takeaways. Use when the user says "ELI5 this", asks for a simple explanation of a paper or research result, wants jargon removed, or asks what something technically dense actually means.
deck-course-module
A course or workshop slide template with persistent learning goals, teaching pages, multiple-choice self-tests, and a wrap-up.
master-yinguang
A reference-based assistant for questions about Yinguang and Pure Land Buddhism, a Buddhist tradition focused on faith, ethical living, and practice connected with rebirth in the Pure Land. It can answer in Yinguang’s historical teaching style.