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-llm-simulatorgit 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-llm-simulator)<a href="https://agentmods.dev/skills/learnprompt/andrej-karpathy-skills/karpathy-llm-simulator"><img src="https://agentmods.dev/badge/skills/learnprompt/andrej-karpathy-skills/karpathy-llm-simulator/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-llm-simulator"><img src="https://agentmods.dev/badge/skills/learnprompt/andrej-karpathy-skills/karpathy-llm-simulator.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 | $0.00102 | $0.01322 |
| Opus 5 | $0.00051 | $0.00661 |
| Sonnet 5 | $0.00020 | $0.00264 |
| Haiku 4.5 | $0.00010 | $0.00132 |
Grade A, and why
karpathy-llm-simulator 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 — 146 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Skill 3: LLM as Simulator（LLM模拟器思维）
Source: https://x.com/karpathy/status/2037921699824607591 | https://x.com/karpathy/status/2049907410303865030 "Drafted a blog post → LLM argue the opposite" — 31k likes
Core Principle
Don't ask what the LLM thinks. Ask it to simulate what a diverse group of experts would argue.
LLMs are trained to please. A direct question gets a sycophantic answer. A simulation request gets a distribution of real perspectives — including the uncomfortable ones.
Karpathy's method: write a draft → ask LLM to argue the strongest possible opposite position → synthesize a better view.
The 4 Simulator Modes
Mode 1: Expert Debate Panel
Best for: technical decisions, architecture choices, research directions
Simulate a structured debate between these 3 expert personas on [TOPIC/DECISION]:
Expert A: [most optimistic / pro position]
Expert B: [most skeptical / con position]
Expert C: [pragmatic outsider / unexpected angle]
For each expert:
- State their core argument in 3 sentences
- Cite 2 specific examples or data points they'd use
- Identify what they'd say is the FATAL FLAW in the opposing view
After the debate, synthesize: what's the strongest hybrid position that survives all three critiques?
Topic: [YOUR_TOPIC]
My current position: [YOUR_DRAFT_VIEW]
Mode 2: Steel Man the Opposite
Best for: before publishing, before committing to a decision
I'm about to [ACTION / PUBLISH / DECIDE]:
[YOUR PLAN OR DRAFT]
Steel man the strongest possible argument AGAINST this. Be merciless.
Don't hedge. Don't say "while this has merit...".
Argue as if you genuinely believe the opposite and need to convince a skeptical expert.
Then: what would it take to make my original position survive this attack?
Mode 3: Pre-Mortem Simulation
Best for: project planning, product launches, major decisions
Imagine it's [DATE 6 MONTHS FROM NOW] and [YOUR PROJECT/PLAN] has failed completely.
Simulate 3 different failure modes — each from a different root cause:
1. Technical failure: what went wrong in the implementation?
2. Strategic failure: what assumption proved wrong?
3. Execution failure: what human/process error occurred?
For each: describe the specific sequence of events that led to failure.
Then: what early warning signals would have been visible by [DATE 1 MONTH FROM NOW]?
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 · 146 lines · 102 tokens per session scan A 3931266c9cf0
karpathy-llm-simulator is a skill published in the GitHub repository LearnPrompt/andrej-karpathy-skills (97 stars, last pushed 2mo ago), licensed MIT. It adds 102 tokens to every session and 1,322 once invoked, about $0.0005 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
systematic-debugging
Use when encountering any bug, test failure, or unexpected behavior, before proposing fixes.
local-ai-agents
Build local-first AI agents that run entirely on a developer workstation with Microsoft Foundry Local and Qwen function-calling models. Covers Small Language Models (SLMs), the OpenAI-compatible local endpoint, sandboxed local tools, local RAG with Chroma, local MCP servers, hybrid cloud/local routing, and the…
next-cache-components-adoption
Turn on Cache Components in a Next.js app and resolve the blocking routes it surfaces. Use when the user wants to enable, adopt, or migrate to Cache Components, flip the cacheComponents flag, work through a flood of blocking-prerender / instant validation errors, run the cache-components-instant-false codemod, or…
insight-error-page
Write or audit an insight-kind error page for the Next.js dev overlay. Use when creating a new errors/ .mdx page, auditing an existing one, or checking that a page matches the framework fix cards. Covers page structure, title alignment, FixCard cards with Copy prompt button, code snippets, terminology verification…
next-cache-components-optimizer
Drive a Next.js route to instant navigation by setting up an agentic loop, under Cache Components / PPR, on initial load (hard navigation) and client-side navigation (soft navigation). Encode the goal as a failing @next/playwright instant() e2e and work it to green, one verified route at a time; the shipped test then…
next-partial-prefetching-adoption
Turn on Partial Prefetching in a Next.js app and work through the insights it surfaces. Use when the user wants to enable or adopt Partial Prefetching, flip the partialPrefetching flag, opt routes in with export const prefetch = 'partial', audit Link prefetch={true} behavior, preserve existing prefetched UI with…