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 toolboxmd/karpathy-wiki --skill using-karpathy-wikigit clone --depth 1 https://github.com/toolboxmd/karpathy-wikiWrote 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/toolboxmd/karpathy-wiki/using-karpathy-wiki)<a href="https://agentmods.dev/skills/toolboxmd/karpathy-wiki/using-karpathy-wiki"><img src="https://agentmods.dev/badge/skills/toolboxmd/karpathy-wiki/using-karpathy-wiki/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/toolboxmd/karpathy-wiki/using-karpathy-wiki"><img src="https://agentmods.dev/badge/skills/toolboxmd/karpathy-wiki/using-karpathy-wiki.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 6 findings, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 53 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Excessive Agency · line 55 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
- medium Agent Snooping · line 123 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
- medium Agent Snooping · line 124 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
- medium Agent Snooping · line 125 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
- medium Agent Snooping · line 126 Skill enumerates or reads other installed skills. Access to other skills' SKILL.md files or the skills directory reveals prompt instructions, capabilities, and secrets that should be invisible to peer skills.Fix: Remove all code or instructions that list or read other skills' files or directories. Skills should operate independently; cross-skill access is a privilege escalation.
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 | $0.00034 | $0.02034 |
| Opus 5 | $0.00017 | $0.01017 |
| Sonnet 5 | $0.00007 | $0.00407 |
| Haiku 4.5 | $0.00003 | $0.00203 |
Grade A, and why
using-karpathy-wiki 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 10d 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 — 129 lines — stays where its author put it; the contents beside it link to each section on GitHub.
karpathy-wiki — loader (read once per session, applies for the whole session)
You are operating in a project that uses the karpathy-wiki plugin. Durable knowledge surfaced in this conversation must be captured to a wiki so future sessions can find it.
Announce when you act
When you write a capture, run an ingest, or answer from the wiki, prefix your reply with one line:
Using the karpathy-wiki skill to [capture this / ingest pending captures / answer from wiki].
This is the ONLY wiki-mechanics text the user sees. Do not narrate orientation, capture authoring, dispatch mechanics, or state-machine progress. Do all the wiki work silently after the announce line.
Instruction priority
If the user's CLAUDE.md / AGENTS.md or an explicit user instruction conflicts with the rules below, follow the user. The wiki rules override the default system prompt where they conflict, but never override the user's direct instructions.
Iron laws
NO WIKI WRITE IN THE FOREGROUND
NO PAGE EDIT WITHOUT READING THE PAGE FIRST
NO SKIPPING A CAPTURE BECAUSE "IT DOESN'T LOOK WIKI-SHAPED"
NO ANSWERING ANY USER QUESTION WITHOUT ORIENTING FIRST
Captures go to <wiki>/.wiki-pending/<timestamp>-<slug>.md. The bounded dispatcher assigns each capture to the configured provider/model profile and ingests it in the background.
Orient before answering — the read protocol
For ANY user message that is asking a QUESTION (vs requesting a code edit, file operation, or pure command), you must orient on the wiki BEFORE drafting the answer. Pre-classifying a question as "general knowledge" or "not wiki-relevant" is the forbidden rationalization Iron Rule 4 was written to forbid.
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.
- 10d ago First seen · 129 lines · 34 tokens per session scan A ac284b273db1
using-karpathy-wiki is a skill published in the GitHub repository toolboxmd/karpathy-wiki (103 stars, last pushed 12d ago), licensed MIT. It adds 34 tokens to every session and 2,034 once invoked, about $0.0002 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
media-ingest
Ingest video, audio, PDF, book, screenshot, and GitHub repo content into the brain. Multi-format handling with entity extraction and backlink propagation. Covers video-ingest, youtube-ingest, and book-ingest subtypes.
mem0-oss-to-platform
Plan and then execute a migration of a project from the mem0 open-source / self-hosted SDK (the local Memory class) to the mem0 Platform / hosted / managed SDK (the MemoryClient class). Use this whenever a developer wants to move, switch, or migrate their mem0 usage off OSS/self-hosted to the hosted API — e.g.…
Cortex
Operate Cortex, the LifeOS memory system — the typed Knowledge Archive (People, Companies, Ideas, Research with typed related: links) plus recall of prior work sessions, ISAs, and conversations. Search, add, harvest, develop, ingest, distill, graph-navigate, recall. USE WHEN cortex, knowledge, knowledge base, search…
memory
Use when the user asks to remember, recall, forget, update, search, or inspect durable OpenSquilla memory, including profile facts in USER.md and long-term notes in MEMORY.md or memory//.md.
ha-data-stores
Map of Hope Agent's local data stores and safe read-only query workflow. Use when the user asks where Hope Agent stores data, wants to inspect sessions/messages/memory/logs/background jobs/knowledge indexes/settings, asks the model to query local app data, or debugging requires checking persisted state. Trigger…
establishing-project-context
Use when the user asks to establish shared project language, or project work exposes a conflicting, renamed, or deprecated domain term that needs active semantic modeling. Routine small tasks stay on the fast path.