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 xiaohei-info/oh-my-multica --skill software-development-karpathy-guidelinesgit clone --depth 1 https://github.com/xiaohei-info/oh-my-multicaWrote 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/xiaohei-info/oh-my-multica/software-development-karpathy-guidelines)<a href="https://agentmods.dev/skills/xiaohei-info/oh-my-multica/software-development-karpathy-guidelines"><img src="https://agentmods.dev/badge/skills/xiaohei-info/oh-my-multica/software-development-karpathy-guidelines/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/xiaohei-info/oh-my-multica/software-development-karpathy-guidelines"><img src="https://agentmods.dev/badge/skills/xiaohei-info/oh-my-multica/software-development-karpathy-guidelines.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector pass
SkillSpector: 1 finding, up to low
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 →
- low Excessive Agency · line 24 Skill's behavior or capabilities extend beyond its stated purpose. Scope creep allows an agent to perform actions unrelated to its documented functionality, increasing the attack surface.Fix: Limit the skill's scope to its documented purpose. Remove instructions that enable the agent to perform actions outside its stated functionality.
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 | $0.00048 | $0.01249 |
| Opus 5 | $0.00024 | $0.00624 |
| Sonnet 5 | $0.00010 | $0.00250 |
| Haiku 4.5 | $0.00005 | $0.00125 |
Grade A, and why
software-development-karpathy-guidelines 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 11d 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 — 101 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Karpathy Guidelines
Behavioral guidelines to reduce common LLM coding mistakes, derived from Andrej Karpathy's observations on LLM coding pitfalls.
Tradeoff: These guidelines bias toward caution over speed. For trivial tasks, use judgment.
5. Answer the Question, Not the Implied Next Step
Research questions are not implementation mandates. When the user asks "does X", "will Y", "is Z", "how does", or "what would happen if" — that is not a request to change anything.
Common failure pattern:
- User: "Will these 40 tools waste my tokens?" → Agent: answers + modifies config
- User: "How does this repo work?" → Agent: clones and installs it
- User: "Clarify the mapping between current repo paths and design-doc terms" → Agent: starts bulk-renaming design docs instead of adding the requested bridge note
Special case for docs and repo-structure work:
- When the user asks for a README, orientation note, or terminology mapping, prefer the smallest bridging artifact that explains the relationship.
- Do NOT silently expand the scope into sweeping doc rewrites, mass renames, or terminology cleanup unless the user explicitly asked for that broader change.
- Preserve the distinction between design-layer terms (architecture roles like Agent Service / Agent Runtime) and physical filesystem names (e.g. app/ and hermes-agent/) when the user wants both to coexist.
Correct pattern:
- Answer the question with data.
- If there's a natural "would you like me to..." follow-up, offer it SEPARATELY after answering.
- Do not bundle the answer and the action in one turn unless explicitly asked.
Messaging/file-delivery pitfall:
- If the user asks you to send a file (e.g. "把文档发给我", "send me the markdown file"), they want an actual attachment or outbound file delivery — not the file contents pasted into chat.
- In that case, use the messaging/file-delivery path first. Do not substitute with an inline dump unless the user explicitly asks for the content.
- If your first attempt used a file marker/path but the user says they did not receive a file, switch to an explicit send step to the concrete channel/target and confirm delivery with the returned handle.
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.
- 11d ago First seen · 101 lines · 48 tokens per session scan A 00724d8fe1c7
software-development-karpathy-guidelines is a skill published in the GitHub repository xiaohei-info/oh-my-multica (26 stars, last pushed 4d ago), licensed MIT. It adds 48 tokens to every session and 1,249 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
cw-gates
Use before claiming any Codewhale change is done, green, or ready to land: the focused-to-broad verification ladder, the budget checks CI enforces, and the rules for what counts as a passing test.
interview
Ask one useful structured question at a time only when material product/implementation choices are genuinely missing; remember answers and produce a brief/spec. Discoverable facts should be investigated instead of asked.
verify
Exercise the real app/API/CLI and collect observable evidence; tests alone do not count as end-to-end verification.
mcp-host-styling-integration
Integrates MCP App UI with host theming system. Applies host CSS variables, handles onhostcontextchanged, safe area insets, display mode detection, and fullscreen configuration.
quality-hooks
Language-specific auto-lint/format/typecheck pipeline. Supports Python (ruff+pyright), TypeScript (prettier+eslint+tsc), Go (gofmt+golangci-lint). Auto-fix and convergence loops.
dd-synthesis
Synthesize findings from multiple specialist agents into a coherent investment recommendation. Use when combining reports from 9+ specialist agents into a single decision document.