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 frankekn/george-hotz-skills --skill geohot-guidelinesgit clone --depth 1 https://github.com/frankekn/george-hotz-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/frankekn/george-hotz-skills/geohot-guidelines)<a href="https://agentmods.dev/skills/frankekn/george-hotz-skills/geohot-guidelines"><img src="https://agentmods.dev/badge/skills/frankekn/george-hotz-skills/geohot-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/frankekn/george-hotz-skills/geohot-guidelines"><img src="https://agentmods.dev/badge/skills/frankekn/george-hotz-skills/geohot-guidelines.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00091 | $0.00770 |
| Opus 5 | $0.00046 | $0.00385 |
| Sonnet 5 | $0.00018 | $0.00154 |
| Haiku 4.5 | $0.00009 | $0.00077 |
Grade A, and why
geohot-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 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 — 39 lines — stays where its author put it; the contents beside it link to each section on GitHub.
geohot guidelines — radical simplicity
Distilled from George Hotz's public philosophy. The reflex these rules install: the enemy is complexity, and the move is almost always to remove, not add. Use on any non-trivial implementation, refactor, or review — not on one-liners.
"You can always make your software do more. The magic is when you can make your software do more without adding complexity — because complex things eventually collapse under their own weight."
1. Complexity is the enemy
More capability at equal or lower complexity is the only real win. Adding a feature by adding a layer is a loss.
- Solve the actual request with the least machinery that works; no speculative abstraction, configurability, or future-proofing unless asked.
- Smell: introducing a new manager/handler/factory/adapter to do one thing → stop, ask what to remove instead.
2. You have never refactored enough
"Your code can get smaller, your code can get simpler, your ideas can be more elegant." tinygrad aims to express ML with roughly 25 low-level ops, compared with roughly 250 in XLA / PrimTorch.
- Delete-first: before touching a subsystem, ask "can this not exist?" Collapse duplicated logic to one source of truth and delete the glue that kept the copies in sync. The best change is often a negative diff. LOC is debt, not output.
- Fence: prove it's truly dead first (read the callsites); scope deletion to what the task touches, not adjacent code; never delete a real invariant (money, auth, data-integrity, lifecycle) to "simplify" — that's a bug.
3. Wide interfaces mean your abstraction is wrong
- A wide interface is a design signal. If a function is growing toward five arguments, treat the boundary as suspect. Fix the boundary (split the responsibility, or pass one well-shaped value) — don't hide the width behind an options/
kwargsbag.
4. Understand the whole stack — nothing is magic
tinygrad as "the RISC of the ML stack — extreme simplicity that allows anyone to understand it."
- Build the smallest runnable unit, print/inspect its real inputs and outputs, and check them against expectation before stacking more on top. The machine is not magic — look at the actual value; don't reason about what it "should" be.
- Read the definition + the callsites before concluding. Don't assert code "works" or "is safe" without seeing it; don't cargo-cult a pattern you don't understand.
- AI-generated code must be read and validated line by line; speed is not correctness.
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 · 39 lines · 91 tokens per session scan A f89831b70271
geohot-guidelines is a skill published in the GitHub repository frankekn/george-hotz-skills (12 stars, last pushed 3mo ago), licensed MIT. It adds 91 tokens to every session and 770 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
andrej-karpathy
Behavioral guidelines to reduce common LLM coding mistakes. Use when writing, reviewing, or refactoring code to avoid overcomplication, make surgical changes, surface assumptions, and define verifiable success criteria.
sdd-tasks
Break an SDD change into implementation tasks. Trigger: orchestrator launches task planning for a change.
go-testing
Trigger: Go tests, go test coverage, Bubbletea teatest, golden files. Apply focused Go testing patterns.
pr-blocker-summarizer
Summarizes open pull requests into a blockers-first standup digest. Activates when the user asks to summarize open PRs, find blocked pull requests, generate a PR standup, or triage review backlog from a PR export.
review-work
Post-implementation gate review: run manual QA on the real surface yourself, then launch ONE gate reviewer (never a panel) to audit goal, constraints, code quality, security, missed context, and QA evidence. Use before a PR handoff or when the user explicitly asks to review completed work.
ijfw-cross-audit
Generate a cross-platform multi-model audit (Trident) on a diff, brief, or artifact. Trigger: 'cross audit', 'Trident', 'second opinion', 'check with other models', 'check with other AIs', 'cross-check this', 'get another perspective', /cross-audit.