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 agentmods add rules/aksoftcode/aicrew/karpathy-guardrailsgit clone --depth 1 https://github.com/AKSoftCode/aicrewWhat 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 | $0.00194 | $0.00194 |
| Opus 5 | $0.00097 | $0.00097 |
| Sonnet 5 | $0.00039 | $0.00039 |
| Haiku 4.5 | $0.00019 | $0.00019 |
Grade A, and why
karpathy-guardrails 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 2d 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.
What it actually says
Karpathy Guardrails
Apply these four principles to every code change in this project.
Full agent spec: ~/Agents/agents/karpathy-guardrails.md
1. Think Before Coding
State the goal in one sentence before touching any file. Identify the entry point, call chain, and files that will change.
2. Simplicity First
Prefer fewer files, fewer abstractions, shortest path to passing tests. No new dependencies unless confirmed necessary.
3. Surgical Changes
Change only what is required for the goal. Do not reformat, rename, or touch code outside the affected region.
4. Goal-Driven Execution
Every edit traces back to the stated goal. Stop when the acceptance criterion is met. Do not gold-plate.
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.
- 2d ago First seen · 27 lines · 194 tokens per session scan A fc8f462cabb3
karpathy-guardrails is a cursor rule published in the GitHub repository AKSoftCode/aicrew (3 stars, last pushed 2mo ago), licensed MIT. It adds 194 tokens to every session, about $0.0010 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 2026-08-31.
Other cursor rules, from other repositories
x-harness
Use only light, standard, deep. Do not use small, medium, large.
ult-autoscaffold-content
Generate real starter content for a project's What-L2 (requirements) and How-L2 (architecture/conventions) CEP layers once ult-repo-layout has resolved their paths but found them empty — an honest, minimal, YAML-frontmatter-first overview document per layer for small/single targets, or graphify-informed per-module…
ult-cep-wizard
Launch a local, localhost-only browser wizard with two journeys — layout onboarding for a project with ult-repo-layout installed, initialized or not — guides an uninitialized or not-yet-discovered repo through Run Discover, then shows resolved layer/slot state through four labeled boxes and a directory picker, and…
ult-onboarding-index
Discover CEP-managed content already present in a target repo (What/How layer docs, compiled guidelines, context packages, decision ledger) via existence checks against layout-slots-registry.yaml-resolved paths, then write one canonical root AGENTS.md onboarding index plus thin per-tool pointer stubs…
project
Core project context for ai-context-kit.
demo-write-user-stories
Reference implementation that writes functional user stories from a feature description, optionally grounded in an approved CEP context package per CONSUMING-CONTEXT-PACKAGE.md, with per-story citations and the standard tag/reverse-index loop. A worked example for downstream teams building their own context-consuming…