karpathy-recipe skillA
karpathy-recipe is agent-read markdown (skill) from andersonlimahw/lemon-ai-hub: Applies Karpathy's "A Recipe for Training Neural Networks" adapted to software engineering. Forces a minimum runnable baseline before optimizing, one knob at a time, with verifiable eval at the beginning. Use when implementing a new feature, doing a non-trivial refactor, integrating an external service, or when the user says "implement X from scratch", "how to start feature Y", "recipe", "incremental approach", "baseline first"..
Indexed from public GitHub and served as immutable, content-addressed versions. Install it pinned to an exact SHA-256 with the mdr CLI, and every file is verified against the hash recorded here before it reaches your agent. The deterministic audit below grades the latest version, and the same file always earns the same grade.
What the file says
# karpathy-recipe Translation of [Recipe for Training Neural Networks](https://karpathy.github.io/2019/04/25/recipe/) to product features. Principle: **be paranoid, go slow, visualize everything**. ## Mandatory Flow ### 1. Become one with the data Before coding, read real data from the domain: - For a financial feature: 10 real transactions from `extratos_exemplo/` or from `financial_transactions` in staging - For AI: 5 examples of expected input/output from the user - List known edge cases (memory, ADRs, past issues) — do not invent them Output: 1 paragraph in the PR `## Data I looked at` with links/IDs. ### 2. Set up end-to-end skeleton + dumb baseline Implement the **complete path** with trivial logic first: - Server action returns a fixed mock - Repository returns `[]` - UI uses hardcoded data Validate that the entire pipeline runs (request → action → repo → DB → UI) before any real logic. **Do not optimize anything yet.** ### 3. Overfit the simplest case Next: make it work on **1 real input**, ignoring edge cases. Mock externals (Stripe, Belvo, OpenAI). Confirm expected logs/output. ### 4. Add one knob at a time …
Read the whole file at its exact version.
How to install
mdr add andersonlimahw/lemon-ai-hub/karpathy-recipe@git:20260614.2f67b30mdr add andersonlimahw/lemon-ai-hub/karpathy-recipe@sha256:192fa25ea5c56830Pin to a label to follow the author's releases, or to a sha256 to freeze the exact bytes forever. Either way the resolved hash is written to mdr.lock, and mdr install reproduces it on any machine.
[](https://markdownregistry.com/a/art_5rdu3ych4iqsi46i)
1 badge views in 30 days
Versions
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (3176 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
andersonlimahw/lemon-ai-hub · 14 stars · license none · pushed 2026-09-20 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_5rdu3ych4iqsi46i GET https://markdownregistry.com/api/v1/resolve?ref=andersonlimahw/lemon-ai-hub/karpathy-recipe GET https://markdownregistry.com/api/v1/blob/192fa25ea5c568306e73a33ca1b9b250593ac51d66ea244d56e2b91f36af9122
Your agent does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.