MCP HubMCP Hub
Retour aux compétences

karpathy-llm-wiki

Astro-Han
Mis à jour 5 days ago
1,047
143
1,047
Voir sur GitHub
Métaaidesign

À propos

Cette compétence aide les développeurs à construire et maintenir une base de connaissances personnelle alimentée par un LLM en gérant les matériaux sources dans un répertoire `raw/` et les articles compilés dans un `wiki/`. Elle s'active pour des tâches telles que l'ingestion de sources, l'interrogation de la base de connaissances, ou le contrôle qualité, suivant le principe que le LLM écrit et maintient le wiki tandis que l'humain lit et pose des questions.

Installation rapide

Claude Code

Recommandé
Principal
npx skills add Astro-Han/karpathy-llm-wiki -a claude-code
Commande PluginAlternatif
/plugin add https://github.com/Astro-Han/karpathy-llm-wiki
Git CloneAlternatif
git clone https://github.com/Astro-Han/karpathy-llm-wiki.git ~/.claude/skills/karpathy-llm-wiki

Copiez et collez cette commande dans Claude Code pour installer cette compétence

Documentation

Karpathy LLM Wiki

Build and maintain a personal knowledge base using LLMs. You manage two directories: raw/ (immutable source material) and wiki/ (compiled knowledge articles). Sources go into raw/, you compile them into wiki articles, and the wiki compounds over time.

Core ideas from Karpathy:

  • "The LLM writes and maintains the wiki; the human reads and asks questions."
  • "The wiki is a persistent, compounding artifact."

Architecture

Three layers, all under the user's project root:

raw/ — Immutable source material. You read, never modify. Organized by topic subdirectories (e.g., raw/machine-learning/).

wiki/ — Compiled knowledge articles. You have full ownership. Organized by topic subdirectories, one level only: wiki/<topic>/<article>.md. Contains two special files:

  • wiki/index.md — Global index. One row per article, grouped by topic, with link + summary + Updated date.
  • wiki/log.md — Append-only operation log.

SKILL.md (this file) — Schema layer. Defines structure and workflow rules.

Templates live in references/ relative to this file. Read them when you need the exact format for raw files, articles, archive pages, or the index.

Initialization

Triggers only on the first Ingest. Check whether raw/ and wiki/ exist. Create only what is missing; never overwrite existing files:

  • raw/ directory (with .gitkeep)
  • wiki/ directory (with .gitkeep)
  • wiki/index.md — heading # Knowledge Base Index, empty body
  • wiki/log.md — heading # Wiki Log, empty body

If Query or Lint cannot find the wiki structure, tell the user: "Run an ingest first to initialize the wiki." Do not auto-create.


Ingest

Fetch a source into raw/, then compile it into wiki/. Always both steps, no exceptions.

Fetch (raw/)

  1. Get the source content using whatever web or file tools your environment provides. If nothing can reach the source, ask the user to paste it directly.

  2. Pick a topic directory. Check existing raw/ subdirectories first; reuse one if the topic is close enough. Create a new subdirectory only for genuinely distinct topics.

  3. Save as raw/<topic>/YYYY-MM-DD-descriptive-slug.md.

    • Slug from source title, kebab-case, max 60 characters.
    • Published date unknown → omit the date prefix from the file name (e.g., descriptive-slug.md). The metadata Published field still appears; set it to Unknown.
    • If a file with the same name already exists, append a numeric suffix (e.g., descriptive-slug-2.md).
    • Include metadata header: source URL, collected date, published date.
    • Preserve original text. Clean formatting noise. Do not rewrite opinions.

    See references/raw-template.md for the exact format.

Compile (wiki/)

Determine where the new content belongs:

  • Same core thesis as existing article → Merge into that article. Add the new source to Sources/Raw. Update affected sections.
  • New concept → Create a new article in the most relevant topic directory. Name the file after the concept, not the raw file.
  • Spans multiple topics → Place in the most relevant directory. Add See Also cross-references to related articles elsewhere.

These are not mutually exclusive. A single source may warrant merging into one article while also creating a separate article for a distinct concept it introduces. In all cases, check for factual conflicts: if the new source contradicts existing content, annotate the disagreement with source attribution. When merging, note the conflict within the merged article. When the conflicting content lives in separate articles, note it in both and cross-link them.

See references/article-template.md for article format. Key points:

  • Sources field: author, organization, or publication name + date, semicolon-separated.
  • Raw field: markdown links to raw/ files, semicolon-separated.
  • Relative paths from wiki/<topic>/ use ../../raw/<topic>/<file>.md (two levels up to project root).

Cascade Updates

After the primary article, check for ripple effects:

  1. Scan articles in the same topic directory for content affected by the new source.
  2. Scan wiki/index.md entries in other topics for articles covering related concepts.
  3. Update every article whose content is materially affected. Each updated file gets its Updated date refreshed.

Archive pages are never cascade-updated (they are point-in-time snapshots).

Post-Ingest

Update wiki/index.md: add or update entries for every touched article. When adding a new topic section, include a one-line description. The Updated date reflects when the article's knowledge content last changed, not the file system timestamp. See references/index-template.md for format.

Append to wiki/log.md:

## [YYYY-MM-DD] ingest | <primary article title>
- Updated: <cascade-updated article title>
- Updated: <another cascade-updated article title>

Omit - Updated: lines when no cascade updates occur.


Query

Search the wiki and answer questions. Examples of triggers:

  • "What do I know about X?"
  • "Summarize everything related to Y"
  • "Compare A and B based on my wiki"

Steps

  1. Read wiki/index.md to locate relevant articles.
  2. Read those articles and synthesize an answer.
  3. Prefer wiki content over your own training knowledge. Cite sources with markdown links: [Article Title](wiki/topic/article.md) (project-root-relative paths for in-conversation citations; within wiki/ files, use paths relative to the current file).
  4. Output the answer in the conversation. Do not write files unless asked.

Archiving

When the user explicitly asks to archive or save the answer to the wiki:

  1. Write the answer as a new wiki page. See references/archive-template.md. When converting conversation citations to the archive page, rewrite project-root-relative paths (e.g., wiki/topic/article.md) to file-relative paths (e.g., ../topic/article.md or article.md for same-directory).
    • Sources: markdown links to the wiki articles cited in the answer.
    • No Raw field (content does not come from raw/).
    • File name reflects the query topic, e.g., transformer-architectures-overview.md.
    • Place in the most relevant topic directory.
  2. Always create a new page. Never merge into existing articles (archive content is a synthesized answer, not raw material).
  3. Update wiki/index.md. Prefix the Summary with [Archived].
  4. Append to wiki/log.md:
    ## [YYYY-MM-DD] query | Archived: <page title>
    

Lint

Quality checks on the wiki. Two categories with different authority levels.

Deterministic Checks (auto-fix)

Fix these automatically:

Index consistency — compare wiki/index.md against actual wiki/ files (excluding index.md and log.md):

  • File exists but missing from index → add entry with (no summary) placeholder. For Updated, use the article's metadata Updated date if present; otherwise fall back to file's last modified date.
  • Index entry points to nonexistent file → mark as [MISSING] in the index. Do not delete the entry; let the user decide.

Internal links — for every markdown link in wiki/ article files (body text and Sources metadata), excluding Raw field links (validated by Raw references below) and excluding index.md/log.md (handled above):

  • Target does not exist → search wiki/ for a file with the same name elsewhere.
    • Exactly one match → fix the path.
    • Zero or multiple matches → report to the user.

Raw references — every link in a Raw field must point to an existing raw/ file:

  • Target does not exist → search raw/ for a file with the same name elsewhere.
    • Exactly one match → fix the path.
    • Zero or multiple matches → report to the user.

See Also — within each topic directory:

  • Add obviously missing cross-references between related articles.
  • Remove links to deleted files.

Heuristic Checks (report only)

These rely on your judgment. Report findings without auto-fixing:

  • Factual contradictions across articles
  • Outdated claims superseded by newer sources
  • Missing conflict annotations where sources disagree
  • Orphan pages with no inbound links from other wiki articles
  • Missing cross-topic references
  • Concepts frequently mentioned but lacking a dedicated page
  • Archive pages whose cited source articles have been substantially updated since archival

Post-Lint

Append to wiki/log.md:

## [YYYY-MM-DD] lint | <N> issues found, <M> auto-fixed

Conventions

  • Standard markdown with relative links throughout.
  • wiki/ supports one level of topic subdirectories only. No deeper nesting.
  • Today's date for log entries, Collected dates, and Archived dates. Updated dates reflect when the article's knowledge content last changed. Published dates come from the source (use Unknown when unavailable).
  • Inside wiki/ files, all markdown links use paths relative to the current file. In conversation output, use project-root-relative paths (e.g., wiki/topic/article.md).
  • Ingest updates both wiki/index.md and wiki/log.md. Archive (from Query) updates both. Lint updates wiki/log.md (and wiki/index.md only when auto-fixing index entries). Plain queries do not write any files.

Dépôt GitHub

Astro-Han/karpathy-llm-wiki
Chemin: SKILL.md
0
agent-skillclaude-codecodexcursorkarpathyknowledge-base

Compétences associées

content-collections

Méta

Cette compétence propose une configuration éprouvée en production pour Content Collections, un outil axé sur TypeScript qui transforme des fichiers Markdown/MDX en collections de données typées de manière sûre avec une validation Zod. Utilisez-la lors de la création de blogs, de sites de documentation ou d'applications Vite + React riches en contenu pour garantir la sécurité de typage et la validation automatique du contenu. Elle couvre tout, de la configuration du plugin Vite et de la compilation MDX à l'optimisation des déploiements et la validation des schémas.

Voir la compétence

polymarket

Méta

Cette compétence permet aux développeurs de créer des applications avec la plateforme de marchés prédictifs Polymarket, incluant l'intégration d'API pour le trading et les données de marché. Elle fournit également une diffusion de données en temps réel via WebSocket pour surveiller les transactions en direct et l'activité du marché. Utilisez-la pour mettre en œuvre des stratégies de trading ou pour créer des outils traitant les mises à jour de marché en direct.

Voir la compétence

creating-opencode-plugins

Méta

Cette compétence aide les développeurs à créer des plugins OpenCode qui s'interconnectent avec plus de 25 types d'événements tels que les commandes, les fichiers et les opérations LSP. Elle fournit la structure du plugin, les spécifications de l'API événementielle et les modèles d'implémentation pour les modules JavaScript/TypeScript. Utilisez-la lorsque vous avez besoin d'intercepter, de surveiller ou d'étendre le cycle de vie de l'assistant IA OpenCode avec une logique personnalisée pilotée par les événements.

Voir la compétence

sglang

Méta

SGLang est un framework de service LLM haute performance spécialisé dans la génération rapide et structurée pour les workflows JSON, regex et agentiques grâce à son cache de préfixe RadixAttention. Il offre une inférence nettement plus rapide, particulièrement pour les tâches avec des préfixes répétés, ce qui le rend idéal pour les sorties complexes et structurées ainsi que les conversations multi-tours. Choisissez SGLang plutôt que des alternatives comme vLLM lorsque vous avez besoin d'un décodage contraint ou que vous construisez des applications avec un partage étendu de préfixes.

Voir la compétence