AI Agent Hub
Back to skills
LLM Wiki Karpathy Knowledge Base Runtime icon

LLM Wiki Karpathy Knowledge Base Runtime

Knowledge Management Updated 2026.08.30

Paste the following prompt into your AI chat to install this skill:

Please follow https://skillhub.cn/install/skillhub.md to install @user_15292d5a/yjkj-llm-wiki-karpathy.

About this skill

Problem

Personal knowledge bases often suffer from scattered source material in raw/, weak note structure, unanswered questions that are never persisted, and drift in source IDs, manifests, and indexes. llm-wiki-karpathy targets Vaults managed by the LLM Wiki Karpathy runtime and turns reading, compiling, maintaining, searching, and persisting notes into a controlled workflow.

Core Capabilities and Workflow

The skill operates the runtime-managed Vault through kb_* tools. Its main paths include:
- Inspection and diagnosis: kb_status and kb_lint report counts, placeholder content, missing representations, and structural issues.
- Source note compilation: text assets use kb_prepare_source, kb_read_raw, and kb_upsert_source_note; PDFs and images use source bundles and representation artifacts before compiling grounded source notes.
- Search and persistence: kb_search and kb_read_notes retrieve evidence; answers can be archived in wiki/outputs/ or promoted into concept, entity, or synthesis pages.
- Continuous maintenance: kb_repair_source_ids runs as a dry run before apply, and kb_rebuild_indexes keeps collection indexes consistent.

Applicability and Caveats

It fits incremental maintenance, batch source-note backfill, AI-topic cleanup, and book-review batches with clear scope. Do not modify raw/ directly, and do not write into wiki/ or .llm-kb/representations/ with generic file tools. When scope is unclear, the skill favors small, high-confidence batches instead of expanding an underspecified request into a long-running rewrite.

Use Cases

  • Maintain an AI-topic Vault by linting placeholder titles and broken links, then repairing a small batch of AI source notes.
  • After collecting book reviews, compile the first 10 missing source notes in the book-review raw folder and rebuild indexes.
  • Answer a scoped query from the wiki, read supporting notes, then persist reusable conclusions as concept or synthesis pages.
  • When source IDs or manifest entries drift, dry-run repairs first, apply only if the plan looks correct, and rebuild indexes.

Best For

  • Researchers maintaining a personal AI knowledge base who want to organize scattered notes into concept, entity, and synthesis pages.
  • Editors curating book-review material who need to compile source notes in batches and keep indexes current.
  • Engineers using LLM Wiki Karpathy Vaults who need to lint issues and repair manifest or source-ID drift.
  • Product managers maintaining multi-source knowledge bases who want to write grounded answers back into output or derived pages.