Hermes LLM Wiki
Paste the following prompt into your AI chat to install this skill:
Please follow https://skillhub.cn/install/skillhub.md to install @user_38654c53/hermes-llm-wiki.
About this skill
The Core Problem
Traditional RAG (Retrieval-Augmented Generation) rediscovers knowledge from scratch for every query. Hermes LLM Wiki solves this by adopting Karpathy's LLM Wiki pattern. It compiles knowledge once into interlinked Markdown files and keeps it current. Cross-references are established, contradictions are flagged, and synthesis reflects everything ingested, creating a compounding knowledge base.
Core Capabilities & Mechanics
The skill operates on a three-layer architecture:
- Raw Sources: Immutable original inputs (URLs, PDFs, pasted text).
- The Wiki: Agent-managed Markdown pages that are continuously updated and cross-referenced.
- The Schema: SCHEMA.md, which defines structure, conventions, and the tag taxonomy.
Key operations include:
- Ingest: Captures raw sources, discusses takeaways, checks existing pages to prevent duplicates, creates or updates entity pages that meet specific thresholds, enforces at least two [[wikilinks]] for cross-referencing, and updates index.md and log.md.
- Query: Locates relevant pages via index.md and full-text search, synthesizes answers grounded in the compiled knowledge with citations, and files high-value novel syntheses back into the Wiki.
- Lint: Automatically detects orphan pages, broken links, index gaps, invalid Frontmatter, stale content, factual contradictions, and oversized pages, outputting prioritized remediation suggestions.
Boundaries & Caveats
- Immutability: Never modify files in the
raw/directory; original sources must remain read-only. - Session Orientation: Before any operation in a new session, the agent must read
SCHEMA.md,index.md, and recentlog.mdentries to prevent duplicate pages and missed cross-references. - Page Thresholds: Do not create pages for passing mentions or footnotes; entities must meet the source mention criteria defined in
SCHEMA.md. - Structural Constraints: All pages require Frontmatter, tags must strictly belong to the defined taxonomy, and individual pages should be kept under 200 lines to maintain scannability.
Use Cases
- Ingest a batch of articles, papers, or web pages into a local wiki and update entity pages, links, index, and log.
- Query an existing knowledge base for technical trade-offs and have the agent cite specific pages in the answer.
- Run periodic lint checks to find broken links, orphan pages, stale entries, and tag issues with prioritized fixes.
- Maintain a Markdown knowledge base on a server while browsing the same vault in Obsidian for links and queries.
Best For
- Researchers maintaining personal or project wikis who want sources distilled into searchable, interlinked Markdown pages.
- Knowledge workers using Obsidian who want an agent to ingest, update, and audit wiki pages.
- Engineers running server-side knowledge pipelines who need headless sync and maintenance for one vault.
- Technical writers managing team documentation indexes who want consistent tags, page structure, and schema rules.
Related Skills
Use a book title and author to search reviews, contents, and reader feedback, then generate a structured breakdown of historical or decision-logic books with key claims, concepts, quotes, and reading notes in Obsidian.
Provides Bazi, Liu Yao, Qimen, Meihua, and date-selection guidance from classical references, with plain-language explanations and actionable suggestions.
A fast CLI for summarizing URLs, local files, and YouTube links with multi-model support, length control, and JSON output.
Automatically classifies ideas and saves them as tagged, structured Markdown notes in a desktop inspiration library.