
from llm-wiki-skill941
Build and maintain a continuously growing personal knowledge wiki from links, files, and text; supports multi-source extraction and automated page generation.
LLM-Wiki turns fragmented content (webpages, social posts, PDFs, local files) into a structured, interlinked personal wiki. The skill orchestrates ingestion, entity/topic extraction, and page generation so knowledge accumulates over time instead of being re-derived each query.
Trigger when the user explicitly asks about creating or managing a knowledge base (keywords: "knowledge base", "wiki", "llm-wiki") or requests actions like init, ingest, batch-ingest, query, lint, status, graph, or crystallize. Do not trigger for single-article summarization unless the user requests adding it to a knowledge base.
Best used with agents that can run shell scripts and call local tools (CLI-capable agents). Works well with Claude/Codex/Copilot-style agents or any runtime that can invoke the repo scripts and manage local files.
llm-wiki is a comprehensive personal knowledge wiki builder based on the Karpathy llm-wiki methodology. It supports multi-source ingestion (URLs, PDFs, local files, plain text), automated wiki page generation, knowledge graph visualization, and health checking. The SKILL.md is extremely thorough (27K+ chars, 10 workflows). Scripts are well-structured with proper error handling and atomic writes. Most script failures in testing were due to path isolation (shared-config.sh sourced via relative paths) or missing wiki context — expected and not indicative of real bugs. The skill uniquely targets Chinese-English bilingual users.
jqnodeperlpython3No security concerns. The curl|sh pattern in print_install_hint is informational only (stderr hint to user, not auto-executed). Well-engineered skill with clear separation of concerns, proper validation, and good defensive scripting.