Agent skill · volcengine
llm-wiki
Compile heterogeneous knowledge sources—including documents, notes, web content, transcripts, research materials, and code repositories—into a Karpathy-style, evidence-grounded LLM Wiki with a maintained index; default entity and concept pages; and selective method, comparison, analysis, or instruction-requested summary pages. Use with ov compile to create or incrementally refresh knowledge that is easy for people and agents to retrieve, navigate, and reuse.
What it needs
About 9k tokens when loaded.
What this skill does
LLM Wiki Objective Turn the supplied sources into durable, connected knowledge rather than a collection of source summaries. Give every page one clear retrieval purpose, a direct opening summary, consistent terminology, explicit relationships, and evidence close to the claims it supports. Follow the core LLM Wiki pattern: keep raw sources immutable, compile their knowledge into persistent Markdown pages, integrate new evidence into existing knowledge, maintain cross-references and contradictions, and keep index.md as the navigation entry point. OpenViking Compile owns writes, derived semantic sidecars, and task history, so do not generate .overview.md, .abstract.md, AGENTS.md, CLAUDE.md, or a duplicate operation log. Keep sources read-only. Follow explicit instructions in the task instruction for scope, audience, language, and depth. Otherwise use the dominant language of the sources and write for a knowledgeable newcomer to the domain. Knowledge model Use the smallest page type that matches the page's primary retrieval purpose: Page type Use for Examples --- --- --- entity A named thing with a stable identity or boundary person, organization, product, project, system, service, module, dataset, standard, named event concept A reusable idea, mechanism, policy, pattern, protocol, or mental model that explains what or why governance rule, architecture pattern, domain theory method A reusable procedure that explains how and has prerequisites, ordered steps or branches, and a verifiable outcome operating procedure, deployment guide, debugging playbook, research method comparison Two or more subjects evaluated side by side on explicit, evidence-supported dimensions product comparison, design tradeoff, version comparison analysis A cross-source conclusion tied to a clear question, scope, assumptions, and uncertainty due-diligence finding, trend analysis, system-wide assessment summary A durable digest that preserves one source's own claims, perspective, and limits paper su …
How to use it
Reference it in AdaL, Claude Code, Cursor or any coding agent — nothing to install:
@skills volcengine/llm-wiki