LLM Wiki Skill: Build a Personal Knowledge Base with InfraNodus Gap Analysis

The LLM Wiki skill is a guided, multi-phase Claude Skill that scaffolds a persistent, compounding knowledge base from your raw sources — with InfraNodus wired into every stage that benefits from network analysis: ontology generation, gap detection, research planning, and GraphRAG retrieval.

What InfraNodus specifically does inside the wiki:

• Builds the knowledge graph — every wiki folder (concepts/, sources/, questions/, systems/…) gets a flat ontology in infranodus/ using [[wikilinks]] syntax that you can paste into infranodus.com to visualize the structure of your thinking as a network.
• Finds the gaps in your thinking — generate_content_gaps identifies the under-connected concept clusters in your wiki, so you can see what you haven't read yet, not just what you have.
• Turns gaps into research priorities — generate_research_questions and generate_research_ideas convert structural gaps into a prioritized, actionable todo list in todos/.
• Powers GraphRAG retrieval — retrieve_from_knowledge_base queries your wiki with graph-aware retrieval grounded in your materials, not the LLM's training data.
• Detects bias, focus, and dispersion — optimize_text_structure diagnoses the cognitive state of any wiki page and recommends whether to broaden, focus, or bridge.

Instead of re-deriving knowledge from raw documents on every query (the classic RAG pattern), the LLM extracts, cross-references, and synthesizes knowledge once into markdown pages — then InfraNodus keeps the graph view current so you always know where the holes are. The wiki compounds over time; you curate sources and ask questions; the LLM does the bookkeeping; InfraNodus does the structural diagnostics.

When to Use This Skill

The skill activates whenever you ask Claude (or another LLM client with skills enabled) to:

• set up a personal knowledge base or research wiki
• organize notes, papers, transcripts, or articles with LLM help
• build a "second brain" or Obsidian + LLM workflow
• create a persistent knowledge graph from a folder of documents
• plan ongoing research priorities and uncover gaps in what you've read
• turn raw PDFs, YouTube transcripts, or web articles into a queryable, linked knowledge structure

It asks a few targeted questions about your domain and goals, then scaffolds the entire wiki structure, schema, and workflows tailored to your needs.

How the Wiki Is Structured

Every LLM Wiki the skill builds has five layers, with InfraNodus owning the structural / network layer:

• Raw sources — immutable input documents (PDFs converted to markdown, transcripts, articles, notes). The LLM reads but never modifies these. Organized by source type (papers, notes, youtube, articles, patents, books, interviews…), not by topic.
• The wiki — LLM-generated markdown pages: summaries, entity pages, concept pages, comparisons, synthesis. The LLM owns this layer entirely.
• The infranodus/ folder — flat folder of [[wikilinks]] ontology files, one per wiki section (concepts-ontology.md, sources-ontology.md, full-wiki-ontology.md…). Generated and incrementally updated by the ontology-creator skill and consumed by the InfraNodus MCP server for graph visualization, gap analysis, and GraphRAG. This folder is the bridge between markdown and network view.
• Output folder — results of your interactions: reports, analyses, gap-analysis exports, slide decks.
• Schema documents (CLAUDE.md and AGENTS.md) — configuration that tells the LLM how the wiki is structured, what InfraNodus tools to call at which stage, and what conventions to follow. Co-evolved by user and LLM over time.

The 10-Phase Workflow

The skill walks through ten distinct phases. Phases 8 and 9 are deliberately separated — acquisition (getting files onto disk) and processing (converting raw materials into wiki pages) are independent, re-runnable operations.

Phases 9 and 10 are where InfraNodus does most of its work: keeping the ontology current as new sources land, then surfacing the missing connections so you know what to read next.

Tiered to Your Project Size

The skill classifies each wiki into a tier based on source volume, entity count, and timeframe — so a one-weekend reading project doesn't get the same indexing infrastructure as a multi-year research program.

What InfraNodus Brings to the Wiki

Markdown alone gives you content. The InfraNodus MCP server gives you structure — the network view of how your concepts actually connect. The LLM Wiki skill calls InfraNodus at every point where structure matters.

1. Append-Only Ontology in [[wikilinks]] Format

After each Phase 9 ingest, the ontology-creator skill reads the updated wiki pages and adds new [[entity1]] [relation] [[entity2]] statements to the corresponding ontology file in infranodus/. Existing relationships are never overwritten — ontology files are curated artifacts that accumulate human-reviewed knowledge. You can paste any ontology file directly into infranodus.com to see the network view.

2. Gap Analysis Drives Phase 10 Research Planning

This is the part that makes the wiki self-directing. After each ingest, Phase 10 calls:

• generate_topical_clusters — identifies the main topical clusters in the current wiki state
• generate_content_gaps — finds the under-connected pairs of clusters: this cluster and that cluster aren't yet bridged
• generate_research_questions / generate_research_ideas — converts each gap into a concrete question or research direction you can act on
• develop_conceptual_bridges — for the gaps worth bridging, suggests how to connect them

The output lands in todos/ as a prioritized list. The wiki tells you what to read next.

3. GraphRAG Retrieval over Your Own Materials

retrieve_from_knowledge_base queries the ontology + wiki together — you get answers grounded in your own sources with the network context (which clusters, which connections, which gaps) instead of a flat top-k chunk dump. generate_responses_from_graph goes further and answers questions using the structure of the graph itself as context.

4. Structural Diagnostics on Any Page or Cluster

optimize_text_structure reads any wiki page (or the whole wiki) and tells you whether it's biased (tunnel vision on one cluster), focused (well-balanced), diversified (spread across multiple clusters), or dispersed (scattered with no through-line) — then suggests the right development move. develop_latent_topics surfaces topics that are implicit in the text but not yet explicit pages.

5. Real-Time External Data for SEO & Search Wikis

For wikis that track a competitive or content-marketing topic, analyze_google_search_results, analyze_related_search_queries, and generate_seo_report pull live SERP data so the wiki's gap analysis includes what the world is searching, not just what you've already collected.

6. Persistent Memory via Knowledge Graphs

memory_add_relations persists key [[wikilinks]] relationships into a named InfraNodus graph that survives across sessions, so future conversations can call memory_get_relations to pick up where you left off — without re-reading the entire wiki.

7. Optional /actionize Integration

Pipe the Phase 10 todo list into the actionize skill to schedule deadlines and Telegram reminders for the research questions InfraNodus surfaced.

Why LLM Wiki + InfraNodus Beats Plain RAG

Most "chat with your documents" workflows re-derive insight from raw chunks on every query. They don't remember what they figured out yesterday, and they have no way to tell you what's missing — only what's there. The LLM Wiki + InfraNodus combination is different:

• Synthesis is persisted, not recomputed. Pages capture insight as durable markdown.
• The graph is persisted too. The infranodus/ ontology folder is the structural counterpart of the markdown wiki — one for reading, one for navigating.
• Cross-references compound. New sources extend existing pages and append new [[wikilinks]] relations rather than starting fresh.
• You can read your wiki without an LLM. Plain markdown, Obsidian-compatible, git-versioned. You can also visualize it without an LLM by pasting the ontology into infranodus.com.
• Gaps are visible, not hidden. InfraNodus tells you which concept clusters aren't yet bridged — the negative space of your reading.
• The wiki tells you what to read next. Phase 10's gap-driven research questions turn the wiki into an active research partner instead of a passive archive.
• The wiki replaces chat history. Insights from a conversation get written into pages, ontology relations, and named InfraNodus memory graphs — so they survive past the context window.

Install the Skill

Download the skill bundle from the InfraNodus skills releases page and install it like any other Claude Skill. See the main installation guide for step-by-step instructions per LLM client (Claude Web / Desktop, Claude Code, ChatGPT, OpenClaw, Cursor).

Claude Code

cd ~/.claude/skills
curl -L -o skill-llm-wiki.zip https://github.com/infranodus/skills/releases/latest/download/skill-llm-wiki.zip
unzip skill-llm-wiki.zip

Claude Web / Desktop

Go to Settings > Capabilities (or Menu > Customize in the newer UI), enable Code execution and file creation, scroll to the Skills section, and upload the skill-llm-wiki.zip file.

OpenClaw

install this skill: https://github.com/infranodus/skills/releases/latest/download/skill-llm-wiki.zip

How to Invoke It

Once installed, the skill activates automatically on phrases like "set up an LLM wiki", "build me a second brain", "I have a folder of papers I want to organize", "help me plan my research", or "create a persistent knowledge graph from these documents."

You can also invoke it explicitly:

Use the llm-wiki skill to set up a research wiki for my reading on
political philosophy. I have ~30 PDFs in ~/Dropbox/PoliPhil and I want
to track contradictions between authors and produce an essay at the end.

The skill responds with the discovery questions, scopes the tier (Light / Medium / Heavy), proposes a directory layout (including the infranodus/ folder), writes the schema files with InfraNodus tool wiring baked in, and scaffolds the wiki on disk.

Then, on every new source you add:

Process the new sources in raw/papers/ and update the wiki.

The skill ingests each file, updates the relevant wiki pages, calls the ontology-creator skill to append new [[wikilinks]] relations, and refreshes the infranodus/ folder.

When you want to know what to read next:

Run Phase 10. What are the biggest gaps in the wiki right now and
what should I read next?

The skill calls generate_topical_clusters → generate_content_gaps → generate_research_questions on the current InfraNodus graph and writes a prioritized todo list to todos/. You can then visualize the same graph at infranodus.com to see the gaps as visual negative space between clusters.

When you want to query the wiki:

Use retrieve_from_knowledge_base to answer: how do the authors I've
read so far disagree about the role of consent in legitimacy?

GraphRAG retrieval over the ontology + wiki returns answers grounded in your own materials, with cluster context.

What's Next

• All Skills & Use Cases
• Improve Skills with the InfraNodus MCP
• Cognitive Variability Skill Use Case
• Installation Instructions
• LLM Wiki Skill on GitHub