llmwiki-init

name	llmwiki:init
description	Initialize the LLM Wiki. Creates directory structure, index, log, sets up qmd collection, and runs initial ingest of all existing content from raw/, docs/, and notes/. Run once per project.
user-invokable	true
allowed-tools	Read, Write, Edit, Bash, Glob, Grep, Agent, mcp__plugin_qmd_qmd__query, mcp__plugin_qmd_qmd__status, mcp__plugin_qmd_qmd__get, mcp__plugin_qmd_qmd__multi_get, TaskCreate, TaskUpdate, TaskList, TaskGet

LLM Wiki: Initialize

Set up a new LLM Wiki in the current project and run the first full ingest.

Pre-flight

1. Check if wiki/index.md already exists. If it does and has content beyond the template, ask the user whether to reinitialize (destructive) or skip.
2. Verify raw/, docs/, and notes/ directories exist. Create any that are missing.
3. Verify raw/attachments/ exists. Create if missing.

Step 1: Create Wiki Structure

Ensure the following directories and files exist:

wiki/
├── index.md          # Master index of all wiki pages
├── log.md            # Chronological log of operations
├── entities/         # Pages about specific things (people, companies, products, tools)
├── concepts/         # Conceptual/topic pages
├── sources/          # Source summaries (one per ingested source)
├── comparisons/      # Comparison tables, side-by-side analyses
└── synthesis/        # Cross-cutting analyses, insights

If files already exist, leave them as-is.

Step 2: Set Up qmd Collection

Run:

qmd collection add wiki ./wiki

This registers the wiki/ directory as a qmd collection named "wiki". If the collection already exists, skip.

Then update the index:

qmd update

If qmd is not installed, warn the user: npm install -g @tobilu/qmd and continue without it.

Step 3: Initial Ingest

Collect ALL existing files that should be ingested:

1. raw/ — all markdown files and other documents (skip raw/attachments/)
2. docs/ — all markdown files recursively
3. notes/ — all markdown files recursively

For each file found, invoke the llmwiki:ingest skill's logic (or invoke the skill itself). Process files in batches using parallel subagents where possible:

• Group files by directory/topic
• For each batch, spawn a subagent to read the files and produce wiki pages
• Each subagent should:
  • Read the source file
  • Determine what wiki pages to create/update (entity pages, concept pages, source summary)
  • Write the source summary to wiki/sources/
  • Create or update entity/concept pages in wiki/entities/ and wiki/concepts/
  • Add [[wikilinks]] between related pages
  • Return the list of pages created/updated

After all subagents complete:

• Rebuild wiki/index.md with all pages
• Append a log entry to wiki/log.md:

  ## [YYYY-MM-DD] init | Initialized wiki with N sources, M pages created
  

Step 4: Build Embeddings (Optional)

If qmd is available, run:

qmd embed

This builds vector embeddings for semantic search. If it fails or takes too long, skip — BM25 search still works.

Step 5: Report

Print a summary:

• Number of source files ingested
• Number of wiki pages created (by category)
• Any files that failed or were skipped
• qmd status (collection registered, embedding status)

Wiki Page Conventions

All wiki pages MUST follow these conventions:

Frontmatter

Every wiki page has YAML frontmatter:

---
title: Page Title
type: entity | concept | source | comparison | synthesis
tags: [tag1, tag2]
sources: [relative/path/to/source.md]
created: YYYY-MM-DD
updated: YYYY-MM-DD
---

Wikilinks

Use [[Page Title]] syntax for cross-references. Obsidian resolves these automatically. When creating a new page, add wikilinks to all related existing pages, and update those existing pages to link back.

Source Summaries

Source summary filenames: kebab-case version of the source title, in wiki/sources/. Example: wiki/sources/cross-platform-tech-stacks-2026.md.

Entity & Concept Pages

Use descriptive kebab-case filenames. Example: wiki/entities/expo.md, wiki/concepts/local-first-architecture.md.

File Naming

• All wiki page filenames: lowercase kebab-case with .md extension
• No spaces, no special characters beyond hyphens
• Keep names concise but descriptive