maria/setup-llm-wiki
Bootstrap a new LLM-maintained wiki at a chosen folder, following the llm-wiki.md pattern (a three-layer memex - raw sources, LLM-generated wiki pages, and a CLAUDE.md or AGENTS.md schema that tells the LLM how to ingest and maintain the wiki). Creates the directory layout, writes a tailored schema file, plus index.md and log.md with a bootstrap entry. Use this skill when the user asks to "set up an llm-wiki", "create an LLM wiki", "bootstrap a wiki", "instantiate the llm-wiki pattern", or invokes /setup-llm-wiki. The skill asks the user about target folder, domain (research deep-dive / personalised work wiki / personal knowledge base / business-team wiki / reading a book / combination), source types (web articles, academic PDFs, meeting/podcast transcripts, own notes), image handling, optional search tooling (qmd), schema filename (CLAUDE.md or AGENTS.md), and optional symlinks to sibling folders if the target sits inside an Obsidian vault.
87
90%
Does it follow best practices?
Impact
83%
2.24xAverage score across 4 eval scenarios
Low
Low-risk findings worth noting
Low
Low-risk findings.
1 low severity finding. Worth noting, but not necessarily harmful.
W011: Third-party content exposure detected (indirect prompt injection risk).
What this means
The skill exposes the agent to untrusted, user-generated content from public third-party sources, creating a risk of indirect prompt injection. This includes browsing arbitrary URLs, reading social media posts or forum comments, and analyzing content from unknown websites.
Why it was flagged
The skill explicitly instructs the agent to pull user-generated content from external systems and public web sources — e.g., the MCP integrations section (assets/fragments/mcp-integrations.md) and the "Source types" handling in SKILL.md describe fetching from Granola, Slack, Notion, Gmail/Drive and web articles into raw/ and then ingesting them, so untrusted third-party content can be read and influence ingestion/agent behavior.
Where we found it
Granola
domain · 2 sites
The skill's generated schema instructs the agent to pull meeting transcripts from Granola via MCP integration (mcp__claude_ai_Granola__*), saving content to raw/ for ingestion.
SKILL.md
89
meeting/call/podcast transcripts (e.g. from Granola)
assets/fragments/mcp-integrations.md
5
**Granola** — meeting transcripts and notes (`mcp__claude_ai_Granola__*`)
Slack
domain · 2 sites
The skill's generated schema instructs the agent to pull channel/thread messages from Slack via MCP integration (mcp__claude_ai_Slack__*), saving content to raw/ for ingestion.
SKILL.md
47
fed by Slack threads, meeting/call transcripts, project documents, and own notes
assets/fragments/mcp-integrations.md
7
**Slack** — channel/thread messages (`mcp__claude_ai_Slack__*`)
Notion
domain · 1 site
The skill's generated schema instructs the agent to pull pages and databases from Notion via MCP integration (mcp__claude_ai_Notion__*), saving content to raw/ for ingestion.
assets/fragments/mcp-integrations.md
8
**Notion** — pages, databases (`mcp__claude_ai_Notion__*`)
Gmail / Calendar / Drive
domain · 1 site
The skill's generated schema instructs the agent to pull content from Gmail, Calendar, and Drive via MCP integration when authenticated, saving content to raw/ for ingestion.
assets/fragments/mcp-integrations.md
9
**Gmail / Calendar / Drive** — when authenticated
Linear
domain · 1 site
The skill's generated schema instructs the agent to pull issues, projects, and documents from Linear via MCP integration (mcp__claude_ai_Linear__*), saving content to raw/ for ingestion.
assets/fragments/mcp-integrations.md
6
**Linear** — issues, projects, documents (`mcp__claude_ai_Linear__*`)
- Audited
- Security analysis