Prompt file imported from nonlinear/literature (.github/prompts/literature.prompt.md). Copyright stays with the author.

Literature Research

Purpose: Search indexed books using RAG to answer questions based on book contents.

Instructions for AI

When this prompt is activated:

1. Check if user specified a folder:

   • Look for patterns like: "use oracles:", "from AI folder:", "search in urbanism:"
   • If folder specified, search ONLY in that folder
   • Check ~/Documents/literature/books/ for available folders

2. If NO folder specified, analyze the question to identify relevant topics:

   • Read all .rag-topics files in ~/Documents/literature/books/*/
   • Match question keywords against topics in each folder
   • Identify which folders contain relevant books

3. Consult the RAG system:

   • Use as ferramentas MCP para consultar os livros relevantes
   • Foque na pasta especificada OU nas pastas cujos tópicos correspondem à pergunta
   • Recupere trechos e citações relevantes

4. Provide answer with sources:

   • When mentioning an answer that has a source, add a reference in parentheses at the end of the sentence or term, like (1), (2), (3).

   • At the end, list sources as a numbered list, each with the file:// link and a 5-word string to search, like this:

📚 Sources:

1. file:///Users/nfrota/Documents/literature/books/folder/book.epub "five sequential words here"

2. file:///Users/nfrota/Documents/literature/books/folder/another_book.epub "another five word example"

3. If no relevant books found:

• Explain which topics you searched
• List available folders and their topics
• Suggest how to rephrase the question

Advanced MCP Arguments

You can use these arguments in your MCP JSON requests for more control:

• question: The main query string.
• book_context: (optional) Specify a book or folder for focused search.
• auto_detect: (optional, default true) Set to false to disable topic auto-detection.
• max_sources: (optional) Limit the number of sources returned (e.g., 5).
• return_snippets: (optional) If true, include text snippets from sources.
• return_metadata: (optional) If true, include extra metadata (author, year, etc.).

Example MCP JSON request:

{
  "method": "tools/call",
    "arguments": {
      "return_snippets": true,
      "return_metadata": true

---

## Available Topics

Current indexed folders and keywords:

- **AI:** ethics, UX, machine learning, algorithmic bias, software design
- **anthropocene:** ecology, forests, indigenous knowledge, climate
- **art direction:** generative design, creative coding, Processing
- **fiction:** sci-fi narratives (limited support)
- **oracles:** I Ching, hexagrams, divination
- **system theory:** complexity, emergence, gradients, simulation
- **urbanism:** legibility, state power, planning, high modernism
- **usability:** content strategy, validation, writing, UX

---
## Debug Commands

# View keywords for a folder
cat ~/Documents/literature/books/AI/.rag-topics

Tips

To ensure RAG activates:

• Mention topics explicitly (e.g., "according to I Ching")
• Ask conceptual questions rather than generic ones
• "What is legibility?" → urbanism/ Examples that might NOT trigger:
• "How are you today?" (social, not knowledge-based)