bw-research

/bw-research — Deep Library Research

You are conducting deep research across the user's personal ebook library using Bookworm. This is a multi-pass process that synthesizes knowledge from multiple books and sections.

Arguments

The research topic: $ARGUMENTS

Research Process

Phase 1: Reconnaissance

1. Run mcp__bookworm__list_books to understand the available library.
2. Identify which books are most likely to cover the topic.

Phase 2: Multi-Query Search

Generate 3-5 different search queries that approach the topic from different angles:

• Direct query: The topic as stated
• Conceptual query: The underlying concept or principle
• Practical query: "how to" or implementation-focused angle
• Related concepts: Adjacent topics that might provide context
• Specific terms: Technical terms, function names, or patterns related to the topic

Run mcp__bookworm__search_library for each query with top_k: 6. Collect all unique results.

Phase 3: Deep Dive

For the top 3-5 most relevant sections found across all queries:

1. Use mcp__bookworm__list_sections to see the TOC of relevant books
2. Use mcp__bookworm__get_chapter to retrieve full chapter content
3. Extract key insights, patterns, code examples, and recommendations

Phase 4: Synthesis

Produce a structured research report:

## Research: [Topic]

### Summary
[2-3 sentence overview of what the library says about this topic]

### Key Findings

#### [Finding 1 — theme or concept]
[Synthesis of what multiple sources say]
— *[Book Title]*, [Chapter] (p.XX)
— *[Book Title]*, [Chapter] (p.XX)

#### [Finding 2 — theme or concept]
...

### Code Examples
[If applicable, include the most relevant code examples found]

### Recommendations
[Actionable takeaways based on what the sources recommend]

### Sources Consulted
- *[Book Title]* by [Author] — [Sections read]
- *[Book Title]* by [Author] — [Sections read]

Guidelines

• Cite sources: Always attribute findings to specific books and sections
• Synthesize, don't summarize: Compare what different authors say about the same concept. Note agreements and disagreements.
• Prioritize practical advice: Focus on actionable patterns and code examples over theory
• Be honest about coverage gaps: If the library doesn't cover an aspect of the topic, say so
• Parallel queries: Use the Agent tool to run multiple searches in parallel for speed when the queries are independent