skill_skill

Overview

This skill lets you search a Zotero library using safe code execution to run multi-strategy searches, automatically deduplicate results, and return only the top matches to LLM context. It minimizes crash risk by keeping large intermediate datasets inside the execution environment and exposing a single, ranked result set. Use it to get comprehensive, reliable search results with minimal manual work.

How this skill works

The skill runs multiple search strategies (semantic, keyword, tag-based) in a code execution environment, fetching large candidate sets per strategy. It combines and deduplicates items in code, applies ranking and optional filters, and returns a small, curated list (e.g., top N) to the LLM. All heavy processing happens off-context so you avoid loading hundreds of items into the model.

When to use it

• When you need thorough Zotero searches that combine semantic and keyword approaches.
• When direct MCP tool calls risk overloading LLM context or causing crashes.
• When you want automatic deduplication and ranking across multiple strategies.
• When you need to apply complex or custom filters in Python before returning results.
• When retrieving recent items, author-specific lists, or tag-focused collections.

Best practices

• Use comprehensive_search(query, max_results=N) for most queries to get multi-angle results with one call.
• Fetch broadly (search_limit_per_strategy 50–100) and filter in code; avoid trying to limit context by constraining fetches too much.
• Wrap calls in try/except to surface errors and suggest broader search terms when no results are found.
• Prefer format_results(...) to produce concise, readable output with optional abstracts.
• Tune search_limit_per_strategy and max_results based on speed vs thoroughness tradeoffs.

Example use cases

• Find top 20 papers on 'embodied cognition' using semantic + keyword + tag searches combined and ranked.
• List recent journal articles on 'machine learning' published since 2020 by fetching broadly then filtering by type and date.
• Retrieve all papers by a given author and sort them by date for a literature review.
• Show items tagged with both 'learning' and 'cognition' and present the most relevant subset.
• Run multi-topic intersection searches (e.g., 'cognition' AND 'learning') and return overlapping results.