Zotero Skill has 10 well-named, read-only search tools with consistent parameter patterns. All tools have descriptions (avg 95 chars) and input schemas with types. However, schemas lack depth: no enums for constrained fields (e.g., 'type' filter), no documented output schemas, and parameter descriptions are minimal (avg 35 chars). Error handling is absent from visible code. No tool annotations (readOnlyHint, etc.). Naming is strong (verb_noun pattern: search, list, browse, get), but parameter documentation falls short of production baseline (94% of A+ tools have full param descriptions; this server ~40%).

• Document output schema for each tool. Example for zotero_search: {items: [{itemID, itemKey, typeName, title, authors, date, doi, url, publication, tags, collections, notes, hasFulltext, hasNotes, zoteroURI}], total: int, limit: int}.
• Add enum constraints for 'type' parameter. Inspect Zotero schema and list valid itemTypes (e.g., 'journal', 'book', 'thesis', 'webpage'). Replace 'Filter by item type (optional)' with 'Item type filter (optional). Valid: journal, book, thesis, webpage, report, conference, etc.'
• Remove 'db' parameter from tool interface. Move to server-side config (environment variable or config file). This eliminates path traversal risk and simplifies the LLM interface.
• Add error recovery guidance. When database open fails, return: 'Database not found at [path]. Ensure Zotero is installed and the database path is correct. Check server logs for details.' When query returns 0 results, suggest: 'No results found. Try a broader search term or use zotero_tag_list to explore available tags.'
• Add readOnlyHint: true to all tool schemas. This signals to agents that these tools are safe to call without side effects and can be retried freely.