Open-Science exposes 17 tools across review, literature, and artifact domains. Most tools have descriptions (15/17), but quality is highly variable. Many descriptions are verbose and lack actionable structure for LLM selection. Input schemas are present but sparse: 8 tools have no visible input parameters, 9 have partial schemas. Parameter descriptions are inconsistent, some detailed, others missing entirely. No tool annotations (readOnlyHint/destructiveHint) despite clear risk classifications. Error handling and recovery guidance are absent. Output schemas are undocumented. This is a domain-specific tool suite with reasonable coverage but significant gaps in LLM-optimization and schema completeness.

• Add explicit input schemas to all 8 tools missing them (read_turn, submit_findings, list_pdf_elements, read_pdf_element, request_skill_import, and 3 others). Include parameter types, descriptions, and constraints.
• Annotate all tools with readOnlyHint (for READ_ONLY risk) and destructiveHint (for WRITE risk). This enables LLMs to reason about safety and retry logic without parsing descriptions.
• Refactor read_artifact description: split into separate tools (read_artifact_trace, read_artifact_content, read_artifact_preview) or compress to ~100 chars with clear WHEN-to-use guidance.
• Document output schemas for all tools. Specify field names, types, pagination (limit/offset/cursor), and result caps. Include chaining IDs (e.g., if search_library returns itemId, read_abstract must accept it).
• Add error handling and recovery guidance to WRITE tools (submit_findings, acquire_pdf, format_document, save, prepare_latex). Include validation rules, actionable error messages, and optional dry-run/confirmation steps.