Server has 5 tools with basic schemas and descriptions, but significant gaps prevent production use. Tool names follow verb_noun convention (list_, who_, get_, create_). Descriptions exist but are generic and lack actionable context for LLM selection. Input schemas present but incomplete: who_knows has enum constraint (good), but create_course lacks critical parameter descriptions and validation guidance. No output schemas documented. Error handling absent, no recovery guidance or categorization. Parameters lack format constraints and dependency documentation. create_course is a mega-tool combining multiple concerns (schema proposal, generation, review, validation, enrollment, media), violates single-responsibility principle.

• Split create_course into 5 focused tools: propose_course_schema (input: title, document_id, intent_density → output: schema_id, nodes), generate_course_content (input: schema_id → output: per-node status, retry counts), validate_course (input: schema_id → output: validation_passed, errors), enroll_user_in_course (input: course_id, user_id → output: enrollment_id), generate_course_media (input: course_id, node_ids, media_types → output: artifact_ids). Each tool does one thing; agent composes them.
• Document output schemas for all tools. Example: list_skills returns {skills: [{id, name, category, proficiency_distribution}], total_count}. who_knows returns {employees: [{user_id, name, proficiency_level, department}], total_count}. Include these in tool definitions so LLMs know what fields to expect.
• Add parameter descriptions and constraints to create_course: intent_density='Course depth (1=light overview, 5=comprehensive mastery). Default 3.' generate_artifacts='Optional array of media types: podcast, infographic, video, quiz. Each generates for first 2 nodes.'
• Enhance tool descriptions with WHEN/WHY context. Example: 'Call list_skills first to see available skills before searching for employees. Use get_gap to identify understaffed skills, then who_knows to find candidates for training.'
• Add error handling guidance to all tools. Example: 'If skill not found, call list_skills to see available skills. If user not found, call search_users() with partial name. If course validation fails, review errors and call propose_course_schema again with adjusted intent_density.'