Critical failures across naming, descriptions, and schemas. Tool names lack action verbs (e.g., 'career_recommendation_engine', 'study_plan_builder', 'resume_review_service' are noun-heavy, not verb_noun). Descriptions are present but generic (avg ~60 chars, well below 194-char baseline). Input schemas visible for only 6/12 tools; most lack parameter type definitions. No output schemas documented. Tools like 'calendar.list_events' and 'time_management_skill' have empty input objects with no description of what they return. No error handling guidance. No tool annotations (readOnlyHint, destructiveHint). Source code shows agent implementations but NOT explicit MCP tool registration with full schemas.

• Rename tools to verb_noun pattern: 'recommend_career_path', 'build_study_plan', 'review_resume', 'search_internships', 'prepare_interview', 'manage_time'. This signals intent to LLMs before they read descriptions.
• Add output schemas to every tool. Document return type (object, array, string), required fields, and field types. E.g., 'Returns: {opportunities: [{title, company, deadline, fit_score}], total_count}' with ISO 8601 dates, not timestamps.
• Expand descriptions to 100-200 chars. Include WHAT (what does it do?), WHEN (when should the LLM call it?), and WHAT IT RETURNS (what fields?). E.g., 'Searches for internship opportunities matching user interests. Returns a list of roles with deadlines and fit scores. Call this before build_study_plan to align timeline.'
• Add parameter descriptions for all inputs. For 'goals' object params, document the expected structure: '{subject: string, exam_date: ISO8601, current_level: enum[beginner|intermediate|advanced]}'.
• Implement error handling with recovery guidance. E.g., 'If no internships found, try broader interests or call search_jobs instead. If API timeout, retry after 5 seconds.' Use pattern:recovery-guide.

Inferred effective spec: <=2025-11-25.

• Relies on Logging (deprecated) - log to stderr or use OpenTelemetry