Grafo Concierge exposes 13 tools with basic schemas and descriptions, but falls short of production quality. Tool names lack consistent verb-noun structure (e.g., 'concierge_mine', 'concierge_wakeup' are vague; 'concierge_resume' is ambiguous). Descriptions are present but generic (10-50 chars for most tools), providing minimal context for LLM selection. Parameter descriptions exist but are sparse. No output schemas documented. Error handling is absent from visible code. The @_mcp_handler decorator suggests timing/error wrapping, but actual error guidance and recovery paths are not visible. Tools like 'concierge_wakeup' and 'concierge_resume' use domain jargon ('Consciousness reactivation', 'Context Compass') that obscures intent. No evidence of input validation, enum constraints, or pagination support despite tools like 'concierge_search' that likely return lists.

• Rename tools to follow verb_noun pattern: 'concierge_mine' → 'ingest_project', 'concierge_wakeup' → 'activate_project_context', 'concierge_resume' → 'get_project_summary'. This makes intent clear from the name alone.
• Expand descriptions to 50-200 chars with explicit WHAT/WHEN/RETURNS structure. E.g., 'Search project code and documentation using hybrid vector+keyword search. Returns ranked results with file paths and line numbers. Use when the user asks about code patterns or documentation.'
• Document output schemas for all tools. For 'concierge_search', specify: returns array of {file_path, line_number, snippet, relevance_score}. For 'concierge_list_facts', specify: returns array of {fact_id, fact_text, scope, created_at}.
• Add enum constraints to parameters. For 'concierge_store_fact', declare scope as enum: ['user', 'session', 'agent', 'org']. For 'concierge_search', add limit bounds (1-100, default 20) and document pagination support.
• Implement error guidance in tool responses. When a project_uuid is invalid, return: 'Project not found. Available projects: [list]. Call concierge_list_projects() to discover.' This enables LLM self-correction without extra calls.

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

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