Five tools with complete JSON schemas and descriptions, but quality is uneven. All tools have descriptions (10 - 200 chars, within baseline), and all expose input schemas with type definitions. However, descriptions lack LLM-optimized guidance on WHEN to use each tool and WHY. Parameter descriptions are minimal (e.g., 'Stable toolbox id' for toolbox_id). Error handling is present but generic ('capability_toolbox_unavailable'). No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Output schemas are documented in code but not exposed to the LLM. Risk levels are declared but not actionable in descriptions.

• Expand tool descriptions to 100 - 200 chars. Add WHEN/WHY guidance: 'Call warehouse_list first to discover available toolboxes. Then call toolbox_open to inspect a specific toolbox before activating capabilities.'
• Add tool annotations to input schemas: capability_activate and capability_release should include 'destructiveHint': true in their schema metadata to signal state modification.
• Document output schemas in tool descriptions. E.g., 'Returns: {entries: [{id, name, description}], total: number, next_offset: number|null}'.
• Enhance error messages with recovery guidance. Instead of 'capability_toolbox_unavailable', return 'Toolbox not found. Call warehouse_list() to see available toolboxes, or check the toolbox_id spelling.'
• Add parameter format hints. E.g., 'toolbox_id: Stable identifier (alphanumeric, 1-64 chars) shown in warehouse_list output.' and 'capability_ids: Array of leaf capability IDs from toolbox_open manifest.items[].id.'

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

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