Three tools with basic schemas and minimal descriptions. All tools have input schemas with type definitions, but descriptions are generic (10-30 chars) and lack actionable context. No parameter descriptions beyond the schema. No output schemas documented. No error handling guidance. No tool annotations (readOnlyHint/destructiveHint). Tools lack the specificity and LLM-optimization needed for reliable agent selection. Naming is reasonable (verb_noun pattern) but descriptions fail to explain WHEN to use each tool or what happens on failure.

• Expand tool descriptions to 100-200 characters. Example: 'route_to_agent: Route a message to the AI agent best suited for the task. Use when you need specialized handling (NLP, vision, code generation). Returns agent_id, confidence score, and reasoning. Fails if no agent matches capabilities.'
• Add descriptions to all parameters. Example for target_capabilities: 'Array of required agent capabilities (e.g., ["nlp", "vision"]). If empty, system selects best general-purpose agent. Valid values: nlp, vision, code, reasoning, planning.'
• Document output schemas in tool descriptions or as a separate schema field. Example: 'Returns {agent_id: string, confidence: 0-1, reasoning: string, estimated_latency_ms: number}'.
• Add tool annotations: mark share_context with destructiveHint=true (modifies shared state), route_to_agent and orchestrate_collaboration with readOnlyHint=true (no side effects).
• Add error handling guidance: 'If no agent matches, returns error with available_agents list. If timeout, retry with longer deadline. If invalid capability, returns error with valid_capabilities enum.'

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

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