Five tools with complete JSON schemas and descriptions, but significant gaps in parameter documentation and error guidance. invoke_agent and context_store have clear enums and required fields, but context_query, context_summarize, and context_stats lack actionable descriptions for their parameters. No output schemas documented. Error handling is absent, tools do not guide recovery or categorize failures. Naming is verb-forward (invoke_, context_*) but descriptions are terse (10-50 chars), below the 50-200 char baseline for LLM-optimized tools. No tool annotations (readOnlyHint, destructiveHint, idempotentHint). Composition is reasonable, each tool has one responsibility, but parameter relationships (e.g., correlationId required only for conversation scope) are underdocumented.

• Expand tool descriptions to 50-200 characters. Example: 'invoke_agent: Invoke another agent (developer, architect, teamlead, qa, productowner) to perform a task. Use this to delegate work across the agent network. Returns the agent's response or error if wait=true, or a job ID if wait=false.'
• Document output schemas for all tools. Example: 'invoke_agent returns {status: string, response?: object, jobId?: string, error?: string}'. This enables LLMs to chain tools and extract needed fields.
• Add parameter descriptions for context_query: 'mode: Query strategy, keys (lookup specific keys), search (full-text search), get-all (retrieve all items in scope). maxTokens: Token budget for results; search results are truncated if they exceed this limit.'
• Add error handling guidance. Example: 'If correlationId is invalid, call context_query with mode=get-all to discover valid IDs. If TTL expires, the item is automatically removed, re-store it if needed.'
• Add tool annotations: mark context_query, context_stats as readOnlyHint=true; mark context_store, invoke_agent as destructiveHint=true (they modify state). This helps clients optimize caching and retry logic.