Two tools with adequate descriptions but significant schema and parameter documentation gaps. Tool names follow verb_noun convention (overlay_*, preview_*). Descriptions are present (194 - 250 chars, within baseline range) but lack actionable detail on when to use each tool vs. the other. No output schema documented. No error handling guidance. No parameter constraints (enums, ranges, formats). Tools are READ_ONLY (good) but lack idempotentHint annotations.

• Add per-parameter descriptions in the FastMCP @mcp.tool decorator or via JSON Schema. E.g., 'context: User requirements and constraints (plain text, 10 - 5000 chars)' and 'output: Agent response to analyze (plain text, 10 - 10000 chars)'.
• Document the output schema explicitly. Add a docstring or schema annotation showing the structure of the returned dict: {user_intent_map: {...}, agent_attention_map: {...}, attention_gaps: [...], ...}.
• Clarify tool selection in descriptions. Rewrite to: 'overlay_attention: Analyze COMPLETED response alignment (post-hoc). Use after the agent has written output.' and 'preview_attention: Analyze PLANNED focus BEFORE writing (pre-hoc). Use before substantial responses.'
• Add tool annotations via FastMCP or MCP schema to expose readOnlyHint=true and idempotentHint=true, signaling to clients that these tools are safe to call repeatedly.
• Enhance error messages with recovery guidance. Instead of 'context must be a non-empty string', return 'context is empty. Provide user requirements and constraints as plain text (10 - 5000 chars).' This guides LLM retry.