Three read-only tools with clear, concise descriptions (60-80 chars each). All tools have documented input schemas with type and description fields. However, output schemas are not formally documented, only inferred from code. Parameter descriptions are adequate but lack constraint details (e.g., window format regex not stated in descriptions). No error handling guidance or recovery paths. Tool names follow verb_noun pattern but are generic ('personal_stats_tool' vs 'get_personal_stats'). No tool annotations (readOnlyHint, etc.). Composition is sound, each tool has one responsibility.

• Add formal output schema documentation to each tool. E.g., personal_stats_tool returns {prompts: int, tokens_saved: int, dollars_saved: float, wh_saved: float, top_5_compressions: [{original_tokens, achievable_tokens, achievable_pct, estimated_dollars_saved}], window: string, source: string}.
• Update 'window' parameter description to include constraint: 'Time window for stats aggregation. Format: <number><unit> where unit is d (days) or h (hours). Examples: 7d, 30d, 1h. Defaults to 7d.'
• Add error handling: if log file not found, return {prompts: 0, tokens_saved: 0, ...} with a note field: 'Log file not found at <path>. Run prompt-squeeze to generate data.' This guides LLM recovery.
• Rename tools to drop '_tool' suffix: personal_stats, team_stats, weekly_report. Cleaner and follows MCP conventions.
• Add readOnlyHint annotation to all three tools in FastMCP registration: @app.tool(readOnlyHint=True) to signal safety to LLMs.