TrainBud has 15 well-named tools with consistent verb_noun patterns (get_*, remember_*, log_*). All tools have descriptions (avg 65 chars, within baseline). Input schemas are present and typed for all tools. However, parameter descriptions are sparse or missing context. Output schemas are undocumented, LLMs cannot predict response structure. Error handling exists (toolErrors.ts) but lacks recovery guidance. No tool annotations (readOnlyHint/destructiveHint). Two write tools (remember_context, log_subjective) lack confirmation patterns. Overall: solid naming and basic structure, but incomplete parameter docs and missing output schemas limit LLM reasoning.

• Document output schemas for all 15 tools. Specify field names, types, and whether results are paginated. Example: get_activities_range returns {activities: [{id, name, date, duration, calories}], total_count, next_cursor}.
• Add min/max constraints to numeric parameters. E.g., 'days: number (1 - 365, default 7)', 'limit: number (1 - 100, default 20)'.
• Implement dry-run or confirmation for write tools. Add a 'confirm' boolean parameter to remember_context and log_subjective, or return a confirmation_token requiring a second call.
• Add tool annotations to the MCP tool registration. Mark all get_* tools with readOnlyHint: true; mark remember_context and log_subjective with destructiveHint: true.
• Enhance error messages with recovery guidance. E.g., 'Authentication failed. Run "trainbud auth" to re-authenticate.' (already present), extend this pattern to all error paths with actionable next steps.

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

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