Scribble has 14 well-named tools with clear action verbs (capture, search, create, complete, update, list, reflect, sync). All tools have descriptions (avg ~120 chars, within baseline 34 - 392). Input schemas are present and use Zod with type definitions. However, output schemas are NOT documented, responses are JSON-stringified text blobs with no declared structure. Parameter descriptions are present but often generic (e.g., 'Date (YYYY-MM-DD)' lacks context on why or when to use). Error handling is minimal, no recovery guidance, no categorization of retryable vs fatal errors. Security: no visible secret injection, no permission gates, no audit logging. Composition is strong: tools chain well (triage_inbox → create_task), and naming is consistent (verb_noun). Overall: solid definitions with gaps in output documentation and error guidance.

• Document output schemas for all tools. Define the structure of each response (e.g., capture returns {eventId: string, message: string}; list_tasks returns {tasks: [{id, title, due, status}], total: number}).
• Add error handling with recovery guidance. E.g., 'Task not found. Call list_tasks() to find valid task IDs.' Categorize errors as retryable (network timeout) vs user-fixable (invalid task ID).
• Implement permission checks and audit logging for destructive tools. Log who called what, when, and with what parameters. Gate complete_task and update_task behind authorization.
• Enrich parameter descriptions with constraints and examples. E.g., 'energyLevel (1 - 5): 1=minimal focus, 5=peak; favors smaller tasks when low. Default: 3.' instead of 'Energy 1-5'.
• Clarify get_intervention_ladder. Add parameters (e.g., date, scope) or explain what signals it reads. Document the L1 - L4 scale (e.g., L1=gentle reminder, L4=escalation).