Rocky Todo List Buddy

Problem Solved

Daily todos often scatter across chat context, notes, and task state: an AI may remember the last turn but cannot reliably persist state; hand-written Markdown tables may drop columns or reorder rows; deferred tasks may remain only verbal; morning planning and review can depend on long conversational memory.

How It Works

Rocky Todo List Buddy separates semantic inference from durable data operations. The AI layer extracts title, area, priority, and schedule_date from natural language, while scripts handle SQLite writes, date calculations, state machines, and Markdown rendering. Key paths include:

• Inbox capture: after the user says “record this,” the AI extracts fields and calls write_task.py, preserving the raw input for audit, then stops for confirmation.
• Planning and inventory: query.py scans the full task set, while build_plan.py / render_table.py emit standard tables that the AI pastes verbatim instead of reconstructing them.
• Morning planning: review_state.py advances step by step, first handling OVERDUE_ACTIVE and BACKLOG, then presenting candidates, waiting for user confirmation, batching writes, and finalizing the plan.
• Review: daily and weekly review are state-controlled, with build_review_stats.py outputting stats and TASK_SEQ_MAP retained for later #N operations.

Boundaries and Caveats

• The single source of truth is ~/.rocky-todo-list-buddy/rocky.db; verbal deferral is not enough—batch_update.py must be called.
• Stop-and-wait points require user confirmation before activation or cancellation.
• Work/life partitioning follows tasks.area, not title semantics.
• The Rocky voice is controlled by rocky_voice and can be switched to standard mode.
• Drills and debugging should use test- prefixed keys to avoid polluting production review state.