Memory Auto Update

Problem

Key details in a conversation do not automatically become durable memory: decisions, action items, commitments, budgets, user preferences, and project milestones can be lost after clarification or complaint. memory-auto-update handles session-triggered memory capture: it does not monitor the background continuously; it recognizes the current context at explicit trigger points, drafts a summary, and writes the memory file only after user confirmation.

How It Works and Limits

The skill offers three update modes: active mode prompts a summary at the end of a session; passive mode waits for manual triggers; and hybrid mode flags higher-value items such as decisions, tasks, and commitments earlier, while ordinary content waits for later review. Recognition covers categories such as decisions, tasks, commitments, facts, preferences, and projects, with frequencies including real-time, every 30 minutes, every hour, manual, and session end. Scripts include extract_memory.py, generate_summary.py, write_memory.py, and user_settings.py for extraction, summarization, writing, and settings.

It is useful for engineers using personal assistants, multi-turn collaboration, or project tracking, especially when conversation outcomes should become reviewable text. Limitations: it only works at trigger points, does not read historical conversations, and extraction may miss items. Critical conclusions still require human review, and the memory template and related files should be backed up regularly.