Back to Skills
    🦞

    claw-progressive-memory

    Meta-skill for implementing

    By @autogame-17
    View on GitHub
    SKILL.md
    ---
    name: progressive-memory
    description: Meta-skill for implementing the Progressive Memory pattern. Offloads heavy details to indexed sub-files to keep the main context window light.
    ---
    
    # Progressive Memory Skill
    
    A meta-skill that formalizes the "Progressive Memory" pattern for AI agents.
    
    **The Problem**: `MEMORY.md` grows indefinitely, consuming context tokens and confusing the agent with too much detail.
    **The Solution**: Keep `MEMORY.md` as a lightweight **Index**, and offload heavy details (logs, lists, configs) to `memory/topic.md`. The Agent only reads the sub-file when the Index points to it being relevant.
    
    ## Tools
    
    ### memorize
    Save content to a sub-file and automatically index it in `MEMORY.md`.
    
    ```bash
    node skills/progressive-memory/index.js memorize "Visual Identity" "White hair, red eyes..."
    ```
    
    - Creates/Updates `memory/visual_identity.md`.
    - Adds `- **Visual Identity**: See memory/visual_identity.md` to `MEMORY.md`.
    
    ### recall
    Read a specific memory topic file.
    
    ```bash
    node skills/progressive-memory/index.js recall "Visual Identity"
    ```
    
    ## Best Practices
    
    1.  **Index First**: Before adding 50 lines to `MEMORY.md`, ask "Do I need this *every* turn?". If no, use Progressive Memory.
    2.  **Naming**: Use clear, searchable topic names (e.g., `suno_presets`, `project_alpha_specs`).
    3.  **Context**: The Index entry in `MEMORY.md` should have a short description so the Agent knows *why* to recall it.