Progressive Memory

What This Skill Does

Progressive Memory is a high-efficiency architecture for managing AI context windows by implementing a hierarchical 'index-first' retrieval system. Instead of flooding an agent's context window with thousands of tokens of historical data, this skill organizes information into a top-level index with estimated token costs. Agents scan this summary index first, identifying relevant entries before choosing to 'fetch' full-text content. Inspired by progressive disclosure principles, it ensures that attention budgets are preserved for high-value data, preventing the common issue of 'context rot' where relevant information is buried under irrelevant history.

Installation

To integrate Progressive Memory into your agent workflow, execute the following command in your terminal: clawhub install openclaw/skills/skills/lxgicstudios/progressive-memory. After installation, ensure your existing markdown-based memory files are updated to include the mandatory index tables and ID mapping schema provided in the documentation. For optimal performance, update your system instructions (AGENTS.md) to explicitly direct your agent to 'scan the index' before attempting any memory retrieval.

Use Cases

This skill is perfect for complex, long-running projects where maintaining state across multiple sessions is critical. Software engineers can use it to track technical debt and architectural decisions, while content creators can use it to maintain style guidelines and branding rules across thousands of posts. It is particularly effective for teams using agents to manage ongoing client projects, where a repository of 'gotchas' and 'decisions' helps avoid repeating past mistakes or duplicating work.

Example Prompts

1. "Scan the memory index, find the 'Auth Bug' entry, and explain the workaround required to bypass the CLI limitation."
2. "Before starting today's sprint, check the memory index for any 'tradeoff' or 'decision' entries related to our database schema."
3. "Review the current MEMORY.md file and summarize all entries marked with the 🔴 (gotcha) or 🚨 (rule) icons for my upcoming deployment task."

Tips & Limitations

To maximize the utility of this skill, strictly enforce the icon usage guidelines. The system relies on the agent's ability to interpret visual patterns (icons), so consistent labeling is key. Note that this is a manual indexing system; you or your agent must update the index whenever new memory files are generated. It does not replace a vector database, but rather acts as a lightweight, human-readable alternative that excels in transparency and cost management.