elite-longterm-memory

What This Skill Does

The elite-longterm-memory skill is an advanced, multi-layered architecture designed to eliminate the context-window limitations of modern AI agents. By integrating a Write-Ahead Log (WAL) protocol with vector search (LanceDB) and persistent Git-backed knowledge graphing, it ensures that your AI agent never "forgets" project requirements, user preferences, or past decisions. It functions by syncing short-term, medium-term, and long-term memory across a tiered system, ranging from a volatile 'Hot RAM' SESSION-STATE.md file to a structured 'Cold Store' for historical permanent knowledge. This skill effectively transforms your agent from a stateless chatbot into a continuous, stateful collaborator that evolves alongside your workflow.

Installation

To install this skill, run the following command in your terminal within your OpenClaw-enabled project:

clawhub install openclaw/skills/skills/itsjustfred/elite-longterm-memory-1-2-3

Once installed, ensure your environment has access to the local storage path for the LanceDB instance to maintain persistence between sessions.

Use Cases

• Long-term Project Management: Retain complex architectural decisions over weeks of development without having to re-explain requirements.
• Preference Personalization: Automatically recall specific coding styles, preferred library versions, or UI design constraints without prompting.
• Contextual Troubleshooting: Search historical logs and prior error resolutions stored in the Knowledge Graph to prevent repeating technical mistakes.
• Vibe-Coding Sessions: Maintain continuity during creative sessions, allowing the AI to 'remember' the specific tone and stylistic goals established at the start of a project.

Example Prompts

1. "What was the core architectural decision we made regarding the database schema last week? Check our permanent memory."
2. "Store this in memory: I prefer using Tailwind CSS for styling instead of raw CSS modules, and keep this preference for all future projects."
3. "Recall all blockers related to the authentication module so we can resume working from where we left off."

Tips & Limitations

• Write Frequency: Since the 'Hot RAM' layer (SESSION-STATE.md) updates before every response, ensure your environment allows for frequent file-write operations.
• Manual Cleanup: Periodically review the MEMORY.md file to summarize or archive outdated information, preventing the context injection from becoming bloated over time.
• Importance Scoring: When manually storing information, utilize the importance parameter (0.0 to 1.0) to help the vector search prioritize key data points during auto-recall.