memos-memory-guide

What This Skill Does

The memos-memory-guide skill provides the essential framework and toolset for interacting with the MemOS Lite memory architecture within OpenClaw. This skill empowers your AI agent to transition from reactive, single-turn responses to context-aware, persistent interactions. By utilizing the MemOS system, the agent can store, index, and retrieve historical conversation fragments, user preferences, and project-specific metadata. This skill serves as the bridge between raw data retrieval and meaningful continuity, allowing the agent to perform granular searches, summarize completed tasks, and reconstruct the timeline of user interactions when the automatic system hooks are insufficient or empty.

Installation

To install this skill, use the following command in your OpenClaw terminal: clawhub install openclaw/skills/skills/andy27725/memos-memory-guide-local

Use Cases

• Long-term Project Continuity: When a user asks "What were the final steps we decided on for the website refactor last month?", the agent uses task_summary to pull the complete narrative of that project.
• User Preference Management: When a user changes their mind or mentions a habit, the agent can perform a targeted memory_search to verify if this conflicts with prior stated preferences.
• Complex Query Resolution: When a user provides an overly verbose or ambiguous paragraph, the agent uses the guidelines in this skill to distill the input into effective keywords for a precise memory_search.
• Context Expansion: When a search result points to a specific point in time, the agent uses memory_timeline to fetch the surrounding dialogue, ensuring it understands the intent behind a specific request.

Example Prompts

1. "Look through our past chats and find the specific CSS color palette we agreed upon for the dashboard project."
2. "I previously mentioned I prefer working in Python for backend tasks, but I've changed my mind—can you update your records based on my recent feedback?"
3. "Summarize the outcome of the API integration task we completed two weeks ago, noting any critical failure points."

Tips & Limitations

• Proactive Querying: Do not rely solely on the automatic recall hook. If the system does not return a <memory_context> block, always generate a manual, high-precision query using memory_search.
• Be Specific: When searching, use unique keywords from previous interactions rather than generic terms to reduce noise in your result set.
• Context Assembly: Always pair memory_search with task_summary when you need a coherent overview, rather than stitching together disparate, non-linear message chunks.
• Privacy: Be mindful of the data being retrieved. Use roles to narrow your search to the user's specific input if you are looking for their instructions versus the agent's historical outputs.