openclaw-mem

What This Skill Does

The openclaw-mem skill acts as an intelligent, automated curator for the OpenClaw agent's cognitive lifecycle. It solves the classic LLM context-window problem by implementing a multi-tiered architecture that separates ephemeral conversation data from durable, long-term knowledge. By proactively moving information from RAM (Session Memory) to disk-based logs and curated fact files (MEMORY.md) before automated compaction, it ensures that your agent retains context across days, weeks, and months of interaction without bloating the active session window. It operates silently, using pre-compaction hooks to ensure your agent 'remembers' the things that actually matter.

Installation

To install the skill, execute the following command in your terminal: clawhub install openclaw/skills/skills/weareallsatoshin/openclaw-mem

After installation, ensure session memory indexing is enabled in your configuration: clawdbot config set agents.defaults.memorySearch.experimental.sessionMemory true

Use Cases

• Project Continuity: Maintain complex project decisions and architectural choices across multiple distinct sessions.
• Preference Retention: Automatically store user preferences for formatting, tone, and tool usage so you don't have to repeat them.
• Automated Work Logging: Automatically timestamp and save your daily progress and experimental results into structured, chronological log files.
• Context Optimization: Keep the immediate chat context lean and clean, ensuring the agent only holds onto relevant information for the current task while offloading background knowledge to the disk.

Example Prompts

1. "We just decided to pivot from PostgreSQL to SurrealDB for this project; please ensure this is recorded in MEMORY.md so you don't suggest SQL migrations later."
2. "Look through the last three days of memory logs and summarize the key blockers we encountered during the API integration."
3. "Save my preference for using arrow functions in all TypeScript files as a formal project rule."

Tips & Limitations

• Be Selective: Only promote information to MEMORY.md if it is a decision, rule, or invariant that needs to persist long-term. Excessive entries will reduce the effectiveness of the search retrieval.
• Use the Hierarchy: If you are unsure if a piece of information is permanent, write it to the daily log first. Promoting it later keeps your primary memory files clean.
• Format Matters: The skill relies on specific IDs and tagging (e.g., DEC- for decisions, PREF- for preferences). Stick to these naming conventions to ensure the memory_search function can index your data effectively.
• Resource Management: Remember that while disk space is cheap, context retrieval consumes tokens. Always use memory_get to bring in only the relevant slices of your long-term memory to keep costs and latency low.