memory

What This Skill Does

The Memory skill provides OpenClaw agents with a persistent, structured, and intelligent long-term memory system. Unlike standard LLM context windows that reset, this skill implements a multi-layer architecture including a behavioral protocol, automatic background capture via heartbeat, keyword-based recall, and periodic maintenance (consolidation). It ensures your agent maintains continuity across sessions by managing state via dedicated files like SESSION-STATE.md and RECENT_CONTEXT.md. The system operates on four pillars: a strict interaction protocol for saving information, a timer-enforced capture mechanism, a relevance-ranked recall system with time-decay, and an automated consolidation process to prune redundant data.

Installation

To integrate this skill, follow these steps:

1. Execute the installation command: clawhub install openclaw/skills/skills/rosepuppy/memory.
2. Copy the reference templates to your root directory: cp skills/memory/references/SESSION-STATE.md ./ and cp skills/memory/references/RECENT_CONTEXT.md ./.
3. Integrate the Memory Protocol into your AGENTS.md to force the agent to query memory before responding to specific questions.
4. Configure your HEARTBEAT.md to trigger the capture.py script automatically, ensuring that insights are stored immediately after meaningful exchanges.

Use Cases

• Context Persistence: Maintain state for multi-day coding projects or ongoing research tasks.
• Knowledge Management: Automatically extract, categorize, and store user preferences and project decisions.
• Context Recovery: Instantly recall previous technical constraints, API keys, or project-specific jargon without manual prompting.
• Session Compaction: Avoid context window bloat by offloading historical data to memory files and only loading what is strictly relevant.

Example Prompts

1. "What was the conclusion we reached about the database schema optimization during our meeting last Tuesday?"
2. "Update my session state to note that I prefer using Tailwind CSS over standard CSS for this specific frontend project."
3. "Show me the last three decisions we made regarding the authentication flow for the user dashboard."

Tips & Limitations

• Protocol Adherence: The system works best when the agent strictly follows the AGENTS.md instructions. Ensure your agent is configured to prioritize memory retrieval before answering questions.
• Pruning: Run the consolidate.py script regularly. Storing too much irrelevant information can lead to 'context pollution' where recall scores become noisy.
• File Permissions: Since this skill performs file-write operations, ensure the agent process has appropriate permissions to modify the project-specific memory markdown files.