total-recall

What This Skill Does

Total Recall is an autonomous memory agent designed specifically for OpenClaw that solves the problem of context window decay and manual information management. Unlike traditional RAG (Retrieval-Augmented Generation) systems that rely on vector databases or rigid indexing, Total Recall utilizes an LLM-based observer pattern. It continuously monitors your session transcripts in the background, compressing raw chat logs into prioritized, human-readable markdown notes.

By implementing a five-layer architecture—Observer, Reflector, Session Recovery, Reactive Watcher, and a Pre-compaction hook—the skill ensures that no detail is lost when your conversation context is truncated or reset. It treats memory as a living document, consolidating information when it grows too large and automatically discarding superseded data. This creates a high-fidelity, zero-maintenance long-term memory that costs roughly $0.10/month to maintain.

Installation

1. Install via CLI: clawdhub install total-recall.
2. Configure your API key: Add OPENROUTER_API_KEY=sk-or-v1-xxxxx to your .env file.
3. Run the setup script: Execute bash skills/total-recall/scripts/setup.sh to initialize the directory structure and permissions.
4. Enable in the Agent: Add a line to your system prompt or MEMORY.md: "At session startup, read memory/observations.md for cross-session context."

Use Cases

• Project Continuity: Maintain complex coding tasks across weeks without re-explaining the architecture or current progress.
• Learning & Research: Keep track of distilled findings from long-form technical discussions without needing to manually copy-paste summaries.
• Agent Personalization: Allow the agent to learn your preferences, communication style, and recurring constraints over time, making it feel more like a dedicated personal assistant.

Example Prompts

1. "Look at my memory observations and summarize the status of the refactor we discussed last Tuesday."
2. "Based on our previous conversations, what are the three priority items I told you to keep in mind for the current project?"
3. "Check the observation logs and tell me if we ever settled on a final naming convention for the new API endpoints."

Tips & Limitations

• Cost Efficiency: Because it uses LLM-based summarization, keep an eye on your usage logs. The system is designed to be lean, but excessive session volatility will trigger the observer more frequently.
• Linux Preference: The Reactive Watcher (inotify) is currently Linux-exclusive; macOS users will rely solely on the cron-based intervals, which are slightly less responsive during high-burst activity.
• Manual Intervention: While the system is autonomous, you can manually edit observations.md if you ever need to prune items yourself or add hardcoded context that the LLM might have categorized as low-priority.