memory-pipeline

What This Skill Does

The Memory Pipeline is a sophisticated cognitive architecture designed for OpenClaw agents to transcend the limitations of stateless session-based AI. Rather than relying on simple vector searches, this skill treats agent intelligence like a human memory system. It operates via a structured lifecycle: it extracts critical decision points, preferences, and commitments from your interaction history; organizes them into a dynamic knowledge graph; and synthesizes them into a daily briefing. By enforcing 'pre-game' routines and post-session after-action reviews, it ensures the agent maintains continuity, reduces hallucinations stemming from forgotten details, and enforces execution discipline without interrupting the live workflow.

Installation

To begin using the Memory Pipeline, ensure you have Python 3 installed in your environment. Run the installation command via the terminal: clawhub install openclaw/skills/skills/joe-rlo/memory-pipeline. Once installed, execute the setup script located at skills/memory-pipeline/scripts/setup.sh. This script scans your local environment to identify existing LLM API keys (OpenAI, Anthropic, or Gemini) and initializes the local memory/ directory structure where your knowledge graph and briefing files will reside.

Use Cases

• Long-term Project Management: Maintaining project context, tech stack decisions, and pending milestones across multiple weeks.
• Knowledge Consolidation: Ingesting external data from ChatGPT exports or local documentation into a unified, searchable agent brain.
• Agent Consistency: Ensuring that agent personality, formatting preferences, and tool usage policies remain consistent across independent work sessions.
• Executive Briefing: Starting each morning with a 'pre-game' cheat sheet that highlights high-priority tasks and previous day learnings.

Example Prompts

1. "Run the memory extraction process on yesterday's transcripts and update my project knowledge graph accordingly."
2. "Generate a briefing for my next session that highlights the technical blockers we identified last week regarding the database migration."
3. "Ingest these chat export logs into the memory directory to help the agent understand my preference for Python documentation standards."

Tips & Limitations

To get the most out of the Memory Pipeline, we recommend automating the full execution cycle using your HEARTBEAT.md file. This ensures that the agent is refreshed every day without manual intervention. Note that the memory extraction process is most effective when your interaction history is clean; avoid using it to process large binary files or irrelevant logs. While the knowledge graph creates powerful bidirectional links, avoid overloading the agent with contradictory information to prevent retrieval biases. Remember that the system is designed for between-session maintenance, not real-time mid-task correction; use it to set the stage for your work sessions, not as a replacement for real-time task oversight.