continuity

What This Skill Does

The Continuity framework represents a paradigm shift from passive logging to active AI cognitive development. By automating the reflection process, Continuity enables OpenClaw to process session data during idle periods, transforming raw interaction logs into structured, high-confidence memories. It categorizes information into distinct types such as facts, preferences, and principles, assigning confidence scores that help the agent discern between explicit user instructions and speculative inferences. The system then generates contextually relevant follow-up questions to refine the agent's self-model, ensuring that the AI continues to grow and align with the user's evolving needs over long-term engagements.

Installation

To integrate Continuity into your OpenClaw agent, execute the following command in your terminal:

clawhub install openclaw/skills/skills/riley-coyote/vektor-continuity

Once installed, you must define the HEARTBEAT logic in your project configuration. Ensure the CONTINUITY_IDLE_THRESHOLD and CONTINUITY_MIN_MESSAGES environment variables are set to tune the reflection trigger frequency according to your specific workflow preferences.

Use Cases

• Project Long-term Continuity: Perfect for complex development projects where design patterns and project-specific preferences need to be remembered across multiple coding sessions.
• Personalized Assistance: Ideal for administrative or creative agents that need to adapt to user stylistic preferences, communication habits, and relationship dynamics.
• Knowledge Management: Use this to distill high volumes of research or brainstorming conversations into persistent, categorized knowledge files that guide future AI output.

Example Prompts

1. "Reflect on our last session and see if you've gathered enough confidence to commit my preferred code style to memory."
2. "Show me the pending questions generated from our morning sync so we can clarify my requirements before starting this feature."
3. "Check the current memory state to see if you have successfully integrated the feedback I gave you yesterday about project constraints."

Tips & Limitations

• Quality over Quantity: Focus on maintaining the questions.md file regularly; clearing out resolved questions keeps the agent's focus sharp.
• Calibration: If the agent is making too many speculative inferences, adjust the confidence score thresholds or manually prune the reflections/ directory.
• Limitations: The skill requires a baseline of message history to function effectively; it is not suited for extremely short or sporadic micro-sessions.