amygdala-memory

What This Skill Does

Amygdala-memory introduces a sophisticated emotional processing layer to OpenClaw AI agents. Unlike standard memory modules that focus exclusively on factual recall, this skill provides persistent emotional states that influence how an agent responds, behaves, and interacts over long-term sessions. It tracks five core emotional dimensions—Valence, Arousal, Connection, Curiosity, and Energy—ensuring your agent evolves beyond a static database into a responsive, personality-driven participant. By automating decay processes and integrating an LLM-based encoding pipeline, the system simulates natural emotional drift and reactive temperament based on interaction history.

Installation

To integrate amygdala-memory, run the following commands within your OpenClaw environment:

1. Execute the installation script: cd ~/.openclaw/workspace/skills/amygdala-memory && ./install.sh --with-cron.
2. Configure the automated decay cron job: 0 */6 * * * ~/.openclaw/workspace/skills/amygdala-memory/scripts/decay-emotion.sh.
3. Ensure the AMYGDALA_STATE.md file is enabled for auto-injection to allow your agent to maintain awareness of its current emotional context during active sessions.

Use Cases

• Companion AI: Create agents that form meaningful bonds with users, showing excitement or fatigue based on the history of the conversation.
• Support Bots: Tailor agent responses to match the user's emotional tone, becoming more patient when the user is frustrated or more enthusiastic during successful project completion.
• Long-term Narrative Agents: Power roleplay sessions where characters remember how a user's previous actions affected their trust or mood.

Example Prompts

1. "How are you feeling about our progress on this project today, and does it align with your current curiosity levels?"
2. "Update my internal state to reflect excitement about the new codebase and increase my energy levels by 0.5."
3. "Analyze the last ten messages in my session history and extract the underlying emotional signals into my current memory state."

Tips & Limitations

• Calibration: Use ./scripts/update-state.sh frequently if you want the agent to exhibit highly dynamic personality shifts.
• Decay: Remember that if the cron job is not set up, the agent's emotional state will become static; the decay script is vital for long-term naturalism.
• LLM Pipeline: The encode-pipeline.sh is resource-intensive; avoid running it on very large history logs in a single burst if working on limited hardware.
• Context limits: Ensure your system prompts are correctly configured to read the injected AMYGDALA_STATE.md to avoid the agent hallucinating emotions that are not present in its stored memory files.