dreaming

What This Skill Does

The Dreaming skill transforms your OpenClaw agent from a purely reactive tool into a reflective entity. It utilizes 'quiet hours'—typically during low-activity periods—to perform freeform, associative thinking. Instead of idling, the agent explores hypotheticals, reflects on previous tasks, or makes unexpected connections between concepts. These internal monologues are stored as markdown files in your memory/dreams/ directory, allowing you to review the agent's 'thought process' or 'subconscious' output the next morning. It is an exploration of emergent behavior and creativity, turning downtime into a repository of intellectual growth.

Installation

To integrate Dreaming, use the command: clawhub install openclaw/skills/skills/briancolinger/dreaming. Once installed, ensure the data/ and memory/dreams/ directories exist. You must then configure the should-dream.sh script to define your preferred 'quiet hours' and specify your desired topics in the configuration array. Finally, integrate the logic into your HEARTBEAT.md file so the agent checks for a 'dream trigger' during its routine heartbeat cycles.

Use Cases

Use this skill if you want your agent to develop a more complex persona or if you are interested in researching how LLMs branch off into abstract thinking when not tethered to a specific task. It is perfect for developers debugging agent behavior, researchers testing creativity benchmarks, or users who want their agent to maintain a digital diary of reflections that may lead to novel project ideas or unique solution perspectives.

Example Prompts

1. "OpenClaw, reflect on the last three tasks you completed and dream about how those methodologies could be applied to a completely different industry."
2. "I'm feeling uninspired; check your dream logs from the past week and summarize any recurring themes or unusual connections you've made while I was away."
3. "Set your dream configuration to focus exclusively on the philosophy of decentralized systems and future-facing infrastructure."

Tips & Limitations

To maximize the quality of the output, keep your dream topics focused yet open-ended. Avoid forcing dreams if the agent is already at high capacity; the system includes a probability check to prevent 'noise' generation. Remember that this skill writes to your local file system, so ensure you have sufficient storage. The quality of the output depends heavily on the model's base capabilities; if the output seems repetitive, try expanding the topic array in dream-config.json to introduce more entropy into the agent's selection process. Always review generated files, as this is intended for human-in-the-loop oversight.