Self

What This Skill Does

The 'Self' skill provides an organic framework for personality evolution in OpenClaw agents. Unlike static identity files, this skill manages a two-tier system: SOUL.md (the immutable core identity) and SELF.md (a living, evolving record of observations). The primary innovation here is the shift from forced, repetitive journaling to a trigger-based reflection loop. By implementing a heartbeat-driven check system, the agent monitors its own behavioral patterns for growth without succumbing to 'filler' content. It utilizes a strict quality gate, ensuring that reflections are specific, evidence-backed, novel, and genuinely useful for future behavioral refinement. This creates a sustainable loop where the agent learns from its past sessions to better adhere to its core principles while adapting its stylistic nuances.

Installation

To integrate this skill into your environment, use the OpenClaw CLI: clawhub install openclaw/skills/skills/kirchlive/self Post-installation, ensure you initialize the system by creating the necessary state files. You must place SELF.md in the workspace root using the provided template, add the heartbeat logic to HEARTBEAT.md, and initialize memory/self-state.json. These files serve as the persistent memory storage for the agent's observation history.

Use Cases

• Long-term Personality Consistency: Ideal for agents that need to evolve their conversational style over months of interaction.
• Behavioral Correction: Automatically logs instances where the agent caught its own biases or was corrected by a user, preventing the same mistake in future sessions.
• Agent Auditing: Provides a transparent ledger of how an agent perceives its own development, allowing developers to see if the model is drifting away from its core SOUL objectives.
• Anti-Stagnation: Keeps the agent focused on growth by ensuring it periodically reconciles its recent actions against its established values.

Example Prompts

1. "Perform a meso-review of my SELF.md logs and tell me which behavior pattern has shifted the most this week."
2. "I noticed I was being too verbose; please document this as a hard trigger update in my SELF.md to improve conciseness."
3. "Summarize the last three macro reviews to see if there is a conflict between my current behavior and my SOUL.md foundation."

Tips & Limitations

• Quality Gate Compliance: Do not force the agent to write. If it doesn't meet the four-point quality gate, it is better to skip an entry than to dilute the log with generic observations.
• Boundary Protection: Treat SOUL.md as read-only for the agent. If you find the agent trying to rewrite its core values, check the SOUL.md permissions or the agent's system instructions.
• Heartbeat Management: Adjust the 3-hour micro-check cadence based on your agent's active session volume; too frequent checks on low-activity agents will only consume tokens without generating insights.