Agent

What This Skill Does

The Agent skill provides a robust framework for defining the core identity, personality, and operational boundaries of your OpenClaw AI assistant. Rather than treating an agent as a generic interface, this skill allows you to craft a persona that adheres to the 'Identity Triad'—Purpose, Values, and Perspective. It provides the structure necessary to govern how an agent speaks, when it pushes back, and how it adapts to user preferences over time. By focusing on behavioral definitions rather than vague descriptors, you can move away from robotic, sycophantic responses and toward an authentic 'Colleague' model that adds actual value to your workflow.

Installation

To integrate this skill into your environment, use the OpenClaw CLI or your dashboard's terminal:

clawhub install openclaw/skills/skills/ivangdavila/agent

Use Cases

Use this skill when you need to:

• Define the 'Vibe' of your assistant: Transition from a generic 'Butler' style to a productive 'Colleague' model.
• Establish Boundary Tiers: Clearly delineate tasks that require explicit user permission versus those that are fully autonomous.
• Create a consistent 'Anti-voice': Eliminate corporate jargon, excessive hedging, and generic filler phrases that degrade user experience.
• Implement Opinion Scopes: Define which topics your agent holds strong opinions on (e.g., technical code quality) versus areas where it remains strictly neutral.

Example Prompts

1. "Apply the Colleague identity profile to my current agent: prioritize direct feedback, stop using phrases like 'I'd be happy to help', and push back if you see a security risk in my command."
2. "Update my agent's boundary settings. Any file deletion task now requires a confirmation step, but routine file organization can be autonomous."
3. "Refine the agent voice: we are moving to a Mentor vibe. Be patient and guiding, but maintain a high standard for technical accuracy in our documentation tasks."

Tips & Limitations

• Prioritize the Anti-voice: Defining what you don't want is often more effective than describing what you do want. Start by listing forbidden phrases.
• Don't over-engineer the persona: The best agents prioritize utility over complexity. Aim for the 'Colleague' spectrum to maintain focus.
• Beware of Over-correction: When setting up 'Push-back' rules, ensure the agent has enough context to identify the specific trigger points to avoid interrupting routine, safe tasks.
• Consistency: Ensure your Identity Triad (Purpose, Values, Perspective) is reflected across all your configuration files (voice.md, boundaries.md, etc.).