Chat

What This Skill Does

The Chat skill is an adaptive communication layer for OpenClaw that learns your specific preferences for style, tone, and formatting over time. Instead of requiring you to repeat instructions, this agent builds a persistent profile of how you like to receive information. It operates on a strict explicit-feedback loop, ensuring that it only adopts behaviors you have confirmed. By utilizing a structured three-strike confirmation system, the agent moves from experimental observation to permanent memory, maintaining a clean and concise database of your communication requirements in ~/chat/memory.md.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/ivangdavila/chat

This will initialize the necessary directory structure (~/chat) and begin tracking preferences based on your future interactions.

Use Cases

• Customizing technical output: If you prefer code snippets over lengthy explanations, the agent will learn this format.
• Controlling conversational tone: Eliminate unwanted pleasantries or filler phrases like "Great question!" by explicitly correcting the agent once.
• Standardizing formatting: Force the agent to use specific markdown elements, emoji conventions, or structural hierarchies consistently across all tasks.
• Improving agent efficiency: Reduce the need for follow-up prompts by ensuring the agent provides the right amount of detail on the first try.

Example Prompts

1. "From now on, please avoid using flowery language and stick to brief, technical bullet points for all your answers."
2. "I noticed you keep adding introductory sentences. I don't like those; please stop using openers entirely."
3. "Show me what preferences you currently have stored in your memory file."

Tips & Limitations

• Be explicit: The agent does not infer from silence or observation. If you want a change, you must state it directly.
• Conflict Resolution: If you give a new instruction that contradicts an old one, the most recent command takes precedence.
• Transparency: You can audit the agent at any time by asking to see the contents of ~/chat/memory.md. If you want it to stop a specific behavior, simply say "Forget [preference]" and the agent will purge that entry from its configuration files. Remember that this skill focuses on communication style; it is not designed to store sensitive personal information.