agent-config

What This Skill Does

The agent-config skill is the primary control interface for modifying OpenClaw's core configuration files. It acts as an intelligent layer that prevents configuration bloat while ensuring that your agent's instructions, personality, and operational rules remain synchronized and performant. Instead of manually editing flat files, this skill validates your changes against a strictly defined directory architecture, ensures character limits are respected to avoid context truncation, and enforces format best practices like imperative structuring for operational logic and narrative-based tone guidance for personality adjustments.

Installation

To integrate this configuration management layer into your OpenClaw environment, execute the following command in your terminal:

clawhub install openclaw/skills/skills/thatguysizemore/agent-config

Use Cases

This skill is essential when you need to evolve your agent's behavior. Use it when:

• You want to refine the agent's tone or ethical boundaries (SOUL.md).
• You need to add, remove, or update operational processes, routing rules, or delegation workflows (AGENTS.md).
• You are adjusting personal identifiers or core 'vibe' settings (IDENTITY.md).
• You are onboarding new user preferences or family-specific data (USER.md).
• You need to update tool definitions, API locations, or command examples (TOOLS.md).

Example Prompts

1. "I want you to be more direct and stop using hedging phrases like 'it seems like' or 'perhaps'. Please update your SOUL.md to reflect this new personality constraint."
2. "We need to add a new step to our project research workflow. Whenever I ask for a market analysis, first check the news API, then update the MEMORY.md file, and finally summarize. Please update AGENTS.md."
3. "Change your core identity to be more formal and add 'Professional Assistant' to your identity header in IDENTITY.md."

Tips & Limitations

• Check Before Writing: Always use the built-in size check feature. If your target file is approaching the 20,000-character limit, the agent may struggle to process the full file at startup. Refactor redundant info into shared references if necessary.
• File Selection: Never put operational logic in SOUL.md. Use AGENTS.md for logical, step-by-step processes. Subagents can only access AGENTS.md and TOOLS.md, so keep mission-critical rules in those files.
• Duplication: Always grep the target file for keywords before adding new instructions to prevent rule collisions.