Escalate

What This Skill Does

The Escalate skill is the engine of cognitive growth for your OpenClaw agent. It functions as an adaptive heuristic memory system that tracks delegation patterns to determine when the agent should act autonomously versus when it must request human intervention. Instead of forcing you to toggle settings, the agent observes your commands and the subsequent feedback you provide.

By following a strict core loop of Observing, Recognizing Patterns, and Confirming, the skill minimizes "notification fatigue" while maintaining high safety standards. It evolves through five specific trust levels—blocked, ask, pattern, confirmed, and locked—ensuring that the agent only takes independent action after you have explicitly validated its competence in a specific domain. The system is hard-coded to prioritize caution, particularly regarding sensitive areas like financial transactions, data deletion, or external communication.

Installation

To integrate this skill into your environment, use the OpenClaw CLI: clawhub install openclaw/skills/skills/ivangdavila/escalate

Use Cases

• Routine Workflow Optimization: Automate repetitive file organization tasks or code formatting after the agent demonstrates consistent performance.
• Reduced Friction for Common Queries: Stop answering "should I proceed?" for low-risk, frequent operational tasks like cleaning temporary cache files.
• Boundary Enforcement: Use the built-in hard-block list to prevent the agent from accidentally triggering dangerous actions, such as production deployments or unauthorized API key rotations, regardless of previous successful minor tasks.

Example Prompts

1. "I'm tired of confirming this every time; just refactor the code however you think best from now on."
2. "You decide how to organize these download files, I trust your judgment on this folder structure."
3. "Go ahead and handle the formatting for these JSON files, I don't need to review them anymore."

Tips & Limitations

• Patience is Key: The agent will not become autonomous after a single command. It requires at least two instances of delegation before it will even suggest taking the "confirmed" path. This protects your workflow from premature automation.
• Hard Limits: Do not attempt to force the agent to automate financial transactions or contract signing. The Escalate skill enforces a hard-coded "blocked" status for these categories for your security.
• Context Awareness: Always review the boundaries.md file in your local implementation to understand exactly which tasks remain strictly off-limits to agent autonomy. If you ever feel the agent is becoming too autonomous, you can manually reset its trust levels.