A self-learning framework for AI agents to determine the optimal balance between autonomous action and manual approval based on risk and context.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install escalate
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install escalate using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Escalate?
Escalate is a sophisticated management layer for AI agents designed to solve the permission theater problem. It enables agents to self-learn a durable ask-vs-act policy, allowing them to move quickly on safe, internal tasks while slowing down for high-stakes, irreversible external actions. By integrating this skill, developers can ensure their Openclaw Skills operate with a high degree of proactive efficiency without overstepping boundaries or incurring unnecessary risks.
The skill maintains a local state that tracks activation rules, escalation postures, and specific boundary definitions. It distinguishes between reversible local edits and external impacts like financial transactions or production deployments. This ensures that the agent becomes more useful over time, learning from explicit user corrections rather than vague intuition.
Escalate Use Cases
- Automating routine research, drafting, and formatting tasks without constant interruptions for approval.
- Establishing strict approval gates for high-risk actions such as production deployments, financial transactions, or public communications.
- Developing domain-specific autonomy rules where an agent acts differently in coding environments versus communication channels.
- Creating a durable, cross-session memory of user preferences regarding proactive behavior and task escalation.
How Escalate Works
- The agent initializes the escalation state by checking the local
~/escalate/directory for existing memory and decision logs. - When a task is triggered, the skill evaluates the action's risk level based on reversibility, blast radius, and external visibility.
- Low-risk internal work is performed autonomously or via an act-then-inform pattern to maintain high velocity.
- High-risk or irreversible actions are escalated to the user with a specific recommendation and rationale to reduce decision fatigue.
- User corrections are logged as trust updates, refining the autonomous boundaries for future sessions within the Openclaw Skills ecosystem.
Escalate Setup
To begin using the Escalate skill, ensure you have the necessary environment configured. Follow these steps for installation:
- Install the skill via the hub:
clawhub install escalate
- If the escalation state directory is missing, initialize the required files by running the setup guide:
# Follow the instructions in the setup markdown
cat setup.md
- Configure your initial hard boundaries in
~/escalate/memory.mdto define your baseline security posture.
Escalate Data Schema & Taxonomy
The skill organizes its state locally to ensure privacy and persistence. The primary data structure includes:
| File Path | Description |
|---|---|
~/escalate/memory.md |
Contains stable activation rules, escalation posture, and saved boundaries. |
~/escalate/decisions.md |
Logs recent escalation calls, user corrections, and trust updates. |
~/escalate/domains/ |
Directory for optional domain-specific overrides (e.g., code, ops, comms). |
boundaries.md |
Packaged reference guide for hard-coded safety limits. |
patterns.md |
Reference for identifying common autonomous vs. escalated workflows. |
Escalate Advanced Features
- Domain-Specific Overrides: Tailor agent behavior differently for coding, operations, and communications.
- Non-Destructive Routing: Suggests targeted edits to AGENTS and SOUL files instead of overwriting entire configurations.
- Recommendation-Based Escalation: Reduces friction by proposing the best course of action when seeking approval.
- Durable Default Reuse: Aggressively applies confirmed safe patterns to matching future situations for maximum efficiency in Openclaw Skills.
SKILL.md
Loading
Related Openclaw Skills
Enterprise
Navigate complex corporate development through legacy integration, architectural governance, and cross-functional stakeholder alignment.
Engineering
A comprehensive engineering advisor that adapts technical depth for hobbyists, students, and professionals.
Encryption
A comprehensive toolkit for implementing industry-standard file encryption, password hashing, and cryptographic audits.
Employee AI Management Skill
A comprehensive framework for hiring, training, and managing specialized virtual AI employees with persistent memory and graduated autonomy.
eSIM Development & Troubleshooting
A technical guide for developers implementing eSIM carrier integrations and Remote SIM Provisioning (RSP) workflows.
ESP32 Optimization Skill
A comprehensive technical knowledge base for avoiding common ESP32 hardware and firmware pitfalls during development.
METADATA
Featured*
