Self-Guardian is a non-negotiable safety layer that prevents AI agents from executing destructive commands or making accidental file modifications.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install self-guardian
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 self-guardian 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 Self-Guardian?
Self-Guardian is a lightweight self-monitoring framework designed to enforce strict safety rules on AI coding agents. It acts as a digital safety net, ensuring that every file operation, shell command, and system modification is vetted against a Pre-Action Checklist. By integrating these rules into your workflow, you ensure that Openclaw Skills maintain high integrity and avoid catastrophic errors such as accidental deletions or risky root-level operations.
The skill categorizes files into protection tiers and monitors shell activity for dangerous patterns. It forces the agent to pause, reason, and seek explicit user confirmation for high-stakes tasks. This makes it an essential tool for developers who want to delegate complex tasks to AI without worrying about unrecoverable data loss or security breaches.
Self-Guardian Use Cases
- Preventing accidental deletion of critical configuration files like .env or cryptographic keys.
- Blocking execution of destructive shell commands such as rm -rf or force pushes without confirmation.
- Managing bulk file refactoring by enforcing batch operation throttles and progress reporting.
- Protecting sensitive system directories and credential stores from unauthorized AI access.
- Ensuring recovery paths exist by prompting for git checkpoints before large-scale code changes.
How Self-Guardian Works
- Keyword Activation: The skill triggers automatically when it detects keywords like delete, sudo, or drop database in the user request.
- Pre-Action Checklist: The agent internally evaluates if the action is reversible, properly scoped, and if a safer alternative exists.
- Tiered Validation: Files are checked against protection tiers (Never Touch, Confirm First, or Safe) to determine the level of user consent required.
- Command Filtering: Shell commands are matched against a blocklist; any matches (e.g., chmod 777) require explicit user justification.
- Batch Throttling: Large-scale operations are broken into manageable chunks (max 5 files) to prevent unmonitored mass changes.
- Post-Action Reporting: The skill provides a summary of actions taken and suggests recovery breadcrumbs if needed.
Self-Guardian Setup
To activate Self-Guardian within your environment, include the skill definition in your agent configuration. Ensure your agent is configured to respect the Openclaw Skills protocol for non-negotiable constraints.
# Standard activation within your agent shell
openclaw-agent install self-guardian
No additional system-level dependencies are required as the skill operates as a set of logical constraints within the agent's reasoning loop.
Self-Guardian Data Schema & Taxonomy
Self-Guardian uses a structured approach to categorize system resources and operational limits:
| Data Type | Description | Examples |
|---|---|---|
| Red Tier | Strictly prohibited without direct per-file consent | .env, id_rsa, .git/ |
| Yellow Tier | Requires impact explanation before change | package.json, Dockerfile, tsconfig.json |
| Blocklist | High-risk shell command patterns | git reset --hard, sudo, rm -rf |
| Throttle Limit | Max operations before mandatory reporting | 3 files for deletions; 5 for edits |
Self-Guardian Advanced Features
- Heuristic Uncertainty Detection: Automatically pauses and asks for clarification if the user's instructions contradict previous session rules.
- Recovery Breadcrumbs: Forces the agent to read file contents before overwriting, ensuring the original state is preserved in the conversation history.
- Path Traversal Guard: Monitors for risky path patterns like .. or absolute root paths to prevent escaping the project context.
- Model-Awareness Logic: Specifically designed to compensate for LLM over-eagerness by requiring plain-language explanations for long or complex shell commands.
SKILL.md
Loading
Related Openclaw Skills
Beauty Recycle Rewards
Beauty Recycle Rewards automates the design of points-based loyalty programs and sustainable empty-container recycling incentives for DTC beauty brands.
Beauty Free-Shipping Top-Up Optimizer
An AI-driven merchandising strategist that designs high-conversion free shipping threshold logic and top-up recommendations for beauty brands.
VolcEngine Video Generation
A professional-grade AI video generation skill that interfaces with VolcEngine to transform natural language prompts into cinematic video content.
Skill Evolver
A comprehensive skill lifecycle manager designed to discover, orchestrate, and evolve reusable AI agent capabilities.
segundo
A zero-friction CLI-based second brain for capturing and retrieving thoughts using daily markdown journals and semantic search.
Nasdaq-100 Futures Quote (Node.js)
A high-performance Node.js skill for AI agents to fetch live Nasdaq-100 futures and Yahoo Finance market data.
Featured*
