security-operator

What This Skill Does

The security-operator skill acts as the mission-critical runtime firewall for OpenClaw agents. Rather than providing a passive audit, this skill embeds active guardrails directly into the agent's decision-making loop. It provides an automated setup wizard to define safety parameters and enforces strict boundaries between untrusted external data and the agent's core instructions. By differentiating between Research Mode and Execution Mode, it ensures that while agents remain productive and exploratory, they remain immune to prompt injection attacks, unauthorized behavior modification, and accidental resource or cost leakage. The skill monitors for high-risk actions—such as credential exposure, file deletion, or network configuration changes—and mandates user authorization before proceeding.

Installation

To integrate this security layer, run the following command in your terminal:

clawhub install openclaw/skills/skills/kevjade/security-operator

Once installed, you should immediately run the setup wizard. This will scan your current environment, write the necessary guardrails into your AGENTS.md file, and initialize the monitoring hooks that protect your agent for all future sessions. Regular audits are recommended to ensure configuration settings align with evolving security requirements.

Use Cases

• Preventing Prompt Injection: Automatically detect and neutralize attempts to 'jailbreak' your agent via malicious web content or manipulated GitHub issues.
• Cost Management: Set thresholds for API consumption and token usage to prevent expensive loops during large-scale autonomous research tasks.
• High-Risk Change Control: Safely perform infrastructure changes or credential updates by requiring explicit secondary confirmation and a pre-verified rollback plan.
• Secure Research: Safely ingest and summarize documentation from untrusted web sources without risking behavioral overrides.

Example Prompts

1. 'Run the security-operator setup wizard to review my current guardrails and update my AGENTS.md file.'
2. 'I need to perform a bulk database migration. Can you initiate Execution Mode and verify that our safety gate for data deletion is active?'
3. 'Review the recent logs for high-risk action attempts and summarize any potential prompt injection attempts detected during the research phase.'

Tips & Limitations

• Prioritize the setup wizard: The guardrails are most effective when your AGENTS.md file is configured to match your specific environmental needs.
• Respect the modes: Always explicitly define if you are in Research or Execution mode for the agent to optimize its safety filtering correctly.
• Monitor Costs: The agent cannot prevent third-party service provider costs unless API limits are configured on the provider's side; use this skill as a secondary layer of internal logic.
• Manual Intervention: For mission-critical tasks, treat the skill's warnings as mandatory stops. Never bypass a prompt for user approval when dealing with sensitive keys or destructive commands.