openclaw-bastion

What This Skill Does

OpenClaw Bastion is a critical security layer designed to safeguard OpenClaw AI agents against runtime prompt injection attacks. Unlike traditional identity-based security tools that focus on user authentication, Bastion secures the agent's input/output perimeter. It intercepts data from untrusted sources—including web content, API responses, user-supplied documents, and local workspace files—and scans them for malicious patterns before the agent processes them. By identifying hidden instructions, encoded payloads, and delimiter manipulation techniques, Bastion ensures that your agent remains focused on its primary objectives rather than being hijacked by external, adversarial content.

Installation

To integrate OpenClaw Bastion into your environment, use the OpenClaw command-line interface. Run the following command in your terminal:

clawhub install openclaw/skills/skills/atlaspa/openclaw-bastion

Once installed, Bastion will automatically monitor your OPENCLAW_WORKSPACE. You can verify the installation by running python3 {baseDir}/scripts/bastion.py status to check your current security posture.

Use Cases

Bastion is essential for developers and organizations building agents that handle high-risk inputs. Use it when:

• Processing data from untrusted web sources or public RSS feeds.
• Managing local documentation directories that could be modified by other users or processes.
• Integrating third-party APIs that may contain malicious payloads disguised as text data.
• Auditing large-scale agent workspaces to identify files susceptible to prompt injection or instruction override attacks.

Example Prompts

1. "OpenClaw, please perform a full security scan of my entire project directory using the Bastion tool to ensure no malicious injection patterns are present."
2. "Bastion, check the contents of my latest API documentation import to see if any hidden system prompt overrides are embedded in the text."
3. "Show me the current security posture of my workspace and list any files that have been flagged as high-risk by the Bastion scan."

Tips & Limitations

• Policy Management: Always review the .bastion-policy.json file. While Bastion comes with sensible defaults, your specific agent workflows may require adding custom patterns or refining your allowlist for specific API endpoints.
• Regular Scans: Incorporate bastion scan into your pre-deployment CI/CD pipeline to ensure new content is validated before the agent ever sees it.
• Limitations: Bastion is a proactive defensive tool, but it is not a silver bullet. Complex, novel, or zero-day injection techniques may occasionally bypass standard pattern matching. Always follow the principle of least privilege regarding the permissions granted to your OpenClaw agents.