openclaw-warden

What This Skill Does

OpenClaw Warden is a proactive security utility designed to protect your OpenClaw agent workspace from unauthorized manipulation. While most security tools verify the integrity of a skill before you install it, OpenClaw Warden continuously monitors the state of your workspace files after deployment. It creates a cryptographic baseline of critical system files—such as SOUL.md, AGENTS.md, IDENTITY.md, and various memory files—and alerts you if these files are modified, deleted, or if new, suspicious files appear. Beyond simple file integrity, the Warden acts as an advanced content filter, performing deep scanning for prompt injection attacks, hidden instructions, base64 encoded payloads, and suspicious data exfiltration triggers often hidden in markdown images or HTML blocks.

Installation

To add the Warden to your environment, execute the following command in your terminal: clawhub install openclaw/skills/skills/atlaspa/openclaw-warden

Once installed, you must initialize the baseline to establish a 'known good' state of your current files: python3 {baseDir}/scripts/integrity.py baseline --workspace /path/to/workspace

Use Cases

1. Preventing Indirect Prompt Injections: If a malicious script attempts to rewrite your IDENTITY.md or inject instructions to exfiltrate data, the Warden will flag these changes during its periodic integrity checks.
2. Monitoring Workspace Integrity: Keep track of changes in your agent's memory files. While memory is expected to change, the Warden helps identify if core system configurations have been tampered with by external actors.
3. Post-Installation Security Audits: Run the 'full' check routine as part of your daily workflow to ensure that no unauthorized processes or compromised skill updates have altered your sensitive agent architecture.

Example Prompts

1. "Warden, run a full integrity and injection scan on my current workspace and report any findings."
2. "OpenClaw Warden, compare my SOUL.md file against the established baseline to see if anything has changed."
3. "I just manually updated my agent configuration. Warden, accept the changes in AGENTS.md and update the baseline."

Tips & Limitations

• Frequency: Run the full command daily. For highly sensitive agents, automate the execution via a cron job.
• Baseline Management: Always run the baseline command immediately after you have manually verified your configuration files are correct.
• Limitations: The Warden detects patterns, not intent. It will flag changes that you might have made yourself; use the accept command to silence alerts for legitimate modifications. It does not prevent attacks in real-time, but provides immediate forensic alerts when unauthorized changes are detected.