sandwrap

What This Skill Does

Sandwrap is a specialized security layer designed for OpenClaw AI that provides soft-sandbox protection for running untrusted third-party skills. It does not utilize virtual machines or system-level isolation but instead employs a multi-layered, prompt-based defense mechanism to mitigate approximately 85% of common prompt injection and malicious intent attacks. By wrapping target skills in a strictly defined instruction hierarchy, it forces external content to adhere to your predefined security parameters.

Installation

To install this skill, run the following command in your terminal: clawhub install openclaw/skills/skills/rubenaquispe/sandwrap

Use Cases

Sandwrap is essential for users who frequently integrate third-party community skills into their workflow. It is highly recommended for:

• Analyzing suspicious URLs or files where you suspect malicious payload intent.
• Testing newly installed skills from unknown sources before granting them full system permissions.
• Processing unstructured data from untrusted sources (like user-submitted web forms or external APIs) that could contain prompt injection vulnerabilities.
• Enforcing strict access control by using presets like 'read-only' to browse sensitive codebases without risking accidental data modification.

Example Prompts

1. "Run github-repo-summarizer in sandwrap read-only"
2. "Analyze this suspicious email link using sandwrap web-only"
3. "Run code-converter-v2 in sandwrap audit preset"

Tips & Limitations

• Understand the Soft-Sandbox Nature: Sandwrap is not an airtight VM or container. It is a prompt-level safety filter. Therefore, do not process high-stakes, hyper-sensitive credentials through sandwrapped skills.
• Preset Selection: Always choose the most restrictive preset that satisfies your workflow needs. Default to 'read-only' whenever possible to minimize the attack surface.
• Pay Attention to Approvals: Layer 4 requires manual human approval for sensitive actions. If you see an injection warning, abort immediately rather than bypassing.
• Limitations: While Sandwrap provides a significant security boost (~85%), sophisticated, adaptive, or novel attack vectors may still bypass these protections. For absolute deterministic security, use a dedicated physical or virtualized sandbox environment instead.