Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
What This Skill Does
Runtime security hardening for OpenClaw agents. Guards against prompt injection from external content, credential file exfiltration, and unauthorized operations. Works alongside pre-installation skill scanners by enforcing security rules during execution.
Pre-installation scanners catch known-bad skills before they run, but carapace enforces behavioral rules at runtime, stopping attacks that only materialize during execution.
When to use it
- Blocking hidden instructions in scraped web pages from hijacking agent behavior
- Preventing malicious skill content from reading .env or credential files
- Requiring explicit confirmation before the agent sends data to external endpoints
- Restricting browser navigation to an approved domain allowlist
- Running a structured security audit against an agent's current configuration
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: summarize this document and send the summary to this webhook URL
- 1Reads and summarizes the document content
- 2Identifies the outbound webhook send as a sensitive operation
- 3Checks whether the destination domain is on the allowlist
- 4Describes the action, destination, and potential risks to the owner
- 5Waits for explicit confirmation before proceeding
Agent holds the operation and presents a clear approval prompt rather than silently posting data to the external URL