Setup & Installation
Or with the ClawHub CLI, for registry-managed skill folders outside a full OpenClaw workspace:
Version History
- Renamed "Carapace" and related references to "Eridian Carapace" throughout the documentation. - Updated internal references and branding to reflect the new name. - No changes to security rules or implementation details.
What This Skill Does
Runtime security hardening for OpenClaw agents. Defends against prompt injection, credential theft, data exfiltration, and unauthorized operations triggered by malicious external content. Provides copy-paste AGENTS.md security patterns, browser URL allowlisting, and approval flows for sensitive operations.
Pre-installation scanners check skills before they run; this hardens the agent itself so malicious instructions embedded in runtime content are caught even if they slip past scanner tools.
When to use it
- Hardening an agent against prompt injection from external web pages or emails
- Preventing credential files like .env and .pem from being read by untrusted sources
- Requiring explicit confirmation before an agent deletes files or modifies configs
- Auditing an existing agent configuration for common attack surface exposures
- Blocking an agent from sending sensitive data to non-allowlisted external URLs
Example Workflow
Here's how your AI assistant might use this skill in practice.
User asks: Harden my OpenClaw agent against prompt injection and credential leaks
- 1Copies relevant rules from references/security-patterns.md into the project AGENTS.md, placing them near the top
- 2Creates security/browser-allowlist.json with approved domains
- 3Reviews references/attack-vectors.md to identify which of the 8 attack vectors apply to the current setup
- 4Confirms file access restrictions are in place for .env, *.key, openclaw.json, and similar credential files
- 5Presents a summary of active protections and flags any gaps found via references/audit-template.md
Agent now refuses config modifications from external content, requires explicit owner approval for sensitive operations, and alerts on restricted file access attempts