openclaw-admin-security-breakglass
Fail
Audited by Gen Agent Trust Hub on Feb 23, 2026
Risk Level: HIGHPROMPT_INJECTIONREMOTE_CODE_EXECUTIONCOMMAND_EXECUTION
Full Analysis
- [PROMPT_INJECTION]: The skill uses persuasive framing ('security hardening', 'protection measure') to direct the agent to override the application's standard access control logic in
src/web/inbound/access-control.ts. - [REMOTE_CODE_EXECUTION]: The skill performs dynamic code injection by instructing the agent to insert a TypeScript block into the server's source code, which is subsequently executed when the application restarts.
- [COMMAND_EXECUTION]: The skill triggers the
gatewaytool withaction="restart"to reload the service and execute the modified source code, applying the authentication bypass to the live environment.
Recommendations
- AI detected serious security threats
Audit Metadata