make-bot-ui
Fail
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: HIGHPRIVILEGE_ESCALATIONREMOTE_CODE_EXECUTIONEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PRIVILEGE_ESCALATION]: The skill instructs the agent to use sudo for installing Tailscale and bringing the service online.
- [REMOTE_CODE_EXECUTION]: The skill uses a piped shell command to download and execute an installation script directly from Tailscale's official website (curl | sudo sh).
- [EXTERNAL_DOWNLOADS]: Fetches an installation script from a well-known service (Tailscale).
- [COMMAND_EXECUTION]: Executes system-level commands to manage network interfaces and services including tailscale status and tailscale up.
- [INDIRECT_PROMPT_INJECTION]: The skill processes external data from webhooks using placeholders like {{body.field}}. * Ingestion points: Webhook request bodies and headers processed by the task scheduler. * Boundary markers: Instructions explicitly state to treat request values as 'untrusted data' and 'outside data'. * Capability inventory: The resulting agent thread performs actions based on the rendered JSON fields. * Sanitization: Relies on the agent following the instruction to name specific fields rather than consuming raw blobs.
Recommendations
- AI detected serious security threats
Audit Metadata