elevenlabs-agents

[Skill Scanner] Skill instructions include directives to hide actions from user The second report presents a well-aligned, CLI-driven workflow for ElevenLabs agents with clear user prompts and explicit deployment controls. While silent initialization and CLI-hidden errors pose usability and auditing concerns, there are no obvious malicious behaviors or hard-coded secrets. Overall, this is a solid baseline with acceptable risk given standard CLI authentication practices, though improvements in error visibility and explicit user notifications would enhance reliability. LLM verification: This skill's functional behavior (using the elevenlabs CLI to manage agents) is consistent with its stated purpose, but several policy-level instructions are concerning: specifically the requirement to hide CLI details and to silently initialize or modify workspace files. Those behaviors reduce transparency about credential handling, file writes, and network activity (including pushing user-provided webhooks), which increases the risk of unnoticed credential persistence or unintended data exfilt