The Agent Skills Directory

[UNVERIFIABLE_DEPENDENCIES_AND_REMOTE_CODE_EXECUTION]: The skill directs users to reference external installation instructions from a third-party GitHub repository: https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md. Additionally, it requires the installation of external executable components using 'npx skills add' for 'belt-sh/cli' and various 'inference-sh' packages, which introduces unverified code into the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill handles user-provided text for audio generation by interpolating it into a shell command, which is a vector for indirect instruction injection. 1. Ingestion points: User-supplied strings within the 'segments' JSON object in the 'belt app run' command. 2. Boundary markers: Absent; there are no delimiters or instructions for the agent to treat the dialogue content as untrusted data. 3. Capability inventory: The skill utilizes the Bash tool to execute system commands. 4. Sanitization: None demonstrated; the examples show direct interpolation of text into a shell-executed JSON string, providing no protection against command breakout or injection via malicious characters.

elevenlabs-dialogue