The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The instructions direct the agent to execute a command string ('data.next_command') returned by the hosted API ('api.luxin.sh'). This creates a critical vulnerability where the remote service could execute arbitrary shell commands on the host system by providing them in the API response.
[EXTERNAL_DOWNLOADS]: The skill frequently uses 'npx -y luxin-cli@latest' to download and run the latest version of a third-party CLI tool. Executing unpinned remote code without integrity verification is a security risk that exposes the environment to supply chain attacks if the package or its registry is compromised.
[COMMAND_EXECUTION]: The skill provides bash commands that modify sensitive environment variables ('HOME', 'CODEX_HOME') and create directories to redirect agent state. While intended for isolation, these actions manipulate the execution environment and can be used to bypass directory-based security controls.
[PROMPT_INJECTION]: The skill processes untrusted user data (prompts) that are interpolated directly into shell commands and sent to an external API without sanitization. * Ingestion points: User-supplied prompts in SKILL.md templates. * Boundary markers: Absent; no delimiters or instructions are provided to the agent to ignore embedded instructions in the prompt. * Capability inventory: Shell command execution via npx and network access to api.luxin.sh. * Sanitization: None; the skill does not define any validation, escaping, or filtering for external content.

ai-audio-generation