The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill directs users to install an external CLI utility using the command npx skills add belt-sh/cli. It also references an external installation guide and script hosted on a public GitHub repository at https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md which is not a pre-verified trusted source.
[REMOTE_CODE_EXECUTION]: The skill facilitates the execution of remote application logic through the belt app run command. It triggers various remote models and utilities (e.g., falai/dia-tts, infsh/media-merger) hosted on the inference.sh platform.
[COMMAND_EXECUTION]: The skill's configuration explicitly requests broad shell execution capabilities for the belt CLI tool via allowed-tools: Bash(belt *). This allows the agent to perform administrative actions such as belt login and run arbitrary remote apps.
[PROMPT_INJECTION]: The skill possesses a surface for indirect prompt injection by interpolating user-provided text directly into CLI command arguments.
Ingestion points: The prompt field within the JSON input for belt app run commands found in SKILL.md.
Boundary markers: The text is wrapped in JSON quotes, but there are no instructions or delimiters to prevent the underlying synthesis model or CLI from being influenced by adversarial content within the dialogue text.
Capability inventory: The skill utilizes the Bash tool to execute belt commands with potentially unsanitized inputs.
Sanitization: No evidence of input validation, escaping, or filtering is present in the skill's instructions.

dialogue-audio