The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes natural language prompts provided by the user which are interpolated into shell command arguments. This creates a surface where maliciously crafted input could attempt to influence the execution of the underlying CLI tool.
Ingestion points: The prompt field within the JSON payload for the belt app run command found throughout SKILL.md.
Boundary markers: Input text is delimited by JSON property quotes and passed as part of a single-quoted shell argument.
Capability inventory: The skill utilizes the Bash tool to execute the belt CLI utility.
Sanitization: The instructions do not define any sanitization, filtering, or validation logic for the input prompts before they are processed.
[EXTERNAL_DOWNLOADS]: The skill instructions guide the user to download installation scripts and add prerequisite skill components from external sources associated with the vendor.
Evidence: References to https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md for CLI setup and the installation of the belt-sh/cli component.
[COMMAND_EXECUTION]: The skill is primarily configured to operate via the belt CLI tool using the Bash execution environment, as limited by the allowed-tools metadata in the frontmatter.

dialogue-audio