The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted user data which is interpolated into shell commands, creating a potential attack surface.
Ingestion points: The skill accepts user-defined strings for voice_script, prompt, voice_prompt, and video_prompt within the belt app run commands (SKILL.md).
Boundary markers: The instructions lack delimiters or explicit warnings to the agent to ignore embedded instructions in user data.
Capability inventory: The skill utilizes the Bash tool to execute the belt CLI (SKILL.md).
Sanitization: No sanitization or escaping of the user-provided scripts is demonstrated before being passed to the shell.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external CLI tool and references remote documentation.
Suggests installing the belt-sh/cli package via npx (SKILL.md).
References external installation instructions hosted at raw.githubusercontent.com/inference-sh/skills (SKILL.md).
[COMMAND_EXECUTION]: The skill is configured to execute shell commands using the Bash tool scoped to the belt binary.
The allowed-tools frontmatter grants access to Bash(belt *), allowing the agent to run various subcommands of the belt utility (SKILL.md).

ai-avatar-video