The Agent Skills Directory

[SAFE]: The skill serves as documentation for the belt CLI tool provided by inference.sh. No malicious behavior, obfuscation, or unauthorized access patterns were found.
[EXTERNAL_DOWNLOADS]: The skill references image assets and installation documentation from the official inference.sh domain and its public GitHub repository. These sources are considered legitimate for the tool's functionality.
[COMMAND_EXECUTION]: The skill instructions utilize the belt command-line utility. The execution is properly scoped within the allowed-tools section of the YAML frontmatter, limiting the agent's actions to the intended toolset.
[PROMPT_INJECTION]: Analyzed the skill for potential indirect prompt injection risks.
Ingestion points: External audio files provided via audio_url in the belt app run commands.
Boundary markers: Absent; transcribed content is not explicitly delimited.
Capability inventory: The skill utilizes Bash(belt *) for operation execution.
Sanitization: Not specified in the instructions. While a transcript could theoretically contain malicious instructions, this is an inherent property of speech-to-text tools and does not represent a unique vulnerability in the skill's design.

speech-to-text