The Agent Skills Directory

REMOTE_CODE_EXECUTION (HIGH): The skill's 'Quick Start' section explicitly directs the execution of a remote script using the command curl -fsSL https://cli.inference.sh | sh. This 'curl-to-shell' pattern is a significant security risk as it executes unverified code from an external domain (inference.sh) that is not included in the trusted source list.
COMMAND_EXECUTION (MEDIUM): The skill requires the Bash tool with infsh * permissions. This allows the agent to execute any subcommand of the infsh CLI, which may have capabilities for file system access, network operations, or credential management (as seen in infsh login).
EXTERNAL_DOWNLOADS (LOW): The skill relies on software downloaded from a non-whitelisted external domain (inference.sh). While necessary for the skill's function, it introduces a dependency on an unverified third party.
PROMPT_INJECTION (LOW): The skill is vulnerable to Indirect Prompt Injection (Category 8).
Ingestion points: Audio and video URLs (audio_url, video_url) provided as input to the transcription models.
Boundary markers: Absent. There are no instructions to ignore spoken commands within the audio or delimiters to separate data from instructions.
Capability inventory: The skill has the ability to run shell commands (infsh) and pipe outputs to other applications.
Sanitization: None. The transcription results (which could contain malicious instructions from the audio) are directly processed and used in subsequent steps, such as subtitle generation or further CLI commands.

speech-to-text