The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external CLI tool and other platform-specific skills using commands like npx skills add belt-sh/cli and npx skills add inference-sh/skills@.... It also references a remote installation script at https://raw.githubusercontent.com/inference-sh/skills/refs/heads/main/cli-install.md. These dependencies are necessary for the skill's core functionality of interacting with the inference.sh service.
[COMMAND_EXECUTION]: The skill executes shell commands via the belt CLI (e.g., belt login, belt app run). The execution environment is restricted by the allowed-tools configuration to only allow commands starting with belt, which mitigates the risk of arbitrary command injection.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external audio and video files, creating a potential vector for indirect prompt injection if the transcribed content contains instructions for the AI.
Ingestion points: The skill accepts audio and video_url parameters in JSON inputs to fetch external media for transcription.
Boundary markers: No explicit delimiters or instructions are provided in the examples to warn the agent about ignoring embedded instructions within the transcribed text.
Capability inventory: The skill uses the belt CLI to perform transcription and can pipe output to other tools (e.g., infsh/caption-videos), which could potentially be influenced by malicious content in the audio.
Sanitization: There is no evidence of sanitization or filtering applied to the transcription output before it is returned to the agent or used in downstream tasks.

elevenlabs-stt