The Agent Skills Directory

[COMMAND_EXECUTION]: The transcribe.js script executes system binaries such as whisper, ffmpeg, and python3 using spawnSync and execSync. This is necessary for verifying dependencies and performing the actual audio transcription locally.
[EXTERNAL_DOWNLOADS]: The documentation and skill metadata guide the user to install external dependencies including the openai-whisper Python package from PyPI and the ffmpeg multimedia framework from standard system package managers (e.g., apt, brew, pacman).
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection (Category 8c) because it processes untrusted audio data and converts it into text that is then fed back to the AI agent.
Ingestion points: Audio files (WAV, MP3, M4A, FLAC, OGG) provided as command-line arguments to transcribe.js in SKILL.md.
Boundary markers: None; the transcription output is returned as plain text without delimiters or instructions to the agent to ignore embedded commands.
Capability inventory: The skill is intended to be used within the OpenClaw agent environment, which may have significant system capabilities (file system access, terminal access).
Sanitization: No sanitization or filtering is performed on the transcribed text before it is returned to the agent context.

local-whisper