The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the Bash tool to execute system commands like ffmpeg and whisper-cli. There is a risk of shell command injection if a local filename contains malicious shell metacharacters (e.g., backticks or semicolons), as the filename is interpolated into the ffmpeg command string. While the agent is instructed to confirm the file, the underlying execution pattern remains vulnerable to crafted filenames.
[INDIRECT_PROMPT_INJECTION]: The skill processes transcription output, which is untrusted external data derived from audio/video content. This data is then fed back to the agent for linguistic correction and brand name checking.
Ingestion points: The output captured from whisper-cli is processed by the agent in Step 4.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the transcribed text during the correction phase.
Capability inventory: The skill possesses significant capabilities including Bash, Write, Edit, and Read tools.
Sanitization: No sanitization or filtering is applied to the transcription output before the agent processes it, meaning spoken instructions in a video could potentially influence the agent's behavior during the correction pass.

transcribe-video