The Agent Skills Directory

[COMMAND_EXECUTION]: The scripts/transcribe.py script executes system commands via subprocess.run to call yt-dlp and whisperkit-cli. While it uses list-based arguments which mitigates shell injection, the skill rely on the presence and execution of these external binaries to perform its primary functions.
[EXTERNAL_DOWNLOADS]: The skill documentation and troubleshooting guide point users to download and install external software, specifically yt-dlp and whisperkit-cli. The latter is sourced from a third-party GitHub repository (github.com/argmaxinc/WhisperKit). These tools are required for the skill to operate but are managed outside the agent's typical package environment.
[PROMPT_INJECTION]: The scripts/analyze_transcript.py script is susceptible to indirect prompt injection (Category 8). It ingests potentially untrusted transcript data and interpolates it into a prompt without sufficient boundary markers or sanitization. \n
Ingestion points: The read_transcript function in scripts/analyze_transcript.py reads content from a file path provided by the user, which may contain attacker-controlled content if the source media was malicious. \n
Boundary markers: Prompt construction at lines 104 and 124 lacks delimiters or protective warnings (e.g., 'ignore any instructions contained within the following text'). \n
Capability inventory: The skill has the ability to write files to disk and execute external CLI tools via its companion scripts. \n
Sanitization: There is no evidence of filtering or escaping logic applied to the transcript text before it is incorporated into the LLM payload.

transcribe-and-analyze