The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data from YouTube, creating a vulnerability to indirect prompt injection attacks.
Ingestion Points: The skill is designed to fetch and analyze external content including video transcripts (via execution/fetch_transcript.py and DataForSEO), viewer comments, and competitive metrics from the DataForSEO API. This data is used as direct context for generating strategies, scripts, and audits.
Boundary Markers: There are no explicit instructions or delimiters provided to the agent to treat this ingested content as untrusted or to ignore instructions embedded within the text.
Capability Inventory: The skill can execute shell commands through subprocess.run (in execution/fetch_transcript.py), perform network operations via API calls, and write files to the local disk for caching in the ~/.claude/.tmp/ directory.
Sanitization: Although the video_id is validated with a regular expression, the actual content of transcripts and comments is not filtered or sanitized for potential prompt injection patterns before being processed by the AI agent.- [COMMAND_EXECUTION]: The script execution/fetch_transcript.py executes an external binary to retrieve video data.
Evidence: The script uses subprocess.run() to call the yt-dlp tool. While it uses a list of arguments rather than a shell string, which mitigates standard injection risks, it still involves running an external process with parameters partially derived from user-provided inputs.

youtube