The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses the shell to execute a local Python script (channel_research_pipeline.py) with arguments derived from user input. Specifically, it runs python3 channel_research_pipeline.py --channel "@channelhandle". While the input is quoted, there is a risk of command injection or unexpected behavior if the @channelhandle variable contains shell-special characters that the agent or the underlying script does not sanitize correctly.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant attack surface for indirect prompt injection. It ingests large amounts of untrusted data from external sources (YouTube channel metrics, video descriptions, and audience comments) and processes them into a narrative report.
Ingestion points: Reads analysis_summary.json, top_comments.json, and research_brief.md, which are populated with data from the YouTube API.
Boundary markers: The instructions do not specify any boundary markers or delimiters to help the AI distinguish between the metadata and potential instructions embedded in video descriptions or comments.
Capability inventory: The skill can execute shell commands (python3), read local files, and write new files (channel_analysis.md).
Sanitization: There is no evidence of sanitization or filtering for the external data before it is processed by the agent to write the 5,000-word analysis.
[DYNAMIC_EXECUTION]: The skill relies on executing an external Python script located at a path determined by an environment variable (CHANNEL_BREAKDOWN_PIPELINE_DIR) or a hardcoded fallback path. This allows the agent to run arbitrary logic defined in the external script, which is not bundled with the skill itself, creating a dependency on the integrity of the external environment.
[DATA_EXPOSURE]: The skill reveals a specific local user directory structure in its fallback path (/Users/nikhilbhansali/...), which may expose internal naming conventions or organizational details of the author's environment.

youtube-channel-breakdown