The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes content from YouTube, which is a source of untrusted data. An attacker could embed malicious instructions in a video's transcript or description to influence the agent's behavior during the analysis phase.
Ingestion points: scripts/fetch_youtube.py fetches transcripts, titles, and descriptions from external YouTube URLs.
Boundary markers: The analysis templates in references/analysis-modes.md and the workflow in SKILL.md do not employ explicit delimiters (e.g., XML tags or unique markers) to clearly separate the fetched transcript content from the agent's internal instructions.
Capability inventory: The skill allows the agent to perform network requests to YouTube, write analysis results to the local filesystem (~/.claude/knowledge), and execute local Python scripts.
Sanitization: No sanitization or filtering is performed on the transcript text or metadata before it is presented to the agent for analysis.
[COMMAND_EXECUTION]: The skill workflow in SKILL.md and the test suite in scripts/test_fetch.py rely on executing shell commands via uv run and subprocess.run to interact with the Python scripts. While these operations target local, skill-provided scripts, they represent a significant capability surface.
[EXTERNAL_DOWNLOADS]: The skill utilizes the well-known libraries yt-dlp and youtube-transcript-api to fetch video data from YouTube's official services. These are standard dependencies for the skill's stated purpose.

youtube-content