The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content (YouTube transcripts) which presents a surface for indirect prompt injection attacks.
Ingestion points: The script scripts/fetch_transcript.py fetches transcript data from external YouTube URLs provided by users.
Boundary markers: The workflow in SKILL.md does not specify the use of delimiters or specific instructions to ignore malicious commands embedded within the fetched transcript text.
Capability inventory: The skill uses a helper script to fetch data and then performs text transformation (summarization, formatting) on that data.
Sanitization: There is no evidence of sanitization or filtering of the transcript content to remove potential prompt injection sequences before processing by the agent.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of an external Python package to function.
Evidence: SKILL.md and scripts/fetch_transcript.py reference the youtube-transcript-api package, which is a widely used and well-known library for this purpose.

youtube-content