The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The skill is vulnerable to indirect prompt injection as it ingests and processes untrusted YouTube transcripts.
Ingestion points: Untrusted transcript text enters the agent context through scripts/extract-transcript.py.
Boundary markers: There are no visible delimiters or safety instructions in the provided scripts to prevent the LLM from obeying instructions embedded in the transcripts.
Capability inventory: The extract-transcript.py script utilizes the youtube-transcript-api for network data retrieval.
Sanitization: No text sanitization or filtering is performed on the extracted transcripts.
[EXTERNAL_DOWNLOADS] (LOW): The skill installs a required third-party Python package from an external registry.
Evidence: scripts/install-dependencies.sh uses pip to install youtube-transcript-api.
Context: The package is well-known and essential for the skill's primary function, but represents a standard external dependency risk.

youtube-summarizer