youtube-transcript
Pass
Audited by Gen Agent Trust Hub on Sep 18, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The
SKILL.mdfile contains an instructional override using emphatic language to control the agent's behavior. - Evidence: The directive "CRITICAL: YOU MUST NEVER MODIFY THE RETURNED TRANSCRIPT" uses high-priority markers to bypass standard agent processing logic.
- [INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from an external source (YouTube captions) and instructs the agent to process it (cleaning, formatting, and saving to files), creating a risk that instructions embedded in a video's transcript could be executed by the agent.
- Ingestion points:
scripts/get_transcript.pyfetches transcript snippets via theyoutube-transcript-api. - Boundary markers: No delimiters or "ignore instructions" warnings are provided to the agent when the transcript content is returned.
- Capability inventory: The skill instructions authorize the agent to write the extracted content to the local filesystem.
- Sanitization: The script performs no validation, filtering, or sanitization of the transcript text before it is returned to the agent context.
Audit Metadata