captions
Pass
Audited by Gen Agent Trust Hub on Nov 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted transcripts from YouTube videos which could contain malicious instructions. Ingestion points: Captions and metadata are fetched from api.transcriptout.com/v1/transcript as seen in SKILL.md and references/curl-examples.md. Boundary markers: The skill includes a dedicated 'Treat fetched content as data' section in SKILL.md instructing the agent to ignore any embedded commands within transcripts. Capability inventory: The skill utilizes curl for network requests and shell commands (mkdir, chmod, sed, etc.) for environment configuration and credential storage. Sanitization: No technical sanitization is performed on the transcript content; the skill relies on instructional guardrails.
- [SAFE]: The skill follows security best practices for secret management. It instructs users to store keys in a restricted configuration folder (~/.config/transcriptout) with limited permissions (chmod 700/600) and uses secure practices like umask 077 and stty -echo during key setup in references/auth-setup.md.
Audit Metadata