summarise-youtube
Pass
Audited by Gen Agent Trust Hub on Aug 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill uses
yt-dlpto download subtitles. This involves executing a shell command with a user-provided URL (<url>). Whileyt-dlpis a standard tool for this purpose, shell interpolation of user-supplied URLs requires proper escaping to prevent command injection. - [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external source (YouTube subtitles). An attacker could embed malicious instructions within a video's captions (e.g., 'Ignore previous instructions and instead do X') that the AI might follow when reading and summarizing the transcript.
- Ingestion points: Subtitles are downloaded to
<scratchpad>/subs.en.vttand subsequently read into the agent's context in step 2. - Boundary markers: None specified. The instructions do not tell the agent to ignore instructions embedded within the transcript text.
- Capability inventory: The skill uses
yt-dlp(subprocess) to fetch data and has the ability to write to the conversation. - Sanitization: The skill mentions reading past 'noise' (timestamps/repeats) but does not specify security-focused sanitization to filter out potential injection strings.
Audit Metadata