content-digest

Pass

Audited by Gen Agent Trust Hub on Mar 4, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes shell commands to process data. It uses yt-dlp for YouTube subtitle extraction and a pipeline of Unix utilities (sed, tr, grep) for text cleaning. It also runs a local Python script to fetch social media content.
  • [EXTERNAL_DOWNLOADS]: The skill downloads content from external sources including YouTube, X/Twitter, and general webpages. It uses browser automation tools to navigate and retrieve text from dynamic websites.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of processing untrusted external content.
  • Ingestion points: Untrusted data enters the agent through YouTube transcripts, scraped webpage text, social media API responses, and local PDF documents.
  • Boundary markers: There are no explicit delimiters or instructions provided to the subagents to ignore or isolate potential commands embedded within the summarized content.
  • Capability inventory: The skill has access to subagent creation, file system writes in specific research directories, web search capabilities, and shell command execution.
  • Sanitization: Sanitization is restricted to removing formatting elements like timestamps and HTML tags, but it does not sanitize for instructional text that could influence the LLM's behavior.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 4, 2026, 11:03 AM