news-digest-video-agent

Pass

Audited by Gen Agent Trust Hub on Apr 14, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted data from news articles and social media profiles to generate narrated scripts.\n
  • Ingestion points: The agent uses browser-use to extract text from news sites (BBC, AP News, etc.) and X.com profiles (detailed in AGENTS.md).\n
  • Boundary markers: There are no explicit delimiters or instructions provided to the agent to separate untrusted web content from its core instructions.\n
  • Capability inventory: The skill utilizes shell command execution for media processing (ffmpeg) and network access via the videodb SDK to upload content.\n
  • Sanitization: No sanitization or filtering of the retrieved web content is specified before the agent processes it for script writing.\n- [COMMAND_EXECUTION]: Potential for shell command injection in media processing instructions.\n
  • The instructions in AGENTS.md (Step 5) direct the agent to execute ffmpeg commands using a file path (outputs/<topic-slug>/...) that incorporates a variable (topic-slug) derived from user-provided input. If the agent does not properly sanitize the topic before generating the slug, it could lead to arbitrary command execution.\n- [EXTERNAL_DOWNLOADS]: Fetches external skill configuration during setup.\n
  • The setup instructions in SKILL.md and README.md download the browser-use skill definition from its official GitHub repository (github.com/browser-use/browser-use).
Audit Metadata
Risk Level
SAFE
Analyzed
Apr 14, 2026, 03:14 PM