thumbnail

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the agent to search for and download external assets, such as company logos and YouTube thumbnails, from the internet.
  • Evidence: "If a company logo is not locally available, you can search for it online and download it using curl" (SKILL.md)
  • Evidence: "Fetches thumbnails from YouTube's image servers for style transfer reference" (SKILL.md)
  • [COMMAND_EXECUTION]: The skill directs the agent to execute shell commands to manage external assets.
  • Evidence: "you can search for it online and download it using curl, then pass the absolute path to the downloaded image in the prompt." (SKILL.md)
  • [DATA_EXFILTRATION]: The skill instructions include accessing an internal agent directory which may contain other skill configurations and logic.
  • Evidence: "check if a brand guidelines skill exists at ~/.claude/skills/ and apply any relevant design constraints." (SKILL.md)
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted visual data from external websites and instructs the agent to analyze it, creating a surface for embedded instructions to influence the agent.
  • Ingestion points: Images downloaded via curl from YouTube and general web searches.
  • Boundary markers: None specified for the analysis of downloaded image content.
  • Capability inventory: Shell command execution (curl), file system writes (./downloads/), file system reads (~/.claude/skills/), and image model invocation (creator-stack:nanobanana).
  • Sanitization: No sanitization or filtering logic is described for the content of the analyzed images.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 11:56 AM