skills/adobe/skills/extract/Gen Agent Trust Hub

extract

Pass

Audited by Gen Agent Trust Hub on May 13, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill analyzes untrusted content from external websites (headings, text, metadata) using an LLM to identify page types and design modules, creating a risk of indirect prompt injection.
  • Ingestion points: HTML content is fetched from external URLs provided as input and stored in "stardust/current/pages/".
  • Boundary markers: There are no explicit delimiters or specific instructions to the LLM to ignore commands that may be embedded in the extracted site content.
  • Capability inventory: The agent has access to the host shell via "npx playwright" and the ability to write files to the local directory.
  • Sanitization: No sanitization or escaping of the crawled content is described before it is processed by the LLM.
  • [COMMAND_EXECUTION]: The skill uses the shell to execute "npx playwright" for browser automation. This is a core functionality required to crawl and render the target websites.
  • [EXTERNAL_DOWNLOADS]: The skill downloads brand assets such as images, logos, and font files from the target website's origin to the local "stardust/current/assets/" directory. This is expected behavior for its stated purpose.
Audit Metadata
Risk Level
SAFE
Analyzed
May 13, 2026, 05:46 PM
Security Audit — agent-trust-hub — extract