od-web-effect-extractor

Warn

Audited by Gen Agent Trust Hub on Nov 11, 2026

Risk Level: MEDIUMINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted content from user-provided URLs, creating an attack surface where a malicious website could influence the agent's behavior. \n
  • Ingestion points: The agent captures data from a website's DOM, JavaScript bundles, and network requests as part of the inspection workflow (SKILL.md, Step 2). \n
  • Boundary markers: No explicit delimiters or specialized safety instructions are provided to the agent to prevent it from obeying instructions that might be embedded in the scraped HTML or JavaScript comments. \n
  • Capability inventory: The skill utilizes network access for content capture and fs:write to generate the final artifact, which could be abused if malicious instructions are successfully injected. \n
  • Sanitization: The instructions include a mandate to exclude private data, auth flows, and business logic, providing some risk reduction by narrowing the scope of what is extracted. \n- [DYNAMIC_EXECUTION]: The skill generates and saves new executable code based on scraped data, which is intended to be executed in a preview or developer environment. \n
  • Script generation: The agent creates standalone HTML/JS files and GLSL shaders by translating observations from a source website into a new implementation (SKILL.md, Step 4). \n
  • Risk: Malicious code or shaders from a target site could be reflected into the generated artifact, potentially leading to the delivery of malicious code under the guise of a recreated visual effect.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 11, 2026, 01:08 AM