html-to-interaction-prompts

Pass

Audited by Gen Agent Trust Hub on Jul 28, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PROMPT_INJECTION]: The skill processes untrusted HTML, CSS, and JavaScript from external URLs or user-provided files, creating a surface for indirect prompt injection. An attacker could embed malicious instructions in code comments or page content to influence agent behavior.
  • Ingestion points: The agent reads HTML/CSS/JS files and fetches live website content via the WebFetch tool.
  • Boundary markers: Absent. The skill instructions do not define delimiters or specific guidance to ignore instructions embedded within the ingested data.
  • Capability inventory: The skill has access to file modification (Write, Edit), shell execution (Bash), and network requests (WebFetch).
  • Sanitization: Absent. The instructions do not direct the agent to filter or sanitize the ingested code before analysis.
  • [COMMAND_EXECUTION]: The skill uses the Bash tool to execute system commands such as ffmpeg, ffprobe, and git. While these are legitimate for verifying media and managing assets, providing an agent with shell access is a powerful capability that must be monitored.
  • [EXTERNAL_DOWNLOADS]: The skill utilizes WebFetch to retrieve data from live websites. This interaction with untrusted remote resources is a necessary part of the skill's purpose but represents a network attack surface.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 28, 2026, 03:02 AM
Security Audit — agent-trust-hub — html-to-interaction-prompts