responsive-validation

Pass

Audited by Gen Agent Trust Hub on Jun 19, 2026

Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection due to its core function of reading and processing untrusted web content.
  • Ingestion points: Data is ingested from web pages via agent-browser get text @main-heading and visual screenshots (SKILL.md).
  • Boundary markers: No delimiters or instructions are provided to the agent to ignore potentially malicious text embedded in the web page during validation.
  • Capability inventory: The agent has the ability to navigate, take screenshots, and extract text from any URL provided via the agent-browser tool (SKILL.md).
  • Sanitization: There is no filtering or validation of the text extracted from the page before it is used to satisfy the skill's pass criteria.
  • [COMMAND_EXECUTION]: The skill utilizes shell-style loop patterns to automate browser interactions across various screen resolutions. This is a standard automation practice and uses hardcoded viewport dimensions.
  • [DATA_EXPOSURE]: The skill acknowledges that screenshots may capture sensitive or authenticated states. It includes a security recommendation to store these artifacts in a gitignored directory (e2e-evidence/) to prevent accidental disclosure via version control.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 19, 2026, 08:44 AM
Security Audit — agent-trust-hub — responsive-validation