The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes an attack surface for indirect prompt injection via the processing of untrusted web application interfaces.
Ingestion points: Uses browser_navigate and browser_snapshot in SKILL.md to load pages and parse their text/aria-tree content.
Boundary markers: Absent; the instructions do not establish explicit boundaries or direct the agent to ignore hidden prompts or text instructions embedded inside target web pages.
Capability inventory: Limited to browser interactions including browser_navigate, browser_snapshot, and browser_take_screenshot. No arbitrary code execution, file system modification, or network exfiltration capabilities are used.
Sanitization: Absent; text and layout properties from the webpage are parsed directly without content filtering.

responsive-testing