The Agent Skills Directory

[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests untrusted data from external websites and processes it within the agent's context.
Ingestion points: External content is retrieved via browser_navigate, browser_snapshot, and browser_evaluate from user-provided URLs. Specifically, the script extracts el.textContent from page elements to provide font samples.
Boundary markers: Absent. The instructions do not define clear delimiters or provide the agent with warnings to ignore instructions that might be embedded in the targeted website's text or metadata.
Capability inventory: The skill uses the Playwright MCP toolset (browser_navigate, browser_evaluate, browser_take_screenshot, browser_snapshot, browser_resize, browser_hover, browser_click) and can write the final report to the local file system (docs/design-references/DESIGN_REPORT.md).
Sanitization: Absent. Data extracted from the website (such as text samples or tech stack metadata) is directly included in the output report without escaping or validation, which could influence the agent if the website contains malicious instructions.

design-inspector