The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from external URLs, which represents a potential indirect prompt injection surface.
Ingestion points: The skill fetches and interprets content from target websites, including robots.txt files, XML sitemaps, full HTML source, and browser accessibility trees using scripts like sitemap_discovery.py and agent_ux_check.py.
Boundary markers: The instructions do not define clear delimiters or "ignore instructions" warnings to encapsulate the fetched content, which could allow maliciously crafted content on a website to influence the agent's context.
Capability inventory: The agent has the ability to execute network-enabled scripts and render pages via the claude-seo environment tools.
Sanitization: No explicit sanitization or filtering of the fetched web data is described before the agent processes it for audit purposes.
[EXTERNAL_DOWNLOADS]: The skill documentation references the use of the Lighthouse tool via npx lighthouse@latest, which involves downloading and executing the package from the official npm registry, a well-known and trusted service.
[COMMAND_EXECUTION]: The skill utilizes a set of specialized utility scripts (sitemap_discovery.py, agent_ux_check.py, render_page.py, pagespeed_check.py, crux_history.py, and gsc_inspect.py) via the claude-seo run command to perform data collection and analysis tasks.

seo-technical