The Agent Skills Directory

[COMMAND_EXECUTION] (SAFE): Uses curl to fetch the first 200 lines of a webpage and specific files like sitemap.xml and robots.txt. This behavior is the core intended functionality of the SEO tool.
[INDIRECT_PROMPT_INJECTION] (LOW): The skill ingests untrusted data from external websites via curl and browser automation (mcp__claude-in-chrome).
Ingestion points: curl -s "$URL" and navigate to URL commands in SKILL.md.
Boundary markers: None. The skill processes fetched text directly.
Capability inventory: Shell command execution (curl) and local script execution (generate_image.py).
Sanitization: No sanitization is performed on the fetched content before it is parsed by the agent. An attacker could theoretically embed instructions in meta tags to manipulate the SEO report.
[DYNAMIC_EXECUTION] (LOW): The skill references an external Python script at ~/.claude/skills/gemini-imagegen/scripts/generate_image.py. This is a cross-skill dependency used for the optional 'Quick Fix' of generating OG images. While this execution happens from a specific local path, it assumes the presence and safety of the secondary skill.
[EXTERNAL_DOWNLOADS] (SAFE): Suggests the installation of next-sitemap via pnpm, which is a standard, reputable package for Next.js SEO optimization.

seo-baseline