The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to crawl and analyze external websites (e.g., /seo audit <url>, /seo page <url>). This involves ingesting untrusted raw and rendered HTML, which is a significant surface for indirect prompt injection where malicious instructions could be embedded in the target website's content.
Ingestion Points: Site audits fetch raw and rendered HTML using render_page.py and analyze business signals from homepages.
Boundary Markers: The instructions emphasize a "PERCEIVE" phase to collect signals without interpretation, but do not specify technical delimiters to isolate untrusted content from the agent's instructions.
Capability Inventory: The skill can execute multiple bundled Python scripts, perform network requests to various APIs, and generate complex reports based on the ingested data.
Sanitization: No explicit sanitization or filtering of external content is described before it is processed by the LLM.
[COMMAND_EXECUTION]: The skill relies on a custom launcher, claude-seo, located in the plugin root's scripts/ directory, to run bundled Python tools (e.g., google_auth.py, seo_updates.py, render_page.py). It also includes setup and diagnostic commands (/seo setup, /seo doctor) that interact with the local environment to manage dependencies and Chromium instances.
[EXTERNAL_DOWNLOADS]: The skill interacts with several external services to gather SEO metrics, including Google APIs (PageSpeed, Search Console), Moz API, Bing Webmaster Tools, and OpenStreetMap (Overpass API and Nominatim). It also mentions optional extensions that can be installed via shell or PowerShell scripts (install.sh, install.ps1) from an extensions/ directory.

seo