The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to modify source files in specific local directories (e.g., /Users/martinpammesberger/Documents/psquared/psquared-websites/) when operating in 'fix' mode. While these paths are hardcoded to the author's environment, the ability to write to files like nuxt.config.js, astro.config.mjs, and various .vue or .html files constitutes a significant capability that must be monitored.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external websites using WebFetch to perform SEO analysis. This creates a surface for indirect prompt injection if the fetched content (meta tags, descriptions, sitemap data) contains instructions designed to manipulate the agent's behavior during the audit or fix phases.
Ingestion points: HTML content (meta tags, headings, images, JSON-LD), robots.txt, and sitemap.xml fetched from user-provided URLs.
Boundary markers: The skill does not explicitly define delimiters or instructions to ignore embedded commands in the fetched data.
Capability inventory: In 'fix' mode, the agent has the capability to write to and modify local configuration and source files based on the analysis of external data.
Sanitization: No explicit sanitization or filtering of the fetched content is described before it is used to determine what 'fixes' to apply.
[SAFE]: The skill uses WebFetch for its network operations, which is a standard tool for this type of task. The external references are directed at the author's own domains (psquared.dev, inboxmate.psquared.dev, etc.), which is consistent with its role as a developer tool.

seo