The Agent Skills Directory

[COMMAND_EXECUTION]: The skill invokes local Python scripts (generate_hero.py, blog_render.py, blog_preflight.py) using python3 in Phase 5.5. These scripts take arguments derived from the blog content, such as titles, tags, and file slugs, which could serve as a vector for command injection if the agent does not properly sanitize these inputs before execution.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process untrusted data in the form of existing blog posts (MDX, markdown, HTML) to perform audits and rewrites.
Ingestion points: The workflow begins by reading the blog post specified by the user in Phase 1.
Boundary markers: There are no explicit instructions for the agent to use delimiters or "ignore" markers to distinguish between the skill's instructions and the content of the blog post being analyzed.
Capability inventory: The skill can execute shell commands via Python scripts, write to the file system (updating frontmatter and content), and perform network searches/downloads.
Sanitization: The skill performs specific checks on image URLs (rejecting javascript:, data:, and file:), but lacks broader sanitization for the text content used in prompts or script arguments.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to search for replacement statistics, YouTube videos, and stock images from providers such as Pixabay, Unsplash, Pexels, and Openverse. While these are reputable sources, the ingestion of external data remains a point of observation for security integrity.

blog-rewrite