The Agent Skills Directory

[PROMPT_INJECTION]: The skill ingests and processes untrusted data from a user-specified <file-path> in Phase 1. This creates an indirect prompt injection surface where instructions or adversarial patterns embedded in the blog post could attempt to influence the agent's behavior during the audit, research, or rewrite phases.
Ingestion points: The content of the file provided as the <file-path> argument is read into the context.
Boundary markers: No explicit delimiters or instructions are provided to the model to ignore potential commands embedded within the processed blog post.
Capability inventory: The skill has the ability to execute shell commands (python scripts/...), perform network operations (searching Pixabay/Unsplash/YouTube), and read/write files in the blog directory.
Sanitization: No explicit sanitization or validation of the input content is mentioned before it is processed for variable extraction and rewriting.
[COMMAND_EXECUTION]: In Phase 5.5, the skill executes several local Python scripts (generate_hero.py, blog_render.py, blog_preflight.py) using subprocess calls. These scripts take arguments derived from the blog post's content, such as the title, tags, and slug. This represents a potential command injection risk if the agent extracts malicious shell metacharacters from the untrusted blog post and includes them in the command line strings.
[EXTERNAL_DOWNLOADS]: The skill conducts automated searches and verifies URLs on well-known services including Pixabay, Unsplash, and YouTube to source visual media and video embeds. These operations are legitimate and consistent with the skill's primary purpose of blog optimization.

blog-rewrite