The Agent Skills Directory

[SAFE]: The skill is entirely composed of documentation and evaluation files. It does not contain any executable code, scripts (e.g., Python or JavaScript), or binary files. All instructions are intended for the AI to follow as a consultant or strategist.
[INDIRECT_PROMPT_INJECTION]: The skill defines a process for ingesting and optimizing user-provided content and queries. This creates a surface for indirect prompt injection if the content analyzed contains malicious instructions. However, because the skill lacks executable capabilities (such as shell access, file writing, or network requests), the risk of exploitation is nullified.
[DATA_EXPOSURE]: The skill refers to reading local context files like .agents/product-marketing.md and .claude/product-marketing.md. These are standard project-specific metadata files used for tailoring AI responses and do not constitute an exposure of sensitive system files or credentials.
[COMMAND_EXECUTION]: There are no shell commands or dynamic context injection patterns (!command) present in the instructions. Mentions of tools like curl or wget occur only in the context of documentation explaining how AI crawlers behave, not as commands for the agent to execute.

ai-seo