The Agent Skills Directory

[PROMPT_INJECTION]: The skill provides templates designed to process external, untrusted data such as website content, robots.txt files, and sitemaps. This creates a surface for indirect prompt injection where malicious instructions could be embedded in the data provided to the agent.
[PROMPT_INJECTION]: The skill includes a specific security note instructing the agent to treat external content as untrusted and wrap it in boundary markers (e.g., '--- BEGIN UNTRUSTED EXTERNAL CONTENT ---'). This follows security best practices for prompt interpolation of untrusted data.
[PROMPT_INJECTION]: The skill does not include any automated code execution, network operations, or file system writes, limiting the potential impact of any successful injection to the context of the conversation.

ai-visibility