The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes browser automation capabilities (Playwright and Chrome Extension) to interact with web pages. It uses commands such as browser_navigate, browser_type, browser_click, and Computer Use tools (left_click, type) to fill and submit forms on external websites.
[EXTERNAL_DOWNLOADS]: The agent is instructed to navigate to and extract data from over 190 external directory URLs listed in directories/DIRECTORIES.md. This includes navigation to various third-party domains, Google Forms, and Tally forms. It also visits arbitrary product URLs provided by the user in the workflows/setup-product.md workflow.
[PROMPT_INJECTION]: The skill presents an attack surface for indirect prompt injection (Category 8). \n
Ingestion points: The workflows/setup-product.md file instructs the agent to visit a user-provided URL and read the full page content to extract product information. \n
Boundary markers: There are no explicit instructions or delimiters defined to prevent the agent from following instructions that might be embedded within the content of the external websites it visits. \n
Capability inventory: The agent possesses high-privilege capabilities including browser navigation, form interaction, file uploads, and writing local markdown files to the profiles/ and tracking/ directories. \n
Sanitization: No sanitization or filtering logic is described for the content retrieved from external web pages before it is used to populate product profiles or influence agent behavior. Note: The impact is reduced by the requirement for user confirmation before any submission occurs.

directory-submitter