The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill provides tools for reading external web content and performing automated browsing, which creates a surface for indirect prompt injection attacks.\n
Ingestion points: Tools like rottentomatoes_com__read_page (SKILL.md) and unbrowse.scrape (SKILL.md) fetch potentially untrusted content from rottentomatoes.com.\n
Boundary markers: The instructions lack explicit delimiters or warnings for the agent to distinguish between its instructions and the data fetched from the website.\n
Capability inventory: The agent can perform complex actions via unbrowse.run and unbrowse.browse.act, including autofill for authentication (SKILL.md).\n
Sanitization: There is no description of sanitization or filtering applied to the fetched content.\n- [DYNAMIC_EXECUTION]: The skill describes a process where automated browsing sessions can result in the server compiling new tools that are subsequently listed for the agent (SKILL.md). While this expands the agent's capabilities dynamically, the compilation and hosting occur on the vendor's infrastructure.\n- [EXTERNAL_DOWNLOADS]: The skill instructions (README.md) suggest adding the skill via npx skills add https://unbrowse.ai and connecting to a remote MCP server at https://unbrowse.ai/mcp/rottentomatoes.com. These resources originate from the vendor's domain.

rottentomatoes