The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and processes external content from mariadb.org, which could potentially contain malicious instructions designed to manipulate the agent.
Ingestion points: Data is ingested through mariadb_org__get_search and mariadb_org__read_page tools, which return readable text and links from external web pages (SKILL.md).
Boundary markers: The instructions do not define specific delimiters or instructions for the agent to ignore potentially malicious content embedded in the fetched data.
Capability inventory: The skill includes powerful capabilities such as unbrowse.run for executing tasks and unbrowse.browse.act for browser actions including form autofilling (SKILL.md).
Sanitization: There is no mention of sanitization, filtering, or validation performed on the external content before it is presented to the agent's context.
[EXTERNAL_DOWNLOADS]: The skill relies on a remote Model Context Protocol (MCP) server for its core functionality.
The skill connects to a remote endpoint at https://unbrowse.ai/mcp/mariadb.org using the streamable-http transport to access tool definitions and execution (server.json).
The documentation provides instructions for users to set up this remote connection in their local agent environment (README.md, SKILL.md).

mariadb