The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the ingestion of untrusted data from the web, creating a surface for indirect prompt injection attacks.
Ingestion points: Tools such as brevo_com__read_page and unbrowse.scrape allow the agent to fetch and process content from brevo.com and other external URLs.
Boundary markers: The skill instructions do not provide delimiters or specific directives to the agent to ignore or isolate instructions that may be hidden within the retrieved external content.
Capability inventory: The skill includes highly capable tools like unbrowse.run (which executes tasks in plain words) and a suite of cloud browser tools (unbrowse.browse.open, unbrowse.browse.act) that allow for automated interaction with websites. These could be misused if the agent follows instructions found in malicious external data.
Sanitization: There is no mention of sanitization, filtering, or validation processes for the data fetched from the remote host before it is presented to the agent.
[EXTERNAL_DOWNLOADS]: The skill relies on external infrastructure and installation scripts to function.
The skill is configured to connect to a remote MCP server at https://unbrowse.ai/mcp/brevo.com to provide its functionality.
Installation instructions include the use of npx to fetch and execute setup scripts from https://unbrowse.ai, which involves downloading and running code from a remote source.

brevo