The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from an external source, creating a potential surface for indirect prompt injection.
Ingestion points: The fastly_com__read_page tool defined in SKILL.md fetches title, readable text, and links from any path or URL on fastly.com.
Boundary markers: The instructions do not specify any delimiters or safety markers (such as "ignore embedded instructions") when processing content retrieved from the web.
Capability inventory: The agent is provided with powerful capabilities including unbrowse.run (which executes tasks described in plain words) and unbrowse.browse.act (which performs browser actions, including credential autofill from a manager).
Sanitization: There is no evidence of content filtering, escaping, or instruction sanitization for the data retrieved from external web pages.
[EXTERNAL_DOWNLOADS]: The skill relies on remote infrastructure provided by the vendor to execute its tools.
MCP Server: The skill connects to a remote streamable HTTP MCP server at https://unbrowse.ai/mcp/fastly.com (documented in server.json and SKILL.md).
Deployment: The README.md suggests using npx to fetch and install the skill from https://unbrowse.ai.

fastly