The Agent Skills Directory

EXTERNAL_DOWNLOADS (MEDIUM): The script fetch.py executes npx -y @smithery/mcp-fetch. This command automatically downloads and runs the latest version of an external package from the npm registry every time the skill is initialized. This introduces a supply-chain risk as the external package source is not a pre-approved trusted repository and could be compromised.
COMMAND_EXECUTION (MEDIUM): The skill uses subprocess.Popen to launch a shell-mediated command (npx). While the arguments provided to the fetch tool itself are structured via JSON-RPC, the reliance on system-level binaries (npx) to fetch remote code poses a medium risk.
PROMPT_INJECTION (LOW): As a Category 8 Indirect Prompt Injection surface, the skill is designed to fetch arbitrary web content and print it to standard output.
Ingestion points: fetch.py reads content from user-provided URLs via @smithery/mcp-fetch.
Boundary markers: None identified. The content is printed directly as text.
Capability inventory: subprocess.Popen is used to launch the MCP server.
Sanitization: None. The script directly prints item.get('text') from the external tool output without validation or escaping, allowing malicious instructions from web pages to enter the Agent's context.

fetch