The Agent Skills Directory

[EXTERNAL_DOWNLOADS] (HIGH): The skill documentation in SKILL.md instructs the agent to configure an MCP server using npx -y next-devtools-mcp@latest. This involves downloading and executing a package from an untrusted author, which poses a supply chain risk.
[REMOTE_CODE_EXECUTION] (HIGH): The setup instructions for the MCP server create a direct path for remote code execution. An agent following these steps would run code from the npm registry at runtime without any prior security validation or human-in-the-loop approval.
[COMMAND_EXECUTION] (MEDIUM): The file upload template in references/server-actions.md is vulnerable to path traversal. The implementation uses path.join with a user-controlled file.name property without sanitization, which could allow an attacker to overwrite sensitive files outside the intended upload directory, such as .env or package.json.
[PROMPT_INJECTION] (HIGH): The skill defines an interface for coding agents to ingest application logs and build errors via tools like get_logs and get_errors. This creates a significant surface for Indirect Prompt Injection. An attacker could trigger specific error messages or log entries (e.g., via a crafted web request) to manipulate the agent's behavior, which is particularly dangerous given the agent likely has file-write and command-execution capabilities.

nextjs