The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the commandline tool implementation (github.com/cloudwego/eino-ext/components/tool/commandline) which enables an AI agent to execute arbitrary shell commands on the host system. This is a high-risk capability that should only be enabled in isolated environments.
[REMOTE_CODE_EXECUTION]: Documentation for the MCP (Model Context Protocol) component includes examples of using npx to dynamically download and execute external MCP servers (e.g., npx -y @modelcontextprotocol/server-xxx), which constitutes remote code execution of unverified packages.
[DATA_EXFILTRATION]: The skill provides implementations for an httprequest utility tool and various search tools (Google, Bing, DuckDuckGo). These components allow for outbound network requests which could be leveraged to exfiltrate sensitive information retrieved through the documented document loaders (S3, local files) or environment variables.
[PROMPT_INJECTION]: The skill's prompt template reference (reference/prompt.md) demonstrates the interpolation of raw user input into system and user messages without illustrating defensive techniques like delimiters or sanitization. This establishes a surface for indirect prompt injection where malicious instructions in processed documents could influence agent behavior.

eino-component