The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides patterns and documentation for the execute_shell_command tool. This tool allows the AI agent to execute arbitrary shell commands on the host environment, posing a high risk of system compromise or data destruction if the agent is improperly prompted or suffers from injection.
[REMOTE_CODE_EXECUTION]: The skill documentation and example scripts (e.g., README.md, references/agents.md) include boilerplate code to register execute_python_code. This capability enables the agent to run arbitrary Python logic, which can be used to perform sensitive file operations, network requests, or bypass environmental constraints.
[DATA_EXFILTRATION]: Because the skill enables shell and code execution, there is a secondary risk of data exfiltration. An agent with these tools could be prompted to read sensitive local files (like environment variables or SSH keys) and send them to a remote server using tools like curl or requests.
[INDIRECT_PROMPT_INJECTION]: The skill author explicitly identifies a risk surface for indirect prompt injection in SKILL.md. When using web-based tools (web_search, web_fetch), malicious content on a processed webpage could attempt to hijack the agent's logic to execute the high-privilege tools (shell/Python) enabled by this skill.

agentscope