The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The calculator tool implemented in SKILL.md and references/agents.md uses the eval() function on input strings. This allows for the execution of arbitrary Python code and is a major security risk if input is untrusted.\n- [COMMAND_EXECUTION]: The references/integration.md file demonstrates the ShellTool, which enables an agent to run arbitrary commands directly in the system shell.\n- [REMOTE_CODE_EXECUTION]: The PythonREPLTool shown in references/integration.md allows the agent to execute arbitrary Python scripts in a REPL environment.\n- [REMOTE_CODE_EXECUTION]: The FAISS vector store loading example in references/integration.md explicitly uses allow_dangerous_deserialization=True, enabling potential code execution via malicious pickle files.\n- [PROMPT_INJECTION]: The risky_api_call tool in references/agents.md accepts a user-provided URL endpoint, creating an attack surface for Indirect Prompt Injection and Server-Side Request Forgery (SSRF) attacks.

langchain