The Agent Skills Directory

[COMMAND_EXECUTION]: The skill's main instruction file, SKILL.md, directs the agent to execute a local Python script save_file.py using command-line arguments to persist generated code to the disk.
[DATA_EXPOSURE]: The included script scripts/save_file.py implements an arbitrary file write capability. It accepts a --path and --content argument and writes the provided data to the specified path using os.makedirs and open().write(). This implementation lacks any path validation or directory sandboxing, allowing the agent (or a malicious user influencing the agent) to write files to sensitive locations such as shell profiles or configuration directories.
[REMOTE_CODE_EXECUTION]: The skill is designed to generate and save executable Python code. Because save_file.py can write to any path, it creates a risk where malicious code could be written to a location where it will be automatically executed by the system or another user, leading to a compromise of the host environment.

veadk-skills