The Agent Skills Directory

[COMMAND_EXECUTION] (HIGH): The skill's primary function is sending text to terminal sessions using iterm2.async_send_text, which allows the execution of any shell command available to the user, including destructive commands or unauthorized software installation.
[CREDENTIALS_UNSAFE] (HIGH): File examples/06-environment-vars.py explicitly demonstrates the extraction of sensitive environment variables, specifically referencing AWS_PROFILE and KUBECONFIG, which are high-value targets for attackers.
[REMOTE_CODE_EXECUTION] (HIGH): The skill instructions (SKILL.md) require the agent to generate Python scripts and execute them using uv run. This creates a direct execution path from AI-generated content to the host operating system.
[PROMPT_INJECTION] (HIGH): The skill has a high vulnerability surface for Indirect Prompt Injection (Category 8):
Ingestion points: External data processed during terminal automation or testing tasks.
Boundary markers: Absent; there are no instructions to use delimiters to separate trusted commands from untrusted data.
Capability inventory: Terminal command execution, Python file system access (open, os.remove), and environment variable reading.
Sanitization: Only basic shell quoting via shlex.quote is used for environment variables; no sanitization is applied to the logic or content of the terminal commands themselves.

iterm2-driver