The Agent Skills Directory

[COMMAND_EXECUTION]: The skill enables the agent to send arbitrary keystrokes and commands to any tmux pane using tmux send-keys. While this is the intended functionality for session control, it allows for arbitrary code execution within the terminal environment managed by the agent.- [DATA_EXFILTRATION]: The skill can read the entire scrollback history of terminal sessions via tmux capture-pane. This capability introduces a risk of sensitive data exposure if credentials, API keys, or private files were previously accessed or displayed in the terminal session.- [PROMPT_INJECTION]: The skill presents a surface for indirect prompt injection by ingesting untrusted data from terminal output.
Ingestion points: Untrusted terminal data is read through tmux capture-pane in SKILL.md and the scripts/wait-for-text.sh script.
Boundary markers: No delimiters or explicit instructions are provided to the agent to distinguish between terminal output and legitimate instructions.
Capability inventory: The skill possesses extensive control over the terminal environment, including sending input (tmux send-keys), managing sessions (tmux new-session, tmux kill-session), and navigating windows.
Sanitization: The skill does not perform any sanitization, filtering, or escaping of terminal output before presenting it to the agent.

tmux