The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill provides a direct channel for untrusted data to influence agent actions.
Ingestion points: tmux capture-pane (documented in SKILL.md) is used to scrape output from terminal sessions, which may display content from untrusted files, web responses, or malicious command outputs.
Boundary markers: Absent. There are no mechanisms described to wrap or delimit captured terminal text to prevent the agent from interpreting embedded instructions as its own directives.
Capability inventory: The skill possesses high-privilege capabilities including tmux send-keys (arbitrary command execution in a shell) and session management (kill-server, kill-session).
Sanitization: Absent. Terminal output is processed as raw text, providing no protection against ANSI escape sequences or text-based injection attacks.
Command Execution (HIGH): The primary purpose of the skill is to facilitate shell command execution through tmux keystroke injection.
Evidence: Multiple examples in SKILL.md demonstrate sending shell commands, including pnpm install, python3, and arbitrary commands to coding agents like Codex using send-keys.

tmux