The Agent Skills Directory

[COMMAND_EXECUTION]: The skill enables the agent to execute arbitrary shell commands in interactive sessions using tmux send-keys. This provides a persistent shell environment that the agent can manipulate.
[PROMPT_INJECTION]: The instructions explicitly direct the agent to bypass security and safety filters of other tools by recommending the use of flags like --dangerously-skip-permissions for Claude Code and --yolo or --full-auto for Codex.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by reading potentially untrusted data from terminal outputs without sanitization.
Ingestion points: Terminal pane history is scraped using tmux capture-pane in both SKILL.md and the scripts/wait-for-text.sh utility.
Boundary markers: There are no delimiters or warnings to ignore instructions embedded in the captured terminal output.
Capability inventory: The skill provides full shell command execution via tmux send-keys.
Sanitization: Captured output is processed as raw text without validation or filtering, allowing content from command outputs (like logs or files) to influence the agent's next actions.

tmux