The Agent Skills Directory

[COMMAND_EXECUTION]: The skill implements functionality to send keystrokes and commands to tmux panes via tmux send-keys. The scripts find-sessions.sh and wait-for-text.sh use shell arrays and double-quoting for all variables (e.g., "$target", "$pattern"), which effectively prevents shell metacharacter injection.
[DATA_EXFILTRATION]: The skill has the capability to capture terminal output and scrollback history using tmux capture-pane. This is an intended feature for monitoring sessions, but users should be aware that any sensitive information (such as credentials or keys) printed to a monitored tmux pane will be visible to the agent.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The agent ingests external, untrusted data via tmux capture-pane in both the SKILL.md examples and the wait-for-text.sh script.
Boundary markers: There are no explicit boundary markers or instructions to the agent to ignore instructions embedded in the captured terminal output.
Capability inventory: The skill provides high-privilege capabilities including the ability to send arbitrary input to terminal sessions and execution of shell commands.
Sanitization: No sanitization or filtering of the captured terminal text is performed before it is processed by the agent.
[EXTERNAL_DOWNLOADS]: The skill metadata specifies the installation of the tmux binary via brew. This targets a well-known package manager and a standard utility, representing no significant security risk.

tmux