The Agent Skills Directory

[COMMAND_EXECUTION]: The tabz-last shell function in references/cli-websocket.md captures the most recent command from the user's history and re-executes it ($($cmd)) to capture output. This can lead to unintended side effects if the command is not idempotent.
[CREDENTIALS_UNSAFE]: The skill's integration workflow depends on reading a sensitive authentication token from a temporary file location (/tmp/tabz-auth-token), which may be accessible to other users on the same system.
[DATA_EXFILTRATION]: Automated functions such as tabz-explain and tabz-last capture current terminal session output or tmux pane content and transmit it to a local service, potentially exposing sensitive information from the terminal buffer.
[PROMPT_INJECTION]: Documentation examples in references/spawn-api.md explicitly instruct the use of flags like --dangerously-skip-permissions, which are designed to bypass the safety and permission filters of the target agent.
[COMMAND_EXECUTION]: The skill provides instructions for users to persist utility functions by modifying shell configuration files such as .bashrc or .zshrc.

tabz-integration