The Agent Skills Directory

[COMMAND_EXECUTION]: The skill is designed to execute arbitrary shell commands within a tmux session using tmux send-keys. This gives the agent full control over the user's terminal environment.
[REMOTE_CODE_EXECUTION]: The skill supports executing commands on remote systems by wrapping tmux commands in ssh "$machine". This extends the agent's reach to any server the user has access to.
[DYNAMIC_EXECUTION]: It employs the eval "$(echo $cmd_b64 | base64 -d)" pattern to execute commands. Using eval on dynamically decoded content is a high-risk practice that can be exploited if the input is not strictly controlled.
[OBFUSCATION]: The use of base64 encoding to transport commands, while intended to prevent shell escaping issues, effectively obfuscates the command payload from direct observation during the execution step.
[DATA_EXFILTRATION]: The tmux capture-pane -pt "$session_name" command allows the agent to read the entire history and current contents of the terminal screen. This can expose sensitive information such as environment variables, passwords, or private keys previously printed to the terminal.
[INDIRECT_PROMPT_INJECTION]: By reading terminal output, the agent is exposed to untrusted data. If a file or website displayed in the terminal contains malicious instructions, the agent could ingest and follow them, leading to unauthorized actions.
[REMOTE_CODE_EXECUTION]: The skill uses scp to upload and execute temporary scripts on remote machines, which is a powerful mechanism for multi-stage command execution.

ttttt