The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The documentation in SKILL.md (Pattern 1) explicitly demonstrates an unsafe practice by sending a clear-text password to an SSH session using the tmx sk command.
[COMMAND_EXECUTION]: The skill requires or suggests the use of elevated privileges via sudo apt install tmux in references/setup.md. Additionally, it executes arbitrary, user-defined shell commands within persistent sessions using tmx create and tmx run.
[EXTERNAL_DOWNLOADS]: The skill recommends installing psmux from an unverified third-party GitHub repository (marlocarlo/psmux) via manual downloads, scoop, or cargo. It also refers to a Python CLI tool (tmx) from agent-sommelier-cli, which is not a verified service.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection through the tmx r command which reads the full scrollback/output of a session. If a remote server or a local process generates output containing malicious instructions, the agent may interpret and execute them.
Ingestion points: tmx r <name> in SKILL.md reads output from potentially untrusted terminal sessions.
Boundary markers: None identified; output is read directly into the agent context.
Capability inventory: The skill can execute shell commands, manage files, and establish SSH connections.
Sanitization: No evidence of output sanitization or filtering before processing session data.

tmux