The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill instructions include a one-liner that downloads a remote shell script from GitHub and pipes it directly into bash for execution.
Evidence: curl -fsSL https://raw.githubusercontent.com/Dicklesworthstone/misc_coding_agent_tips_and_scripts/main/wezterm-mux-tune.sh | bash found in SKILL.md and references/PERFORMANCE-TUNING.md.
Risk: This execution pattern allows the remote source to execute arbitrary commands on the local system without verification or version pinning.
[PRIVILEGE_ESCALATION]: The skill directs the user to perform actions that weaken system security and require root access.
Evidence: sudo sysctl -w kernel.yama.ptrace_scope=0 (SKILL.md). Disabling ptrace_scope removes a key security layer that prevents processes from attaching to and inspecting other running processes.
Software Installation: Instructs the use of sudo apt-get install to install software like reptyr and wezterm (references/PERSISTENT-SESSIONS.md).
[PERSISTENCE]: The skill provides configuration for maintaining persistent access and background processes across user sessions.
Evidence: Creation of a systemd user service (wezterm-mux-server.service) and the use of sudo loginctl enable-linger $USER to ensure the user's processes continue running after logout (references/PERSISTENT-SESSIONS.md).
[EXTERNAL_DOWNLOADS]: The skill downloads external resources and configures third-party package repositories.
Evidence: Fetches a GPG key from https://apt.fury.io/wez/gpg.key to configure an external APT repository for WezTerm (references/PERSISTENT-SESSIONS.md).
[COMMAND_EXECUTION]: The skill relies on the wezterm cli to programmatically interact with the terminal environment, including injecting text into active panes.
Evidence: Extensive documentation of wezterm cli send-text, wezterm cli spawn, and wezterm cli split-pane used to control session behavior (references/COMMANDS.md).
[INDIRECT_PROMPT_INJECTION]: The skill is designed to manage environments where multiple AI agents produce high-volume output, creating a vulnerability surface for indirect injection.
Ingestion points: Processes continuous output streams from multiple agents such as Claude Code and Codex (SKILL.md).
Boundary markers: There are no specified delimiters or boundary markers to isolate agent output from the management context.
Capability inventory: The skill has full terminal control capabilities, including the ability to split panes and inject text/commands via CLI.
Sanitization: No sanitization or filtering of ingested agent output is implemented.

wezterm