The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent on using the cmux send and cmux send-key commands, which allow for the execution of arbitrary commands within terminal surfaces managed by the cmux application.\n
Evidence: Examples such as cmux send --surface "${CMUX_SURFACE_ID:-}" "echo ok\n" in SKILL.md and references/commands.md demonstrate this capability.\n- [REMOTE_CODE_EXECUTION]: The CLI provides a set of commands for interacting with and executing code on virtual machines (VMs).\n
Evidence: The cmux vm exec <vm-id> -- <command> command listed in references/commands.md facilitates execution on remote or virtualized targets.\n- [PROMPT_INJECTION]: The skill features an attack surface for indirect prompt injection as it ingests untrusted data that could influence subsequent agent actions.\n
Ingestion points: Untrusted content can be read from terminal outputs and web pages using cmux read-screen, cmux capture-pane, and cmux browser snapshot (referenced in SKILL.md and references/commands.md).\n
Capability inventory: The agent possesses the ability to execute commands through cmux send and cmux vm exec (found in references/commands.md).\n
Boundary markers: The instructions advocate for using the --json output format for automation, which provides structure but does not act as a security boundary against embedded instructions.\n
Sanitization: There are no explicit instructions for sanitizing or escaping the content retrieved from these ingestion points before it is processed.

cmux-cli