The Agent Skills Directory

[COMMAND_EXECUTION]: The skill relies on the cmux CLI to perform various terminal operations, including creating workspaces, splitting panes, and focusing specific terminal areas. It specifically uses cmux send to inject and execute shell commands inside these panes, such as sourcing environment files and launching the hf run process.
[PROMPT_INJECTION]: The skill uses cmux read-screen to capture output from the agent dashboard and orchestrator logs into the agent's context. This creates an indirect prompt injection surface where untrusted data (such as logs or task statuses from the swarm) could contain embedded instructions. The ingestion occurs in SKILL.md without explicit boundary markers or sanitization, while the agent possesses the capability to execute further shell commands via the cmux tool chain.
[CREDENTIALS_UNSAFE]: The skill references the management of sensitive environment variables like DASHSCOPE_API_KEY, INCEPTION_API_KEY, and ZAI_API_KEY. However, it adheres to safe practices by instructing the user to store these in a .env file and sourcing them locally rather than hardcoding or transmitting them.

hf-cmux