The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill exposes the agent to untrusted data by reading terminal screens and web content.
Ingestion points: The muxy read-screen command (SKILL.md) and the muxy browser suite including read, snapshot, and eval (SKILL.md) ingest external content into the agent context.
Boundary markers: There are no instructions for using delimiters or boundary markers to isolate ingested terminal or browser data from the agent's instructions.
Capability inventory: The skill can execute shell commands (muxy send), run JavaScript (muxy browser eval), and modify global environment settings (muxy install-skills).
Sanitization: No sanitization or escaping mechanisms are described for processing external data before it is interpreted by the agent.
[COMMAND_EXECUTION]: The skill facilitates the execution of arbitrary shell commands in other terminal panes using muxy send and muxy send-keys.
[DYNAMIC_EXECUTION]: The muxy browser eval command allows for the execution of arbitrary JavaScript code within browser tabs.
[DATA_EXFILTRATION]: The skill provides access to potentially sensitive data through muxy browser cookies get, which retrieves session cookies, and muxy read-screen, which can capture credentials or private data visible in terminal output.
[PERSISTENCE]: The muxy install-skills command modifies the global configuration of various AI coding harnesses (such as Claude Code and Cursor) to ensure the skills remain active across the system and future sessions.

muxy-cli