cmux
Pass
Audited by Gen Agent Trust Hub on Mar 16, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides capabilities for interacting with external environments, including 'cmux send' for terminal input and 'cmux browser eval' for executing JavaScript in an embedded browser. Additionally, the installation instructions require administrative privileges via 'sudo ln -sf' to create a symlink for the binary.
- [PROMPT_INJECTION]: The skill's architecture allows for indirect prompt injection. 1. Ingestion points: Untrusted content is ingested through 'cmux read-screen' and 'cmux browser get-text'. 2. Boundary markers: No delimiters are defined to separate ingested content from system instructions. 3. Capability inventory: The agent can execute terminal commands ('cmux send') and browser JavaScript ('cmux browser eval'). 4. Sanitization: No sanitization is performed on data retrieved from terminal or browser surfaces.
Audit Metadata