cmux-agent-role
Pass
Audited by Gen Agent Trust Hub on Apr 12, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill defines behavioral standards and communication protocols for agents working within a project-specific framework.
- [COMMAND_EXECUTION]: The agent uses the cmux-team CLI and standard shell utilities like ls, grep, and tail to manage tasks and monitor system status within the local project directory. These operations are part of the intended multi-agent workflow.
- [DATA_EXFILTRATION]: All inter-agent communication is restricted to local files in the shared .team/ directory, with no network-based operations or external data transfer instructions identified.
- [PROMPT_INJECTION]: The skill identifies a surface for Indirect Prompt Injection inherent to its multi-agent design. Ingestion points: Data from files in the .team/ directory, such as other-role.md and task definitions, are read into the agent context. Boundary markers: Output structure follows specific Markdown templates defined in the protocol. Capability inventory: Agents can execute the cmux-team CLI and write to project files. Sanitization: No explicit sanitization of peer-generated content is defined, which is consistent with the local, coordinated team environment described.
Audit Metadata