cmux-backend
Warn
Audited by Gen Agent Trust Hub on Sep 2, 2026
Risk Level: MEDIUMDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [DATA_EXFILTRATION]: The skill instructions direct the agent to access provider secrets from sensitive local file paths including
~/.secrets/cmux.env,~/.secrets/cmuxterm-dev.env, and~/.secret/cmuxterm.envto facilitate cloud VM builds and local development. - [INDIRECT_PROMPT_INJECTION]: The skill outlines the creation of Vercel route handlers that ingest and parse data from external request inputs, which constitutes a potential attack surface.
- Ingestion points: Request inputs are parsed at the boundary of route handlers within
web/app/api/**andweb/services/**. - Boundary markers: There are no instructions for using delimiters or explicit warnings to the agent to ignore instructions embedded in the external data.
- Capability inventory: The agent is empowered to perform database writes (Postgres), manage Cloud VM lifecycles through provider APIs, and interact with authentication and pricing services.
- Sanitization: The skill does not provide requirements for sanitizing, validating, or escaping the external request input before it is passed to Effect services.
Audit Metadata