new-modular-project

Warn

Audited by Gen Agent Trust Hub on Sep 14, 2026

Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PRIVILEGE_ESCALATION]: The instructions include commands that use sudo to install the gcc compiler (e.g., sudo apt install gcc) or initialize WSL2 on Windows, which involves high-privilege operations required for standard system setup.
  • [COMMAND_EXECUTION]: The skill builds and executes shell commands by interpolating user-provided project names and project types into commands like pixi init [PROJECT] and uv init [PROJECT].
  • [EXTERNAL_DOWNLOADS]: The skill fetches environment configuration and software packages from official modular.com subdomains (conda.modular.com, whl.modular.com) and uses the prefix.dev service for Pixi installation.
  • [INDIRECT_PROMPT_INJECTION]: Ingestion points: User-specified parameters for project names and types are gathered from natural language requests. Boundary markers: No explicit delimiters or instructions are provided to the agent to treat user-provided strings as non-executable data when used in shell commands. Capability inventory: The skill uses environment managers (pixi, uv, pip, conda) and system package managers that possess broad file system and network capabilities. Sanitization: There are no explicit instructions for validating or sanitizing the project names before they are interpolated into shell command templates.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Sep 14, 2026, 03:30 PM
Security Audit — agent-trust-hub — new-modular-project