skills/muxy-app/muxy/e2e-test/Gen Agent Trust Hub

e2e-test

Pass

Audited by Gen Agent Trust Hub on Nov 19, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPRIVILEGE_ESCALATIONDATA_EXFILTRATIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill fetches development tools and installers from reputable sources, including the Rust toolchain (sh.rustup.rs), Homebrew (github.com/Homebrew), Zig (ziglang.org), and official Microsoft, Python, and Git repositories. These downloads are integral to the automated setup of testing templates.- [REMOTE_CODE_EXECUTION]: Setup scripts execute downloaded installers to configure the guest VM environments. The skill utilizes checksum verification (SHA-256) for many of these downloads to ensure the integrity of the executed code.- [DATA_EXFILTRATION]: A dedicated source export script (scripts/export-source.py) actively protects against the exposure of sensitive information. It identifies and excludes directories containing credentials, such as .ssh, .aws, and .env files, before archiving code for use in the guest VMs.- [PRIVILEGE_ESCALATION]: The harness performs necessary administrative actions, such as using sudo on the host to manage DHCP leases for the VM network and adjusting execution policies within Windows guests. These actions are limited in scope to the operational requirements of the testing harness.- [COMMAND_EXECUTION]: The skill uses Python and shell scripts to automate VM management and UI interaction. This involves running local commands like git and ifconfig to gather environment data and drive the swiftvm control utility.- [INDIRECT_PROMPT_INJECTION]: While the agent reviews logs and screenshots from the VM—creating a surface for indirect prompt injection—the skill mitigates this risk by providing subagents with a strict, mechanical protocol that prohibits autonomous debugging or reasoning based on the observed data.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 19, 2026, 08:13 AM