ux-foundation
Warn
Audited by Gen Agent Trust Hub on Nov 12, 2026
Risk Level: MEDIUMPRIVILEGE_ESCALATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PRIVILEGE_ESCALATION]: The file
references/browser-engine.mdcontains instructions to executesudo safaridriver --enable. This command requires administrative privileges to modify system-level settings to allow remote automation on the host machine. - [COMMAND_EXECUTION]: Instructions in
SKILL.md,references/scenario-format.md, andreferences/system-map.mddirect the agent to execute local Python and JavaScript scripts, such aspython3 docs/ux/lint.py,python3 docs/ux/doctor.py, andbin/super-ux.js. These files are not provided in the skill package, resulting in the execution of unverified local code. - [COMMAND_EXECUTION]: The skill recommends installing and executing external tools using
npx(e.g.,npx sheleg-design-skill,npx task-pipeline). While these are identified as vendor-owned resources, the runtime execution of packages from public registries is a notable risk. - [INDIRECT_PROMPT_INJECTION]: The skill defines a clear attack surface for indirect prompt injection. The
templates/agent-panel.mdfile includes a template for an 'Adversarial agent' (AG-05) designed to handle 'injected content in what it reads.' The core workflow of the skill involves processing untrusted external data, such as competitor funnel URLs and store reviews (references/funnel-research.md), which may contain malicious instructions. - [EXTERNAL_DOWNLOADS]: The skill references several external vendor URLs on GitHub (e.g.,
github.com/ssheleg/*) for synchronization and documentation purposes.
Audit Metadata