run-upl-prompts

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (medium risk: 0.30). The workflow performs recursive discovery of prompt files under ~/.upl/prompts or a user/agent-supplied directory and then renders the selected .txt/.upl prompt via upl build-from-json, meaning outsider-authored text can be ingested when an attacker can place or point to prompt files.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill instructs the runtime download and installation of the upl executable from explicit GitHub release URLs (for example: https://github.com/DavidValin/universal-prompt-language/releases/download/0.1.1-rc.2/upl-x86_64_linux.tar.gz), which fetches remote executable code that the skill will run to render prompts.

MEDIUM W013: Attempt to modify system services in skill instructions.

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The skill explicitly instructs using sudo to install the upl executable into /usr/local/bin and to run sudo make install, which modify system files and require elevated privileges.