skill-installer

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

• Potentially malicious external URL detected (high risk: 0.90). The installer explicitly fetches skill code at runtime from GitHub (e.g. https://github.com/openai/skills/tree/main/skills/.curated and URLs of the form https://github.com///tree//), and those fetched files are installed as skills which can contain prompts or executable code that will control agent behavior, so this is a runtime dependency that can directly control prompts/execute code.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The prompt explicitly instructs the agent to "request escalation" when running the networked installer scripts in the sandbox and allows overwriting preinstalled system skills, which encourages obtaining elevated privileges and modifying the host state.