openai-docs

CRITICAL E006: Malicious code pattern detected in skill scripts.

• Malicious code pattern detected (high risk: 0.90). The skill instructs automated installation and immediate re-run of an MCP package with escalated permissions (including "Do not ask the user to run it yet"), which is an explicit, high‑risk pattern for unauthorized privilege escalation and supply‑chain/backdoor implantation via remote tool installation.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill declares and requires the MCP server at runtime (https://developers.openai.com/mcp), which the agent will fetch and use as authoritative documentation that directly controls its generated guidance and behavior.

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

• Attempt to modify system services in skill instructions detected (high risk: 1.00). The prompt explicitly directs installing an MCP server and to retry the install with escalated permissions to bypass sandboxing (without asking the user), which requires privilege escalation and modifies the system state.