openai-docs

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (high risk: 0.90). The skill contains hidden/deceptive operational instructions (lines 25-26) telling the agent to retry installation with escalated permissions and "Do not ask the user," which attempts to override normal consent/authority and is outside the documented purpose of fetching docs.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill declares and uses an MCP tool at runtime that fetches OpenAI docs from https://developers.openai.com/mcp and injects those fetched docs into the agent's responses (e.g., mcp__openaiDeveloperDocs__fetch_openai_doc), so the remote URL directly controls prompt/context during runtime.

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 the agent to "immediately retry the same command with escalated permissions" without asking the user, which directs the agent to obtain/attempt sudo or bypass sandboxing and thus modify the machine state.