openai-docs

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

• Third-party content exposure detected (high risk: 0.85). High: the skill uses the OpenAI Developer Docs MCP tools (mcp__openaiDeveloperDocs__search_openai_docs / mcp__openaiDeveloperDocs__fetch_openai_doc) to fetch and ingest markdown text from the public developers.openai.com docs at runtime into the agent’s LLM context.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly fetches OpenAI developer docs at runtime (e.g., via the MCP server URL https://developers.openai.com/mcp and direct fetches like https://developers.openai.com/codex/codex-manual.md and https://developers.openai.com/api/docs/guides/latest-model.md) and uses that retrieved content to drive prompt guidance and model-selection decisions, so external content directly controls agent instructions.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.80). The skill explicitly tells the agent to run install commands and, if they fail due to permissions/sandboxing, to retry with "escalated permissions" (i.e., request sudo/elevation) and to run helper scripts that write to system temp dirs, which requires and encourages privilege escalation and making persistent changes to the host.