openai-docs

CRITICAL E004: Prompt injection detected in skill instructions.

• Potential prompt injection detected (medium risk: 0.30). The skill includes explicit instructions to retry an install with "escalated permissions" and to provide a justification, which steers users toward privilege escalation and weakening a security boundary.

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

• Third-party content exposure detected (low risk: 0.10). In SKILL.md, at runtime the workflow ingests outsider-provided free text only indirectly via the user’s query, then fetches OpenAI-authored documentation pages using the developers.openai.com MCP tools (search_openai_docs / fetch_openai_doc) and, if needed, bounded official-domain web fallback, without reading attacker-submitted items from a queue/feed.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill calls the Docs MCP server (https://developers.openai.com/mcp) and fetches live OpenAI markdown guides at runtime (for example https://developers.openai.com/api/docs/guides/latest-model.md, https://developers.openai.com/api/docs/guides/model-guidance?model=, and https://developers.openai.com/codex/codex-manual.md) and uses their content to derive prompting and migration instructions that directly control agent outputs.

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

• Attempt to modify system services in skill instructions detected (high risk: 0.90). The skill explicitly tells the agent to retry an install command "with escalated permissions" (i.e., request/obtain elevated/sudo privileges) and to run local install/resolve scripts, which directs the agent to perform privilege escalation and system-level changes.