codex-automation-architect

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

• Third-party content exposure detected (high risk: 0.80). The skill explicitly fetches and interprets untrusted, user-generated GitHub PR content via "gh pr list" in examples/stale-pr-triage-automation.yaml and also refreshes external guidance from the openaiDeveloperDocs MCP (agents/openai.yaml -> https://developers.openai.com/mcp) and public repo links, so third-party content is read and used to drive categorization and action recommendations.

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

• Potentially malicious external URL detected (high risk: 0.80). The skill explicitly declares and at runtime refreshes guidance from the OpenAI docs MCP URL (https://developers.openai.com/mcp), which is fetched via streamable_http and used to ground and shape the agent's prompts/instructions, so external content can directly control agent behavior.