openai-docs

# OpenAI documentation Use mounted search/fetch tools or an authorized documentation connector to find the user's specific topic on official OpenAI sources. Fetch the supporting page, not only its search snippet. Prefer developers.openai.com and platform.openai.com for APIs; use the product's official documentation for ChatGPT and Codex. If no fetch tool is mounted but an authorized process tool has network access, retrieve a known official HTTPS page with `curl` or Python's `urllib.request`. Check status and final URL, retain the supporting response for inspection, and verify redirects remain on an official documentation domain. This provides direct retrieval, not a search engine or an authenticated product connector. Preserve explicitly requested models, endpoint versions, and product context. Check changing details such as models, prices, limits, and supported parameters against current pages. Keep citations adjacent to the claims they support. Distinguish documented behavior from inference and runtime observations. If the source cannot be fetched, state the limit rather than supplying invented links or calling stale knowledge current. For implementation, inspect the actual SDK and host integration. Keep API keys in the host's authorized secret configuration. Do not treat Codex-only tools, ChatGPT product features, or subscription benefits as public API capabilities. When the user says "this app" inside Amplifier, inspect Amplifier's own contracts; this skill does not redefine the current host as Codex.