openai-docs
Use whenever the user asks anything about OpenAI products, APIs, models, or SDKs — including factual lookups ("what is the newest model", "does X support Y"), build/coding requests ("write a script using the Responses API"), latest/current/default-model prompting guidance, model migrations, and citation requests. Invoke before calling any openaiDeveloperDocs MCP tool.
- License 许可
- Apache-2.0
- 最近更新
- 2026-11-03
- Stars
- ★ 13
- Forks
- 7
01 前言 Front Matter
- name
- openai-docs
- description
- Use whenever the user asks anything about OpenAI products, APIs, models, or SDKs — including factual lookups ("what is the newest model", "does X support Y"), build/coding requests ("write a script using the Responses API"), latest/current/default-model prompting guidance, model migrations, and citation requests. Invoke before calling any openaiDeveloperDocs MCP tool.
03 SKILL.md 正文
OpenAI Docs
Use the Cursor-configured OpenAI Docs MCP server at https://developers.openai.com/mcp?source=cursor first for current OpenAI developer guidance.
API key setup
For model and prompt migrations, missing credentials block only the live call. They do not block requested migration source or configuration edits, documentation retrieval, or offline, mocked, static, fixture-based, or syntax validation. For other requests to build, run, configure, debug, or implement API-backed functionality, use openai-platform-api-key first when available.
For model migration or prompting-guidance requests, retrieve the relevant current guidance before any live-call credential check. Continue with requested source edits and offline validation when no API key is available; skip only validation that actually sends a live API request, and say that it was skipped.
Model guidance routing
Classify the model request before inspecting the project or checking credentials:
- Dynamic target: For latest, current, newest, recommended, default, flagship, or otherwise unspecified model selection, prompting, or migration, fetch
https://developers.openai.com/api/docs/guides/latest-model.md. Read itslatestModelInfometadata. For migrations or prompting guidance, resolve the exactmigrationGuideandpromptingGuidepaths againsthttps://developers.openai.comand fetch those exact URLs. Treat the metadata values as opaque; do not derive or substitute another route. - Explicit target: Preserve the requested model and fetch
https://developers.openai.com/api/docs/guides/latest-model?model=<requested-model>. For example, usehttps://developers.openai.com/api/docs/guides/latest-model?model=gpt-5.6for GPT-5.6 andhttps://developers.openai.com/api/docs/guides/latest-model?model=gpt-5.3-codexforgpt-5.3-codex. Do not replace the explicit target with the current model.
If an exact URL is unavailable or lacks substantive content, retry it once through Docs MCP, then run a compact Docs MCP search for that exact model or page. If necessary, fall back only to official OpenAI domains such as developers.openai.com and platform.openai.com. Do not widen to Slack, internal docs, private repositories, or historical launch artifacts.
If current-model metadata or an explicit model guide still cannot be retrieved, say what could not be verified and return bounded uncertainty. Do not infer current model facts, migration steps, limits, pricing, availability, parameters, or capabilities from static knowledge.
General workflow
- Use the OpenAI Docs MCP tools exposed by the configured
openaiDeveloperDocsserver. - Start general lookup with a compact, title-like search query of 2-6 essential terms. If search is noisy, narrow it.
- When a plausible official docs URL is known, fetch it through Docs MCP before web fallback.
- For API schemas, parameters, or required fields, use
get_openapi_specwhen available alongside the guide. - Use
list_openai_docsonly when no clear query or URL exists. - Fetch the exact page or section needed before answering.
- Cite retrieved official sources for links, quotes, and precise claims.
Migration rules
- Keep upgrades behavior-preserving. Update active OpenAI model defaults, directly related prompts, and only the registries, routing, pricing, capability, or picker surfaces the user placed in scope.
- Preserve multi-model cost, latency, and quality roles. Do not collapse a router or picker into one flagship model.
- Leave historical docs, examples, eval baselines, fixtures, provider comparisons, intentionally pinned fallbacks, and ambiguous older usage unchanged unless explicitly requested.
- Keep SDK, tooling, IDE, plugin, shell, auth, and provider-environment migrations out of a model-and-prompt upgrade unless explicitly requested.
- If a safe change requires endpoint or schema rewiring, tool-handler changes, state changes, or another compatibility migration outside scope, report the exact blocker or ask for confirmation.
Validation
After changes, run targeted tests for changed behavior and the most relevant type, lint, build, or smoke checks available. For model migrations, verify every active default surface in scope and exercise representative requests or evals. If validation cannot run, state why and name the next best check.
Rules
- Treat live OpenAI documentation as the source of truth; never use bundled model guidance as a fallback.
- Preserve explicit model targets.
- Prefer concise summaries over long excerpts.
- If sources conflict, state the difference. If evidence is missing, narrow the answer instead of guessing.
04 同仓库更多 Skills
openai-platform-api-key
Use when Cursor is asked to build, run, test, debug, or configure an OpenAI-backed or provider-unspecified AI app, UI, script, CLI, generator, or tool, especially requests phrased only as "using AI" or generators driven by forms/user input; also use for OPENAI_API_KEY or sk-proj setup. Do not gate model or prompt migrations that can proceed with source edits and offline validation. Otherwise treat this as the credential gate: inspect safely, ask reuse-vs-new before API work, guide manual local setup when needed, verify presence, and never expose plaintext.
openai-api-troubleshooting
Use when an OpenAI API request fails or the user asks what an error means, including blocked access to api.openai.com, 401 invalid_api_key, missing OPENAI_API_KEY, 429 insufficient_quota, 429 rate_limit_exceeded, or 403 model_not_found. Classify the failure, name the next step, route credential setup to openai-platform-api-key, and current guidance to openai-docs.
chatgpt-app-submission
Use when a developer needs a ChatGPT Apps submission import file from an MCP server. Inspect the implementation, generate chatgpt-app-submission.json with app info, tool-hint justifications, positive and negative test cases, then report review findings and outputSchema warnings.
build-chatgpt-app
Use when the user explicitly wants a ChatGPT App, Apps SDK project, MCP server plus widget UI, ChatGPT-compatible tool UI, or submission/deployment work for that surface. Build, scaffold, refactor, or troubleshoot those projects from Cursor, consult openai-docs before generating code, and use openai-platform-api-key before any app path that requires live OpenAI API credentials.
agents-sdk
Use when the user wants to build, adapt, run, or evaluate an OpenAI Agents SDK app from Cursor. Prefer a docs-first workflow and route API credential decisions through openai-platform-api-key before API-backed work.