技能 openai-docs
📦
openai-docs
内容修订 r1 高风险 🌐 网络访问⚙️ 外部命令📁 文件系统访问⚡ 包含脚本🔑 环境变量
Find Current OpenAI and Codex Documentation
OpenAI products, models, and APIs change quickly, making remembered guidance unreliable. This skill retrieves current official documentation and provides concise, cited answers.
支持: Claude Codex Code(CC)
安装前请停下并请求确认。
请审查安装计划，并在更改文件前获得用户明确同意。
使用我的智能体安装
将此请求复制给你的智能体，其中包含权威 Skill 页面和清单。
智能体请求
Review the Skillstore skill "openai-docs" from https://skillstore.io/skills/openai-docs.md and its manifest at https://skillstore.io/api/skills/openai-docs/manifest. Verify the artifact. Stop and obtain explicit user consent before installing or changing files.你的智能体仍应展示其计划，并按安全策略请求必要确认。
Agent 可读资源
当 AI Agent、爬虫或脚本需要干净上下文，而不是读取完整页面时，请使用这些链接。
测试它
正在使用“openai-docs”。 Which OpenAI API should handle a stateful tool-using workflow?
预期结果:
Recommendation: use the documented Responses API workflow. The answer explains state handling, tool calls, limitations, and cites the relevant official guide.
正在使用“openai-docs”。 Plan our migration from a named model to a newer target.
预期结果:
- Target decision based on current official model guidance.
- Compatibility changes for requests, tools, reasoning, and state.
- Prompt updates only where documented behavior requires them.
- Validation checks and citations for each consequential change.
正在使用“openai-docs”。 Why is the OpenAI documentation MCP server unavailable in my local Codex client?
预期结果:
A focused diagnostic covering supported setup, enabled state, policy, authorization, refresh requirements, and evidence needed before claiming success.
安全审计
高风险v1 • 2026/8/17 打开版本化报告
Most static matches are false positives caused by SVG coordinates, Markdown text, regular-expression calls, or JavaScript template literals. Confirmed behavior includes official network requests, cache writes, environment reads, shell execution, and a curl fallback. Unrestricted sources and cache destinations create SSRF, argument-injection, local-read, and overwrite risks.
16
已扫描文件
2,028
分析行数
37
复核项
0
已忽略误报
已确认安全问题 (3)
高
Unvalidated Manual URL Enables SSRF and Curl Option Injection
The manual URL accepts arbitrary targets and reaches curl without an option terminator, enabling internal requests or curl option injection.
The CLI value is appended directly to curl arguments and no scheme, host, or option-prefix validation occurs.
中
Unrestricted Model Source Reads Local Files or Arbitrary URLs
Environment variables and command arguments can select any local path or HTTP URL, bypassing the skill's official-domain boundary.
The resolver explicitly accepts environment and CLI sources, then reads local files or fetches any HTTP endpoint.
中
scripts/fetch-codex-manual.mjs:286-296scripts/fetch-codex-manual.mjs:394-405scripts/fetch-codex-manual.mjs:511-512
Unrestricted Cache Directory Can Replace Files
A user-selected cache directory receives fixed filenames through rename operations, which can replace existing files in any writable directory.
The override accepts any writable directory and atomically renames fixed cache filenames over existing entries.
能力复核项 (37)
这些是真实的本地能力，对此技能可能是预期行为，因此需要复核，但不会按已确认恶意行为计分。
高
Node.js child_process module
import { execFile } from "node:child_process";
The helper imports child_process and invokes curl through execFile as a network fallback.
中
Proxy environment variable
process.env.HTTP_PROXY ||
The helper reads standard proxy variables and changes its network transport when they are present.
中
Proxy environment variable
process.env.HTTPS_PROXY ||
The helper reads standard proxy variables and changes its network transport when they are present.
中
Proxy environment variable
process.env.http_proxy ||
The helper reads standard proxy variables and changes its network transport when they are present.
中
Proxy environment variable
process.env.https_proxy;
The helper reads standard proxy variables and changes its network transport when they are present.
中
Temp directory access
pushCandidate(`/private/tmp/${DEFAULT_CACHE_DIR_NAME}`);
The helper selects and writes to an operating-system temporary cache directory.
中
Temp directory access
pushCandidate(`/tmp/${DEFAULT_CACHE_DIR_NAME}`);
The helper selects and writes to an operating-system temporary cache directory.
中
Shell command substitution
SCRIPT_DIR=$(CDPATH= cd -- "$(dirname -- "$0")" && pwd)
The POSIX wrapper uses command substitution to locate its script directory or Node runtime.
中
Shell command substitution
PATH_NODE=$(command -v node 2>/dev/null || true)
The POSIX wrapper uses command substitution to locate its script directory or Node runtime.
中
Unix shell invocation
#!/bin/sh
The executable wrapper invokes a POSIX shell and then runs a compatible Node runtime.
中
Hidden file access
"$HOME/.cache/codex-runtimes/codex-primary-runtime/dependencies/node/bin/node" \
The wrapper checks executable Node binaries under the user's hidden Codex runtime cache.
中
Hidden file access
"$HOME/.cache/codex-runtimes/codex-primary-runtime/dependencies/bin/node" \
The wrapper checks executable Node binaries under the user's hidden Codex runtime cache.
中
Node.js fs operations
return fs.readFile(new URL(source), "utf8");
The resolver reads a local file when its configurable source is not an HTTP URL.
中
Node.js fs operations
return fs.readFile(path.resolve(source), "utf8");
The resolver reads a local file when its configurable source is not an HTTP URL.
低
Hardcoded URL
codex mcp add openaiDeveloperDocs --url https://developers.openai.com/mcp
The source contains an explicit official OpenAI network endpoint.
低
Hardcoded URL
url = "https://developers.openai.com/mcp"
The source contains an explicit official OpenAI network endpoint.
低
Fetch API call
fetch(url, {
The helper performs an HTTP request with the Fetch API.
低
Hardcoded URL
const DEFAULT_MANUAL_URL = "https://developers.openai.com/codex/codex-manual.md";
The source contains an explicit official OpenAI network endpoint.
低
Environment variable access (dot notation)
process.env.HTTP_PROXY ||
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable access (dot notation)
process.env.HTTPS_PROXY ||
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable access (dot notation)
process.env.http_proxy ||
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable access (dot notation)
process.env.https_proxy;
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable access (dot notation)
[process.env.TMPDIR, process.env.TEMP, process.env.TMP].forEach((baseDir) => {
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
process.env.HTTP_PROXY ||
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
process.env.HTTPS_PROXY ||
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
process.env.http_proxy ||
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
process.env.https_proxy;
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
[process.env.TMPDIR, process.env.TEMP, process.env.TMP].forEach((baseDir) => {
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Standard device file access
"$1" -e 'const major = Number(process.versions.node.split(".")[0]); process.exit(major >= 18 ? 0 : 1
The wrapper redirects runtime-probe output to the standard null device.
低
Standard device file access
PATH_NODE=$(command -v node 2>/dev/null || true)
The wrapper redirects runtime-probe output to the standard null device.
低
Fetch API call
const response = await fetch(source, {
The helper performs an HTTP request with the Fetch API.
低
Hardcoded URL
"https://developers.openai.com/api/docs/guides/latest-model.md";
The source contains an explicit official OpenAI network endpoint.
低
Hardcoded URL
const DEFAULT_BASE_URL = "https://developers.openai.com";
The source contains an explicit official OpenAI network endpoint.
低
Environment variable access (dot notation)
source: process.env.LATEST_MODEL_URL || DEFAULT_URL,
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable access (dot notation)
baseUrl: process.env.LATEST_MODEL_BASE_URL || DEFAULT_BASE_URL,
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
source: process.env.LATEST_MODEL_URL || DEFAULT_URL,
The helper reads the named environment variable to choose a proxy, cache location, or model source.
低
Environment variable object
baseUrl: process.env.LATEST_MODEL_BASE_URL || DEFAULT_BASE_URL,
The helper reads the named environment variable to choose a proxy, cache location, or model source.
风险因素
🌐 网络访问 (14)
assets/openai-small.svg:2 references/mcp-diagnostics.md:12 references/mcp-diagnostics.md:19 references/upgrading-to-gpt-5p6-sol.md:262 scripts/fetch-codex-manual.mjs:175 scripts/fetch-codex-manual.mjs:19 scripts/fetch-codex-manual.mjs:47 scripts/fetch-codex-manual.mjs:48 scripts/fetch-codex-manual.mjs:49 scripts/fetch-codex-manual.mjs:50 scripts/resolve-latest-model-info.cjs:44 scripts/resolve-latest-model-info.cjs:9 scripts/resolve-latest-model-info.cjs:10 SKILL.md:14
⚙️ 外部命令 (35)
references/codex-self-knowledge.md:33 scripts/fetch-codex-manual.mjs:12 scripts/fetch-codex-manual.mjs:79 scripts/fetch-codex-manual.mjs:324 scripts/fetch-codex-manual.mjs:82 scripts/fetch-codex-manual.mjs:104-106 scripts/fetch-codex-manual.mjs:161 scripts/fetch-codex-manual.mjs:202 scripts/fetch-codex-manual.mjs:211 scripts/fetch-codex-manual.mjs:219 scripts/fetch-codex-manual.mjs:279 scripts/fetch-codex-manual.mjs:280 scripts/fetch-codex-manual.mjs:318-363 scripts/fetch-codex-manual.mjs:363-369 scripts/fetch-codex-manual.mjs:369-378 scripts/fetch-codex-manual.mjs:378-379 scripts/fetch-codex-manual.mjs:379-396 scripts/fetch-codex-manual.mjs:396-403 scripts/fetch-codex-manual.mjs:403-463 scripts/fetch-codex-manual.mjs:463-472 scripts/fetch-codex-manual.mjs:472-518 scripts/fetch-codex-manual.mjs:518-545 scripts/fetch-codex-manual.mjs:545 scripts/fetch-codex-manual.mjs:548-551 scripts/fetch-codex-manual.mjs:551 scripts/fetch-codex-manual.mjs:572-586 scripts/resolve-latest-model-info:4 scripts/resolve-latest-model-info:23 scripts/resolve-latest-model-info:1 scripts/resolve-latest-model-info.cjs:154 scripts/resolve-latest-model-info.cjs:158 SKILL.md:18 SKILL.md:34 SKILL.md:36 SKILL.md:37
📁 文件系统访问 (18)
references/codex-self-knowledge.md:22 references/codex-self-knowledge.md:24 references/codex-self-knowledge.md:25 references/codex-self-knowledge.md:33 scripts/fetch-codex-manual.mjs:79 scripts/fetch-codex-manual.mjs:318 scripts/fetch-codex-manual.mjs:324 scripts/fetch-codex-manual.mjs:279 scripts/fetch-codex-manual.mjs:280 scripts/fetch-codex-manual.mjs:551 scripts/resolve-latest-model-info:29 scripts/resolve-latest-model-info:30 scripts/resolve-latest-model-info:8 scripts/resolve-latest-model-info:23 scripts/resolve-latest-model-info.cjs:34 scripts/resolve-latest-model-info.cjs:38 scripts/resolve-latest-model-info.cjs:37 scripts/resolve-latest-model-info.cjs:104
🔑 环境变量 (14)
scripts/fetch-codex-manual.mjs:47 scripts/fetch-codex-manual.mjs:48 scripts/fetch-codex-manual.mjs:49 scripts/fetch-codex-manual.mjs:50 scripts/fetch-codex-manual.mjs:272 scripts/fetch-codex-manual.mjs:47 scripts/fetch-codex-manual.mjs:48 scripts/fetch-codex-manual.mjs:49 scripts/fetch-codex-manual.mjs:50 scripts/fetch-codex-manual.mjs:272 scripts/resolve-latest-model-info.cjs:14 scripts/resolve-latest-model-info.cjs:15 scripts/resolve-latest-model-info.cjs:14 scripts/resolve-latest-model-info.cjs:15
审计者: codex
分享与引用此报告
分享版本化评估报告、中性徽章、嵌入卡片和引用。Skillstore 只报告证据，不判断该 Skill 是否安全。
复制报告链接
https://skillstore.io/skills/openai-docs/audits/1?utm_source=security_passport&utm_medium=share&utm_campaign=versioned_reportMarkdown 徽章
[](https://skillstore.io/skills/openai-docs?utm_source=security_passport_badge)HTML 徽章
<a href="https://skillstore.io/skills/openai-docs?utm_source=security_passport_badge"><img src="https://skillstore.io/badges/skills/openai-docs/security.svg" alt="Skillstore security assessment" loading="lazy"></a>嵌入卡片
<iframe src="https://skillstore.io/embed/skills/openai-docs.html" title="Skillstore Security Assessment" sandbox="allow-popups allow-popups-to-escape-sandbox" loading="lazy" referrerpolicy="no-referrer" width="420" height="180"></iframe>学术引用 (APA · BibTeX · CFF)
APA 引用
openai. (2026). openai-docs security audit report (audit version 1) [Author version unspecified]. Skillstore. https://skillstore.io/skills/openai-docs/audits/1BibTeX 引用
@techreport{openai-openai-docs-2026,
author = {openai},
title = {openai-docs security audit report (audit version 1)},
institution = {Skillstore},
year = {2026},
number = {1},
url = {https://skillstore.io/skills/openai-docs/audits/1},
note = {Author version unspecified}
}CITATION.cff
cff-version: 1.2.0
message: "If you use this Skill, cite its author and this versioned security audit report."
title: "openai-docs security audit report (audit version 1)"
version: "unspecified"
type: report
authors:
- name: "openai"
date-released: "2026-08-17"
url: "https://skillstore.io/skills/openai-docs/audits/1"
identifiers:
- type: other
value: "skillstore:openai-docs:audit:1"
description: "Skillstore immutable audit report identifier"
Skillstore 评分
评分依据 证据置信度: 中 82
架构
85
可维护性
87
内容
65
社区
83
规范符合性
你能构建什么
Verify API Behavior
Confirm endpoints, request fields, SDK behavior, and error handling against current official documentation.
Plan a Model Migration
Compare targets and identify model, prompt, tool, state, compatibility, and testing changes.
Configure Codex
Find current guidance for Codex settings, skills, plugins, MCP, automations, setup, and troubleshooting.
试试这些提示
Find an Official Answer
Find current official OpenAI documentation for [topic]. Answer [question] concisely and cite the supporting page.
Compare Two Models
Compare [model A] and [model B] for [workload] using current official documentation. Evaluate quality, latency, cost, modalities, and availability.
Plan a Verified Migration
Plan a migration from [current model] to [target model] for [application]. Identify API, prompting, tool, state, compatibility, and testing changes with citations.
Audit an OpenAI Integration
Audit this OpenAI integration against current official documentation: [system description]. Identify unsupported assumptions, schema mismatches, migration risks, and the smallest verified fixes.
最佳实践
- Name the exact product, model, API, or Codex feature in the request.
- Preserve explicitly requested models and state workload constraints before asking for recommendations.
- Use cited official pages for current facts and disclose any bundled fallback.
避免
- Do not ask for current pricing, availability, or limits without requiring official citations.
- Do not treat a search snippet, memory, or bundled snapshot as current documentation.
- Do not request configuration changes when only a documentation answer is needed.
常见问题
Which sources does this skill use?
It uses official OpenAI developer documentation, platform documentation, ChatGPT Work learning pages, and narrowly selected bundled references.
Can it choose an OpenAI model?
Yes. It compares documented workload fit, quality, latency, cost, modality, and availability while preserving any explicitly requested model.
Can it update my application?
It can guide or implement requested changes after verifying current API and model documentation. API-backed execution may require a separate credential workflow.
Does it know my account limits or access?
No. It states uncertainty when official documentation does not establish account-specific access, quotas, pricing, billing, or rollout status.
What happens when live documentation is unavailable?
It may use a relevant bundled fallback, clearly disclose that choice, and warn that the information may be outdated.
Can it configure the OpenAI documentation MCP server?
Yes, for a supported local Codex client after explicit authorization. It verifies current setup guidance and does not claim success without evidence.