AI 整合指南 | API 文件

AI 串接提示詞

You are integrating with API Hop's public APIs.

Base URL:
https://api.apihop.ai

Canonical Base URL for this session:
https://api.apihop.ai

When creating the skill, store this exact Base URL string. Do not replace it with another URL.

API Key:
<API_KEY>

Authentication:
Use the same Bearer API Key for every request:
Authorization: Bearer <API_KEY>

AI installation guide and public API docs:
https://apihop.ai/tw/docs/api-reference/ai-installation-guide

Machine-readable public API catalog:
https://apihop.ai/docs/public-api-catalog.json

Documentation fallback rule:
First open the AI installation guide URL above. If the hash-routed docs page cannot be read directly by fetch, you MUST fetch this exact machine-readable catalog URL:
https://apihop.ai/docs/public-api-catalog.json

Do not guess endpoints, parameters, response shapes, streaming events, pagination, or errors from memory, old docs, training data, or unrelated local files.

Credential precedence:
The value <API_KEY> is a placeholder, not a credential. Ask the user to configure APIHOP_API_KEY locally through an environment variable or secret store. Never ask the user to paste a key into an AI conversation, or retrieve a full key from the console.

Your task:
Create a reusable agent skill for API Hop's public APIs. Paid or live API verification requires explicit user approval.

Skill name:
api-hop-public-api

Skill creation requirements:
1. If your environment supports persistent skills, plugins, tools, agent memory, custom instructions, or project rules, install this as a reusable capability named api-hop-public-api. Follow the current agent platform's official installation convention. If you cannot find one, create a SKILL.md or equivalent platform-specific skill spec in the current workspace or conversation.
2. Keep the skill lightweight. Do not copy the full installation guide or endpoint documentation into the skill. Store only the project name, the exact Base URL from this prompt, Bearer authentication rule, this installation guide URL, this exact catalog URL, and the workflow for reading docs on demand.
3. The skill must instruct future calls to open the installation guide URL first, read its public API documentation link list, then open only the endpoint docs needed for the current task. Tool integration docs such as Codex setup are helpful for configuring that tool, but they are not public API endpoint docs. If the hash docs page is not readable without a browser, you MUST fetch https://apihop.ai/docs/public-api-catalog.json. Treat the Public API Docs section and catalog as the source of truth for every public endpoint, parameter, response, streaming event, pagination rule, and error shape.
4. "Every public API" means every endpoint link under the Public API Docs section of the installation guide, including future links added there, and no endpoints outside those docs. Do not treat links under Tool Integration Docs as endpoints unless they also appear in Public API Docs or the machine-readable catalog.
5. The skill must preserve the API Hop project name from this prompt.
6. Do not put the API Key into the skill, generated source files, tests, logs, screenshots, or commits. Keep <API_KEY> as a placeholder in examples and instructions. Do not make a live request until a valid API Key is configured locally and the user explicitly approves the call. Recommend a local environment variable or secret store for credentials.

The skill content should be concise and roughly follow this structure:
- name: api-hop-public-api
- description: Use API Hop's public APIs by reading the official public API docs on demand.
- connection: Base URL https://api.apihop.ai. Store this exact Base URL string. Send Authorization: Bearer <API key from the current session or configured secret>.
- docs: Always open https://apihop.ai/tw/docs/api-reference/ai-installation-guide, read the Public API Docs section for endpoint docs, and optionally read Tool Integration Docs only when configuring a named tool such as Codex. If a browser-only hash URL cannot be fetched directly, fetch https://apihop.ai/docs/public-api-catalog.json. Do not guess.
- workflow: read docs -> build request -> call API -> parse response, errors, SSE events, pagination, and generated image URLs according to the endpoint docs.
- security: Never store or reveal the full API Key.

Runtime behavior requirements:
1. Do not make an API request while the Authorization value is <API_KEY>. Use the locally configured APIHOP_API_KEY without printing or exposing it.
2. Before calling an endpoint, read the installation guide and the relevant endpoint doc instead of relying on memory or copied documentation. If the installation guide cannot be fetched because it is a hash-routed page, fetch https://apihop.ai/docs/public-api-catalog.json before making any API call.
3. Implement or use SSE parsing for streaming endpoints and stop on completion signals described in the endpoint docs.
4. For paginated Twitter timeline APIs, read meta.next_token and send it back as pagination_token when fetching the next page.
5. For image generation, handle non-streaming responses with data[].url and streaming image_generation.completed events with final image URLs.
6. Implement unified error handling for flat OpenAI-compatible errors and Twitter errors shaped as { success: false, error, requestId }.
7. Only call endpoints listed in the installation guide's Public API Docs section or the machine-readable catalog.
8. Before any explicitly approved verification call, confirm the request URL starts with https://api.apihop.ai and the Authorization header uses the locally configured credential without displaying it.
9. After explicit user approval and local credential setup, make one minimal verification call through the skill using a model confirmed available to the current account. Read the endpoint documentation and agree on the model and request first; do not copy endpoint docs into the skill. If the call returns 401 or an invalid/expired token error, ask the user to update their local credential configuration. Do not retry with another stored key. For any failure, report the HTTP status, redacted response body, and likely next step without revealing the full API Key.

複製「AI 適用範本」並交給您的 AI 工具。所有 AI 整合範本都使用 <API_KEY>，絕不讀取或包含帳號金鑰。

請在提示詞中保留 <API_KEY>。透過本機環境變數或機密管理服務設定 REQUESTBEAM_API_KEY。切勿將憑證貼入 AI 對話、程式碼儲存庫、日誌或螢幕截圖。

AI 適用整合

使用本頁頂端的「AI 適用範本」入口。複製不會選取或取得任何金鑰。

準備連線時，登入主控台並在 API 金鑰中管理憑證。請將憑證安全地存放於提示詞之外。

步驟

登入主控台，在 API 金鑰中建立新金鑰。
將其存放在本機環境變數或機密管理服務中。「複製給 AI」只複製整合指示與 <API_KEY>，不會複製金鑰。
將含預留位置的範本貼入 AI 工具以產生整合程式碼。在明確授權付費或實際驗證之前，請確認模型、請求與預估費用。

工具整合文件

Codex 整合: Codex base_url = https://api.apihop.ai/openai
Claude Code 整合: Claude Code ANTHROPIC_BASE_URL=https://api.apihop.ai/anthropic

文字模型指南

DeepSeek 模型: V4 Flash / V4 Pro
Gemini 模型: 3.6 Flash / 3.7 Flash / 3.8 Flash / 3.5 Flash-Lite
Kimi 模型: Kimi K3
MiniMax 模型: MiniMax M3

公開 API 文件

如果目標 AI 無法直接讀取已渲染的頁面，可以取得機器可讀目錄：/docs/public-api-catalog.json。

模型目錄: GET https://api.apihop.ai/v1/models
錯誤代碼與疑難排解: 錯誤代碼、重試邊界及已遮蔽敏感資訊的診斷範本
通用聊天補全: POST https://api.apihop.ai/v1/chat-completions
OpenAI 多模態回應: POST https://api.apihop.ai/openai/responses
GPT-Image-2 圖像生成: POST https://api.apihop.ai/openai/v1/images/generations
Seedance 2.0 影片生成: POST https://api.apihop.ai/v1/video/seedance
Seedance 2.5 影片生成: POST https://api.apihop.ai/v1/video/seedance
查詢非同步任務: GET https://api.apihop.ai/v1/jobs/{id}
Claude 訊息: POST https://api.apihop.ai/anthropic/v1/messages
依使用者名稱查詢 Twitter 使用者: GET https://api.apihop.ai/v1/twitter/users?username=...
依 ID 查詢 Twitter 使用者: GET https://api.apihop.ai/v1/twitter/users/{id}
查詢 Twitter 使用者時間軸: GET https://api.apihop.ai/v1/twitter/users/{id}/tweets