OpenAI 導入マニュアル

# AGENTS.md

## Setup and commands
- Install dependencies with `pnpm install`.
- Run unit tests with `pnpm test`.
- Run lint with `pnpm lint`.

## Working agreements
- Before editing, summarize the design and propose a plan.
- Make the smallest safe change.
- Do not add dependencies without approval.
- Do not modify database migrations.

## Definition of done
- Tests, lint, and type checks pass.
- Changed files and rationale are summarized.
- Any risk or manual verification step is listed.

# pip install openai / OPENAI_API_KEY を環境変数に設定
from openai import OpenAI

client = OpenAI()

response = client.responses.create(
    model="gpt-5.5",
    instructions=(
        "あなたは日本語のカスタマーサポートAIです。"
        "根拠がない場合は推測せず、"
        "必要なら問い合わせ窓口への案内を提案します。"
    ),
    input="領収書を再発行したいです。"
)

print(response.output_text)

from typing import Literal
from pydantic import BaseModel
from openai import OpenAI

class TicketClassification(BaseModel):
    category: Literal["billing","login","bug","other"]
    priority: Literal["low","medium","high"]
    summary: str          # 日本語で50字以内の要約
    needs_human: bool
    reason: str

client = OpenAI()
response = client.responses.create(
    model="gpt-5.5",
    input="請求書の宛名を変更したいがエラーになります。",
    text={"format": {
        "type": "json_schema",
        "name": "ticket_classification",
        "schema": TicketClassification.model_json_schema(),
        "strict": True,
    }},
)
print(response.output_text)  # パースして検証する

{
  "type": "function",
  "name": "request_refund",
  "description": "返金申請を作成する。即時実行しない。
    ユーザー確認後に申請のみ作成する。",
  "parameters": {
    "type": "object",
    "properties": {
      "order_id":        {"type": "string"},
      "amount":          {"type": "number"},
      "reason":          {"type": "string"},
      "idempotency_key": {"type": "string"}
    },
    "required": ["order_id","amount","reason",
                 "idempotency_key"],
    "additionalProperties": false
  },
  "strict": true
}

# 1. Vector Storeを作成
vector_store = client.vector_stores.create(
    name="internal-handbook")

# 2. ファイルを追加
file = client.files.create(
    file=open("employee_handbook.pdf", "rb"),
    purpose="assistants")
client.vector_stores.files.create(
    vector_store_id=vector_store.id, file_id=file.id)

# 3. File Searchツールを使って回答
response = client.responses.create(
    model="gpt-5.5",
    tools=[{"type": "file_search",
            "vector_store_ids": [vector_store.id]}],
    input="有給休暇の申請期限を、根拠付きで説明して。",
)

import { Agent, run } from "@openai/agents";

const billingAgent = new Agent({
  name: "Billing Agent",
  instructions: "請求・領収書・返金申請に対応します。",
});

const techAgent = new Agent({
  name: "Technical Support Agent",
  instructions: "ログイン・エラー等の技術的質問に対応します。",
});

const triageAgent = Agent.create({
  name: "Triage Agent",
  instructions: "内容を判断し、適切な専門Agentに引き継ぎます。",
  handoffs: [billingAgent, techAgent],
});

const result = await run(triageAgent,
  "請求書の宛名変更ができません");
console.log(result.finalOutput);

# 手順:
# 1. 規程/FAQを文書種別・部署・版・公開日で整理
# 2. Vector Storeに投入しメタデータを付ける
# 3. instructionsで根拠外回答を禁止
# 4. 権限に応じて検索対象を絞る
# 5. 代表質問でRAG評価を行う

response = client.responses.create(
    model="gpt-5.5",
    instructions=(
        "社内文書に根拠がない場合は"
        "「確認できない」と答える。"
    ),
    tools=[{"type": "file_search",
            "vector_store_ids": [store_id]}],
    input=user_question,
)