Copilot-Skills openai.instructions.md

OpenAI Instructions

Auto-loaded when: **/*.js, **/*.ts, **/*.tsx, **/*.jsx, **/*.py, **/openai*

Default Behaviors

When working with OpenAI code:

1. Never hardcode API keys - Always use environment variables
2. Handle rate limits - Implement exponential backoff retry logic
3. Stream long responses - Use streaming for better UX
4. Monitor tokens - Be aware of token costs
5. Error handling - Always wrap API calls in try-catch
6. Use latest models - gpt-4o, gpt-4o-mini, or gpt-5 for best results
7. Safety identifiers - Use safety_identifier (not deprecated user)
8. Optimize caching - Use prompt_cache_key for better performance
9. Reasoning models - o3/o4-mini don't support stop parameter
10. Service tiers - Use service_tier for priority processing

Common Workflows

Initialize OpenAI Client

import os
from openai import OpenAI

client = OpenAI(api_key=os.getenv("OPENAI_API_KEY"))

import OpenAI from "openai";

const client = new OpenAI({
    apiKey: process.env.OPENAI_API_KEY
});

Basic Chat Completion

# Latest recommended approach (Oct 2024)
response = client.chat.completions.create(
    model="gpt-4o",  # or gpt-4o-mini, gpt-5
    messages=[
        {"role": "system", "content": "You are a helpful assistant."},
        {"role": "user", "content": "Hello!"}
    ],
    temperature=0.7,
    max_tokens=100,
    safety_identifier="hashed_user_id",  # New: replaces 'user'
    service_tier="auto"  # New: default, flex, or priority
)

print(response.choices[0].message.content)

Using Reasoning Models

# For complex reasoning tasks
response = client.chat.completions.create(
    model="gpt-5",  # or o3, o4-mini
    messages=[
        {"role": "user", "content": "Solve: If x + 5 = 12, what is x?"}
    ],
    verbosity="low",  # low, medium, or high
    reasoning_effort="medium",  # control thinking depth
    temperature=1.0
    # Note: NO 'stop' parameter on o3/o4-mini
)

print(response.choices[0].message.content)