Эффективная постановка контекста

Эффективные начальные формулировки

Формулировка «Вы…»

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

persona_prompts = [
    'You are a Socratic philosophy professor. Ask 3 probing questions about this claim: AI will replace programmers.',
    'You are a skeptical venture capitalist who has seen 500 pitches. Give brutal feedback on this pitch: We are building an AI writing assistant.',
    'You are a patient kindergarten teacher. Explain what a computer does in 3 sentences for 5-year-olds.'
]

for prompt in persona_prompts:
    response = client.messages.create(
        model='claude-opus-4-5',
        max_tokens=150,
        messages=[{'role': 'user', 'content': prompt}]
    )
    print(f'--- Persona ---')
    print(prompt[:80] + '...')
    print(response.content[0].text.strip())
    print()

Рамка «Вы…» в системных сообщениях

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

# Persona set once in system; stays active for all turns
system_persona = (
    'You are Marcus, a senior software architect at a Fortune 500 company. '
    'You have 20 years of experience with distributed systems. '
    'Your communication style: direct, pragmatic, no buzzwords. '
    'You always ask about scale and failure modes before giving architecture advice. '
    'If a question lacks context, ask one clarifying question before answering.'
)

response = client.chat.completions.create(
    model='gpt-4o',
    messages=[
        {'role': 'system', 'content': system_persona},
        {'role': 'user', 'content': 'Should we use microservices for our new product?'}
    ]
)
print(response.choices[0].message.content)

Рамка «Контекст…»

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

response = client.messages.create(
    model='claude-opus-4-5',
    max_tokens=300,
    messages=[{
        'role': 'user',
        'content': (
            'The context is: we are a 5-person startup with $2M in seed funding. '
            'Our Python monolith handles 10,000 users and is starting to show performance issues. '
            'We have one backend engineer and cannot hire more for 6 months. '
            'We need to choose between refactoring the monolith vs migrating to microservices.\n\n'
            'Give a recommendation with 3 supporting reasons. Be direct.'
        )
    }]
)
print(response.content[0].text)

Рамка «При условии, что…»

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

scenarios = [
    'Given that our user base will grow 10x in 6 months, what architecture changes should we make today?',
    'Given that we must launch in 2 weeks with the current team, which features should we cut from the MVP?',
    'Given that our API key was exposed publicly for 3 hours, what steps should we take in the next 24 hours?'
]

for scenario in scenarios:
    response = client.chat.completions.create(
        model='gpt-4o',
        max_tokens=120,
        messages=[{
            'role': 'user',
            'content': scenario + ' (Answer in 3 bullet points.)'
        }]
    )
    print(f'Scenario: {scenario[:60]}...')
    print(response.choices[0].message.content.strip())
    print()

Рамка «Цель — …»

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

goal_frames = [
    (
        'The goal is to get the reader to schedule a 30-minute demo call. '
        'Write a 100-word cold outreach email for our AI data pipeline tool '
        'targeting data engineers at e-commerce companies.'
    ),
    (
        'The goal is to help the reader pass a senior Python interview at a FAANG company. '
        'Explain Python decorators with one conceptual explanation and one practical code example.'
    )
]

for prompt in goal_frames:
    response = client.messages.create(
        model='claude-opus-4-5',
        max_tokens=200,
        messages=[{'role': 'user', 'content': prompt}]
    )
    print('--- Goal Frame ---')
    print(prompt[:80] + '...')
    print(response.content[0].text.strip())
    print()

Сочетание начальных рамок

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

combined_frame_prompt = (
    'You are a senior product manager with 10 years of B2B SaaS experience.\n'
    'The context is: our team is debating whether to build a native mobile app '
    'or keep investing in our responsive web app.\n'
    'The goal is: to help our leadership team make a clear go/no-go decision at '
    'next week\'s board meeting.\n'
    'Given that: we have 3 engineers, $300k runway, and 85% of current users are on desktop.\n\n'
    'Write a 250-word recommendation memo with a clear position (build or wait) '
    'and 3 supporting arguments. End with one risk to monitor.'
)

response = client.chat.completions.create(
    model='gpt-4o',
    messages=[{'role': 'user', 'content': combined_frame_prompt}]
)
print(response.choices[0].message.content)

Задание тона и манеры речи с помощью рамок

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

frames = [
    'You are a warm, patient mentor. Explain why learning to code is hard but worth it.',
    'You are a no-nonsense military logistics officer. Explain why learning to code is hard but worth it.',
    'You are a witty tech journalist writing for Wired. Explain why learning to code is hard but worth it.'
]

for frame in frames:
    response = client.messages.create(
        model='claude-opus-4-5',
        max_tokens=80,
        messages=[{'role': 'user', 'content': frame + ' (2 sentences only)'}]
    )
    print(f'Frame: {frame[:55]}...')
    print(response.content[0].text.strip())
    print()

Рамки для аналитической строгости

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

analytical_frames = [
    'You are a critical reviewer whose job is to find fatal flaws. Review this startup idea: a subscription box for AI prompt templates.',
    'Play devil\'s advocate. Challenge this claim: AI will make every knowledge worker 10x more productive.',
    'Steelman the weakest argument against remote work, then give the strongest counter-argument.'
]

for frame in analytical_frames:
    response = client.chat.completions.create(
        model='gpt-4o',
        max_tokens=120,
        messages=[{'role': 'user', 'content': frame + ' (3 sentences max)'}]
    )
    print(f'Frame type: analytical/critical')
    print(f'Prompt: {frame[:60]}...')
    print(response.choices[0].message.content.strip())
    print()

Когда NOT использовать рамку персоны

import anthropic

client = anthropic.Anthropic(api_key='sk-ant-your-key-here')

# Persona frame is unhelpful here — just adds tokens
prompt_with_unnecessary_frame = (
    'You are an expert data processing specialist with years of experience. '
    'Convert the following to JSON: Name: Alice, Age: 30, City: London'
)

# Clean, direct instruction
prompt_direct = (
    'Convert to a JSON object with keys name, age, city:\n'
    'Name: Alice, Age: 30, City: London'
)

for label, prompt in [('WITH UNNECESSARY FRAME', prompt_with_unnecessary_frame), ('DIRECT', prompt_direct)]:
    response = client.messages.create(
        model='claude-opus-4-5',
        max_tokens=50,
        messages=[{'role': 'user', 'content': prompt}]
    )
    print(f'[{label}]')
    print(response.content[0].text.strip())
    print()

Проверка ваших рамок

import openai

client = openai.OpenAI(api_key='sk-your-key-here')

task = 'Explain the pros and cons of using TypeScript over JavaScript.'

frames = {
    'No frame':       task,
    'Persona frame':  f'You are a TypeScript advocate who also knows JavaScript deeply. {task}',
    'Goal frame':     f'The goal is to help a JavaScript developer decide if switching to TypeScript is worth it. {task}',
    'Combined frame': f'You are a pragmatic senior engineer. The goal is to help a JavaScript developer decide. {task} Give a balanced view in 3 bullet points.'
}

for label, prompt in frames.items():
    response = client.chat.completions.create(
        model='gpt-4o', max_tokens=80,
        messages=[{'role': 'user', 'content': prompt}]
    )
    print(f'[{label}]: {response.choices[0].message.content.strip()[:120]}...')
    print()

Проверка знаний

Задание контекста — повторение