Basic Prompt Structure

Messages API: Required Parameters

Every Claude call via the Messages API requires three parameters:

model — the model name (e.g., claude-opus-4-5)
max_tokens — hard limit on response tokens
messages — an array of objects with role and content fields

Roles must alternate, and the array must always start with user.

Correct Format

PROMPT = "Hi Claude, how are you?"
# → correct: {"role": "user", "content": PROMPT}

# Error — two user messages in a row:
messages=[
  {"role": "user", "content": "Celine Dion birth year?"},
  {"role": "user", "content": "Tell me more about her."}
]
# API will return 400 error

System Prompt

The system prompt sets context before the first user message:

SYSTEM_PROMPT = "Your answer should always be a series of critical thinking questions that further the conversation (do not provide answers). Do not actually answer the user question."
PROMPT = "Why is the sky blue?"
# Claude responds with questions, not an explanation

Why This Matters

The system prompt is your primary tool for controlling Claude's behavior. It sets the role, tone, constraints, and output format before the user writes a word.

When to Use Prefill

In the last assistant message, you can start the response for Claude — it will continue from that point:

messages=[
  {"role": "user", "content": "List three fruits."},
  {"role": "assistant", "content": "1."}  # Claude continues: " Apple
2. Banana
3. Cherry"
]

Report a bug

Task grader

Prompt sandbox

Quiz — 3 questions

Discussion

Messages API: Required Parameters

Correct Format

System Prompt

Why This Matters

When to Use Prefill