anthropic-compat

Drop-in fix for the Claude 4.6 breaking change that removed assistant message prefilling.

If your code does this and now gets a 400:

response = client.messages.create(
    model="claude-opus-4-6",
    messages=[
        {"role": "user", "content": "What is 2+2?"},
        {"role": "assistant", "content": "The answer is:"},
    ],
)

Change one import and it works again:

# before
import anthropic

# after
import anthropic_compat as anthropic

That's it. Everything else stays the same — client, methods, exceptions, all of it.

Install

pip install anthropic-compat

What it does

Intercepts prefilled assistant messages before they hit the API and converts them into system prompt instructions. The model still starts its response where your prefill left off.

Also handles the output_format → output_config.format rename so you don't have to.

What it doesn't do

Monkey-patch anything. It wraps the official SDK and passes everything else through untouched.