Checked other resources

• This is a bug, not a usage question.
• I added a clear and descriptive title that summarizes this issue.
• I used the GitHub search to find a similar question and didn't find it.
• I am sure that this is a bug in LangChain rather than my code.
• The bug is not resolved by updating to the latest stable version of LangChain (or the specific integration package).
• This is not related to the langchain-community package.
• I read what a minimal reproducible example is (https://stackoverflow.com/help/minimal-reproducible-example).
• I posted a self-contained, minimal, reproducible example. A maintainer can copy it and run it AS IS.

Example Code

from langchain.agents import create_agent
from langchain_anthropic import ChatAnthropic
# Create model
model = ChatAnthropic(
 model="claude-sonnet-4-20250514",
 api_key="test-key",
)
# Claude Code OAuth requires system_prompt as array of message blocks
system_array = [
 {"type": "text", "text": "You are Claude Code, Anthropic's official CLI for Claude."},
 {"type": "text", "text": "You are a security auditor.", "cache_control": {"type": "ephemeral"}},
]
# This doesn't work - create_agent() only accepts string for system_prompt
agent = create_agent(
 model=model,
 tools=[],
 system_prompt=system_array # TypeError or validation error
)

Error Message and Stack Trace (if applicable)

The silent error is that everything is concatenated into a single message, but anthropic can use

[ 
 { 
 "text": "Today's date is 2024年06月01日.", 
 "type": "text" 
 } 
 ]

Which is needed for some things and how it's supported by vercel ai sdk

Description

I'm trying to use langchain.agents.create_agent() with system prompts that leverage multiple text blocks with different cache control settings.

Expected behavior: create_agent(system_prompt=...) should support both string and array formats, matching the Anthropic API spec:

# String format (current support)
system_prompt="You are a helpful assistant."
# Array format (not currently supported)
system_prompt=[
 {"type": "text", "text": "Base instructions"},
 {"type": "text", "text": "Dynamic instructions", "cache_control": {"type": "ephemeral"}}
]

Actual behavior: create_agent() only accepts system prompts as strings. This prevents using advanced Anthropic API features like separate cache control per text block.

Current workaround: Concatenating all text blocks into a single string:

# Build system prompt as concatenated string
system_prompt = "Base instructions\n\nDynamic instructions"
agent = create_agent(
 model=model,
 tools=[],
 system_prompt=system_prompt # Concatenated string instead of array
)

This works but loses the ability to use separate cache control blocks and proper message structure that the Anthropic API supports.

System Info

langchain-anthropic==0.3.9
langchain-core==0.3.32
anthropic==0.42.0
python==3.12