langchain-ai/langgraph
1
0
Fork 0
mirror of https://github.com/langchain-ai/langgraph.git synced 2026-07-15 12:35:43 -04:00
Code Issues 314 Packages Projects Releases 17 Wiki Activity

Command(goto="__end__") returned from tool does not stop agent loop in create_agent, causes invalid message order error #1092

New Issue
Closed
opened 2026-02-20 17:43:02 -05:00 by yindo · 6 comments
yindo commented 2026-02-20 17:43:02 -05:00
Owner
Copy Link
Copy Source

Originally created by @errajibadr on GitHub (Dec 12, 2025).

Checked other resources

  • This is a bug, not a usage question. For questions, please use the LangChain Forum (https://forum.langchain.com/).
  • I added a clear and detailed title that summarizes the issue.
  • I read what a minimal reproducible example is (https://stackoverflow.com/help/minimal-reproducible-example).
  • I included a self-contained, minimal example that demonstrates the issue INCLUDING all the relevant imports. The code run AS IS to reproduce the issue.

Example Code

from langchain.agents import create_agent
from langchain.tools import ToolRuntime, tool
from langchain_core.messages import ToolMessage, AIMessage, HumanMessage
from langgraph.types import Command

@tool(description="Request clarification from the user when intent is ambiguous.")
def clarify_user(
    question: str,
    runtime: ToolRuntime,
) -> Command:
    return Command(
        goto="__end__",
        update={
            "messages": [
                ToolMessage(
                    content="success",
                    tool_call_id=runtime.tool_call_id,
                    name="clarify_user"
                ),
                AIMessage(content=question)
            ],
            "awaiting_clarification": True,
        },
    )

agent = create_agent(
    model=your_chat_model,  # e.g., ChatOpenAI()
    tools=[clarify_user],
    system_prompt="For testing purposes, ALWAYS run clarify_user tool."
)

# This should terminate after clarify_user returns Command(goto="__end__")
result = agent.invoke({
    "messages": [HumanMessage(content="hello, run clarify_user tool please")]
})
print(result)

Error Message and Stack Trace (if applicable)

File ".../openai/_base_client.py", line 1047, in request
    raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {
    'object': 'error', 
    'message': 'Expected last role User or Tool (or Assistant with prefix True) for serving but got assistant', 
    'type': 'invalid_request_message_order', 
    'param': None, 
    'code': '3230'
}

Description

Issue Body
Description
When using create_agent from LangGraph 1.0, returning a Command(goto="end", ...) from a tool does not terminate the graph as expected. Instead, the agent node is invoked again after the tool execution, which causes an OpenAI API error due to invalid message ordering.

Expected Behavior
Returning Command(goto="end", ...) from a tool should immediately terminate the graph and prevent further model nodes (LLM calls) .

Actual Behavior
The graph continues to the model/agent node after the tool returns the Command, causing an API error:

openai.BadRequestError: Error code: 400 - {'object': 'error', 'message': 'Expected last role User or Tool (or Assistant with prefix True) for serving but got assistant', 'type': 'invalid_request_message_order', 'param': None, 'code': '3230'}
This happens because:

The tool updates state with [ToolMessage, AIMessage]
The graph does not respect goto="end" and proceeds to call the model node
The model sees the last message is an AIMessage, which violates OpenAI's message ordering constraints
Reproduction
from langchain.agents import create_agent
from langchain.tools import ToolRuntime, tool
from langchain_core.messages import ToolMessage, AIMessage, HumanMessage
from langgraph.types import Command

@tool(description="Request clarification from the user when intent is ambiguous.")
def clarify_user(
question: str,
runtime: ToolRuntime,
) -> Command:
return Command(
goto="end",
update={
"messages": [
ToolMessage(
content="success",
tool_call_id=runtime.tool_call_id,
name="clarify_user"
),
AIMessage(content=question)
],
"awaiting_clarification": True,
},
)

agent = create_agent(
model=your_chat_model, # e.g., ChatOpenAI()
tools=[clarify_user],
system_prompt="For testing purposes, ALWAYS run clarify_user tool."
)

This should terminate after clarify_user returns Command(goto="end")

result = agent.invoke({
"messages": [HumanMessage(content="hello, run clarify_user tool please")]
})
print(result)
Error Traceback
File ".../openai/_base_client.py", line 1047, in request
raise self._make_status_error_from_response(err.response) from None
openai.BadRequestError: Error code: 400 - {
'object': 'error',
'message': 'Expected last role User or Tool (or Assistant with prefix True) for serving but got assistant',
'type': 'invalid_request_message_order',
'param': None,
'code': '3230'
}
Current Workaround
Using a before_model middleware to check if the last message is an AIMessage and manually jump to end:

from langchain.agents.middleware import AgentMiddleware

@jump_config(can_jump_to=["end"])
def before_model_check(state, config):
if state["messages"] and isinstance(state["messages"][-1], AIMessage):
return {"jump_to": "end"}
return None

Apply middleware to agent

This workaround is not ideal as it relies on message type inspection rather than proper Command handling.

Environment
langgraph version: 1.0.x (please specify exact version)
langchain version: >1.0
Python version: 3.12
OS: macOS
Questions
Is Command(goto="end") supposed to work when returned from a tool inside create_agent?
If so, is this a bug in the prebuilt agent's edge routing logic?
If not, what is the recommended pattern for terminating an agent graph early from within a tool?

System Info

System Information

OS: Darwin
OS Version: Darwin Kernel Version 23.6.0: Wed May 14 13:52:22 PDT 2025; root:xnu-10063.141.1.705.2~2/RELEASE_ARM64_T6000
Python Version: 3.12.7 (main, Oct 16 2024, 07:12:08) [Clang 18.1.8 ]

Package Information

langchain_core: 1.1.0
langchain: 1.1.0
langchain_community: 0.4.1
langsmith: 0.4.49
langchain_anthropic: 1.2.0
langchain_classic: 1.0.0
langchain_mcp_adapters: 0.1.14
langchain_openai: 1.1.0
langchain_text_splitters: 1.0.0
langgraph_api: 0.5.27
langgraph_cli: 0.4.7
langgraph_runtime_inmem: 0.19.0
langgraph_sdk: 0.2.10

Optional packages not installed

langserve

Other Dependencies

aiohttp: 3.13.2
anthropic: 0.75.0
blockbuster: 1.5.25
click: 8.3.1
cloudpickle: 3.1.2
cryptography: 44.0.3
dataclasses-json: 0.6.7
grpcio: 1.76.0
grpcio-tools: 1.75.1
httpx: 0.28.1
httpx-sse: 0.4.3
jsonpatch: 1.33
jsonschema-rs: 0.29.1
langgraph: 1.0.4
langgraph-checkpoint: 3.0.1
mcp: 1.22.0
numpy: 2.3.5
openai: 1.109.1
opentelemetry-api: 1.38.0
opentelemetry-exporter-otlp-proto-http: 1.38.0
opentelemetry-sdk: 1.38.0
orjson: 3.11.4
packaging: 25.0
protobuf: 6.33.1
pydantic: 2.12.5
pydantic-settings: 2.12.0
pyjwt: 2.10.1
pytest: 9.0.1
python-dotenv: 1.2.1
pyyaml: 6.0.3
PyYAML: 6.0.3
requests: 2.32.5
requests-toolbelt: 1.0.0
rich: 14.2.0
sqlalchemy: 2.0.44
SQLAlchemy: 2.0.44
sse-starlette: 2.1.3
starlette: 0.50.0
structlog: 25.5.0
tenacity: 9.1.2
tiktoken: 0.12.0
truststore: 0.10.4
typing-extensions: 4.15.0
uvicorn: 0.38.0
watchfiles: 1.1.1
zstandard: 0.25.0

Originally created by @errajibadr on GitHub (Dec 12, 2025). ### Checked other resources - [x] This is a bug, not a usage question. For questions, please use the LangChain Forum (https://forum.langchain.com/). - [x] I added a clear and detailed title that summarizes the issue. - [x] I read what a minimal reproducible example is (https://stackoverflow.com/help/minimal-reproducible-example). - [x] I included a self-contained, minimal example that demonstrates the issue INCLUDING all the relevant imports. The code run AS IS to reproduce the issue. ### Example Code ```python from langchain.agents import create_agent from langchain.tools import ToolRuntime, tool from langchain_core.messages import ToolMessage, AIMessage, HumanMessage from langgraph.types import Command @tool(description="Request clarification from the user when intent is ambiguous.") def clarify_user( question: str, runtime: ToolRuntime, ) -> Command: return Command( goto="__end__", update={ "messages": [ ToolMessage( content="success", tool_call_id=runtime.tool_call_id, name="clarify_user" ), AIMessage(content=question) ], "awaiting_clarification": True, }, ) agent = create_agent( model=your_chat_model, # e.g., ChatOpenAI() tools=[clarify_user], system_prompt="For testing purposes, ALWAYS run clarify_user tool." ) # This should terminate after clarify_user returns Command(goto="__end__") result = agent.invoke({ "messages": [HumanMessage(content="hello, run clarify_user tool please")] }) print(result) ``` ### Error Message and Stack Trace (if applicable) ```shell File ".../openai/_base_client.py", line 1047, in request raise self._make_status_error_from_response(err.response) from None openai.BadRequestError: Error code: 400 - { 'object': 'error', 'message': 'Expected last role User or Tool (or Assistant with prefix True) for serving but got assistant', 'type': 'invalid_request_message_order', 'param': None, 'code': '3230' } ``` ### Description Issue Body Description When using create_agent from LangGraph 1.0, returning a Command(goto="__end__", ...) from a tool does not terminate the graph as expected. Instead, the agent node is invoked again after the tool execution, which causes an OpenAI API error due to invalid message ordering. Expected Behavior Returning Command(goto="__end__", ...) from a tool should immediately terminate the graph and prevent further model nodes (LLM calls) . Actual Behavior The graph continues to the model/agent node after the tool returns the Command, causing an API error: openai.BadRequestError: Error code: 400 - {'object': 'error', 'message': 'Expected last role User or Tool (or Assistant with prefix True) for serving but got assistant', 'type': 'invalid_request_message_order', 'param': None, 'code': '3230'} This happens because: The tool updates state with [ToolMessage, AIMessage] The graph does not respect goto="__end__" and proceeds to call the model node The model sees the last message is an AIMessage, which violates OpenAI's message ordering constraints Reproduction from langchain.agents import create_agent from langchain.tools import ToolRuntime, tool from langchain_core.messages import ToolMessage, AIMessage, HumanMessage from langgraph.types import Command @tool(description="Request clarification from the user when intent is ambiguous.") def clarify_user( question: str, runtime: ToolRuntime, ) -> Command: return Command( goto="__end__", update={ "messages": [ ToolMessage( content="success", tool_call_id=runtime.tool_call_id, name="clarify_user" ), AIMessage(content=question) ], "awaiting_clarification": True, }, ) agent = create_agent( model=your_chat_model, # e.g., ChatOpenAI() tools=[clarify_user], system_prompt="For testing purposes, ALWAYS run clarify_user tool." ) # This should terminate after clarify_user returns Command(goto="__end__") result = agent.invoke({ "messages": [HumanMessage(content="hello, run clarify_user tool please")] }) print(result) Error Traceback File ".../openai/_base_client.py", line 1047, in request raise self._make_status_error_from_response(err.response) from None openai.BadRequestError: Error code: 400 - { 'object': 'error', 'message': 'Expected last role User or Tool (or Assistant with prefix True) for serving but got assistant', 'type': 'invalid_request_message_order', 'param': None, 'code': '3230' } Current Workaround Using a before_model middleware to check if the last message is an AIMessage and manually jump to __end__: from langchain.agents.middleware import AgentMiddleware @jump_config(can_jump_to=["end"]) def before_model_check(state, config): if state["messages"] and isinstance(state["messages"][-1], AIMessage): return {"jump_to": "end"} return None # Apply middleware to agent This workaround is not ideal as it relies on message type inspection rather than proper Command handling. Environment langgraph version: 1.0.x (please specify exact version) langchain version: >1.0 Python version: 3.12 OS: macOS Questions Is Command(goto="__end__") supposed to work when returned from a tool inside create_agent? If so, is this a bug in the prebuilt agent's edge routing logic? If not, what is the recommended pattern for terminating an agent graph early from within a tool? ### System Info System Information ------------------ > OS: Darwin > OS Version: Darwin Kernel Version 23.6.0: Wed May 14 13:52:22 PDT 2025; root:xnu-10063.141.1.705.2~2/RELEASE_ARM64_T6000 > Python Version: 3.12.7 (main, Oct 16 2024, 07:12:08) [Clang 18.1.8 ] Package Information ------------------- > langchain_core: 1.1.0 > langchain: 1.1.0 > langchain_community: 0.4.1 > langsmith: 0.4.49 > langchain_anthropic: 1.2.0 > langchain_classic: 1.0.0 > langchain_mcp_adapters: 0.1.14 > langchain_openai: 1.1.0 > langchain_text_splitters: 1.0.0 > langgraph_api: 0.5.27 > langgraph_cli: 0.4.7 > langgraph_runtime_inmem: 0.19.0 > langgraph_sdk: 0.2.10 Optional packages not installed ------------------------------- > langserve Other Dependencies ------------------ > aiohttp: 3.13.2 > anthropic: 0.75.0 > blockbuster: 1.5.25 > click: 8.3.1 > cloudpickle: 3.1.2 > cryptography: 44.0.3 > dataclasses-json: 0.6.7 > grpcio: 1.76.0 > grpcio-tools: 1.75.1 > httpx: 0.28.1 > httpx-sse: 0.4.3 > jsonpatch: 1.33 > jsonschema-rs: 0.29.1 > langgraph: 1.0.4 > langgraph-checkpoint: 3.0.1 > mcp: 1.22.0 > numpy: 2.3.5 > openai: 1.109.1 > opentelemetry-api: 1.38.0 > opentelemetry-exporter-otlp-proto-http: 1.38.0 > opentelemetry-sdk: 1.38.0 > orjson: 3.11.4 > packaging: 25.0 > protobuf: 6.33.1 > pydantic: 2.12.5 > pydantic-settings: 2.12.0 > pyjwt: 2.10.1 > pytest: 9.0.1 > python-dotenv: 1.2.1 > pyyaml: 6.0.3 > PyYAML: 6.0.3 > requests: 2.32.5 > requests-toolbelt: 1.0.0 > rich: 14.2.0 > sqlalchemy: 2.0.44 > SQLAlchemy: 2.0.44 > sse-starlette: 2.1.3 > starlette: 0.50.0 > structlog: 25.5.0 > tenacity: 9.1.2 > tiktoken: 0.12.0 > truststore: 0.10.4 > typing-extensions: 4.15.0 > uvicorn: 0.38.0 > watchfiles: 1.1.1 > zstandard: 0.25.0
yindo added the bugpending labels 2026-02-20 17:43:02 -05:00
yindo closed this issue 2026-02-20 17:43:02 -05:00
yindo commented 2026-02-20 17:43:02 -05:00
Author
Owner
Copy Link
Copy Source

@KarthikeyaKollu commented on GitHub (Dec 12, 2025):

@errajibadr made a PR on the same issue let me know if that works for you!!

@KarthikeyaKollu commented on GitHub (Dec 12, 2025): @errajibadr made a PR on the same issue let me know if that works for you!!
yindo commented 2026-02-20 17:43:02 -05:00
Author
Owner
Copy Link
Copy Source

@errajibadr commented on GitHub (Dec 12, 2025):

i'm mitigated on your solution. which is basically my Troubleshoot.

It doesn't really respect the
Command(goto="end")

It just uses a trick based on last AI message for circuit breaker. which is not what i want nor the Expected behavior.
Command goto should be respected whatever the last message

But i really need command to work. for other edge cases i need to implement.

@errajibadr commented on GitHub (Dec 12, 2025): i'm mitigated on your solution. which is basically my Troubleshoot. It doesn't really respect the Command(goto="__end__") It just uses a trick based on last AI message for circuit breaker. which is not what i want nor the Expected behavior. Command goto should be respected whatever the last message But i really need command to work. for other edge cases i need to implement.
yindo commented 2026-02-20 17:43:02 -05:00
Author
Owner
Copy Link
Copy Source

@Zi-Ling commented on GitHub (Dec 23, 2025):

This feels like an execution boundary issue.

If a tool explicitly returns a terminal command,
continuing the loop seems more harmful than helpful.

Is the intention here that stop signals are best-effort,
or should they always hard-interrupt the agent loop?

@Zi-Ling commented on GitHub (Dec 23, 2025): This feels like an execution boundary issue. If a tool explicitly returns a terminal command, continuing the loop seems more harmful than helpful. Is the intention here that stop signals are best-effort, or should they always hard-interrupt the agent loop?
yindo commented 2026-02-20 17:43:02 -05:00
Author
Owner
Copy Link
Copy Source

@sydney-runkle commented on GitHub (Jan 9, 2026):

This is expected actually - a Command with goto adds a dynamic edge, but the other static edges still execute, hence the continuation of the agent loop. You might want to use return_direct=True on this tool instead.

@sydney-runkle commented on GitHub (Jan 9, 2026): This is expected actually - a `Command` with `goto` adds a dynamic edge, but the other static edges still execute, hence the continuation of the agent loop. You might want to use `return_direct=True` on this tool instead.
yindo commented 2026-02-20 17:43:02 -05:00
Author
Owner
Copy Link
Copy Source

@Prudhvi0717 commented on GitHub (Jan 21, 2026):

@sydney-runkle return_direct doesn't work here. The Command.update added an AIMessage and https://github.com/langchain-ai/langgraph/blob/main/libs/prebuilt/langgraph/prebuilt/chat_agent_executor.py#L964-L977 will return entrypoint -> which doesn't end the graph

@Prudhvi0717 commented on GitHub (Jan 21, 2026): @sydney-runkle `return_direct` doesn't work here. The `Command.update` added an `AIMessage` and https://github.com/langchain-ai/langgraph/blob/main/libs/prebuilt/langgraph/prebuilt/chat_agent_executor.py#L964-L977 will return `entrypoint` -> which doesn't end the graph
yindo commented 2026-02-20 17:43:02 -05:00
Author
Owner
Copy Link
Copy Source

@arthurdorigueto commented on GitHub (Jan 29, 2026):

Maybe AgentState's "jump_to" should be usable by tools?

@arthurdorigueto commented on GitHub (Jan 29, 2026): Maybe `AgentState`'s `"jump_to"` should be usable by tools?
Sign in to join this conversation.
No Branch/Tag Specified
Branches Tags
Labels
Clear labels
agents
awaiting author revision
bug
change
chat models
checkpointer
ci
code quality
code quality
dependencies
documentation
drawing
duplicate
enhancement
good first issue
help wanted
invalid
maintainer
meta
missing information
not maintained by langchain
open-swe-auto
open-swe-max
open-swe-max-auto
pending
performance
pull-request
Mirrored from GitHub Pull Request
 question
requires design
sdk
server
stale
studio
tests
type checking
usage question
v0.5
v1.1
No Label bug pending
Milestone
No items
No Milestone
Projects
Clear projects
No project
Assignees
Clear assignees
yindo
No Assignees
1 Participants
Notifications
Due Date
No due date set.
Dependencies

No dependencies set.

Reference: langchain-ai/langgraph#1092
Delete Branch "%!s()"

Deleting a branch is permanent. Although the deleted branch may continue to exist for a short time before it actually gets removed, it CANNOT be undone in most cases. Continue?