Langgraph Code Review

LangGraph Code Review

When reviewing LangGraph code, check for these categories of issues.

Anti-confabulation (gate 0 — runs before every other gate)

Before issuing any finding — flag a bug, anti-pattern, or improvement — you MUST echo the exact artifact you are judging, quoted from a source you read in this turn:

• The code finding: its file:line plus the cited code, read freshly now.
• The graph/state code under review: the StateGraph, node, edge, or state-schema snippet your finding depends on, quoted from the file you just read.

The artifact is the only source of truth. Never infer what you are reviewing from the branch name, the working directory, surrounding files, or recollection. If your mental model differs from the freshly read source, the source wins. A finding issued without a same-turn echo of its target is invalid — emit the echo first, or do not emit the finding.

This gate exists because an LLM under contextual priming will confidently flag code that is not in the file. It runs before the gates below.

Review gates (sequenced)

Complete in order. Each step has an objective pass condition before moving on.

1. Locate graph code — Search the review scope for StateGraph, compile(, invoke, ainvoke, add_node, add_edge, add_conditional_edges. Pass: a short list of file paths (or explicit “none in scope” after searching).

2. Map state schema — For each graph state type (TypedDict, BaseModel, etc.), list fields that hold lists, dicts, or messages and whether Annotated + reducers (add_messages, operator.add, …) are present. Pass: every such field is either covered by a reducer pattern below or explicitly flagged as intentional overwrite.

3. Trace persistence — If interrupts, thread_id, or checkpoint APIs appear, follow them to compile(..., checkpointer=...) and invocation config. Pass: behavior matches the interrupt/checkpointer/thread_id guidance below—or you document a concrete mismatch with file:line.

4. Report with evidence — For each finding you will deliver, record file path and line number(s) (or a minimal quoted snippet). Pass: no critical or high-severity issue is stated without that citation.

5. Run the checklist — Use the checklist at the end of this skill; each item is satisfied, not applicable (with reason), or open with evidence. Pass: no item left silently unchecked.

Critical Issues

1. State Mutation Instead of Return

pythonCopy code
# BAD - mutates state directly
def my_node(state: State) -> None:
    state["messages"].append(new_message)  # Mutation!

# GOOD - returns partial update
def my_node(state: State) -> dict:
    return {"messages": [new_message]}  # Let reducer handle it

2. Missing Reducer for List Fields

pythonCopy code
# BAD - no reducer, each node overwrites
class State(TypedDict):
    messages: list  # Will be overwritten, not appended!

# GOOD - reducer appends
class State(TypedDict):
    messages: Annotated[list, operator.add]
    # Or use add_messages for chat:
    messages: Annotated[list, add_messages]

3. Wrong Return Type from Conditional Edge

pythonCopy code
# BAD - returns invalid node name
def router(state) -> str:
    return "nonexistent_node"  # Runtime error!

# GOOD - use Literal type hint for safety
def router(state) -> Literal["agent", "tools", "__end__"]:
    if condition:
        return "agent"
    return END  # Use constant, not string

4. Missing Checkpointer for Interrupts

pythonCopy code
# BAD - interrupt without checkpointer
def my_node(state):
    answer = interrupt("question")  # Will fail!
    return {"answer": answer}

graph = builder.compile()  # No checkpointer!

# GOOD - checkpointer required for interrupts
graph = builder.compile(checkpointer=InMemorySaver())

5. Forgetting Thread ID with Checkpointer

pythonCopy code
# BAD - no thread_id
graph.invoke({"messages": [...]})  # Error with checkpointer!

# GOOD - always provide thread_id
config = {"configurable": {"thread_id": "user-123"}}
graph.invoke({"messages": [...]}, config)

State Schema Issues

6. Using add_messages Without Message Types

pythonCopy code
# BAD - add_messages expects message-like objects
class State(TypedDict):
    messages: Annotated[list, add_messages]

def node(state):
    return {"messages": ["plain string"]}  # May fail!

# GOOD - use proper message types or tuples
def node(state):
    return {"messages": [("assistant", "response")]}
    # Or: [AIMessage(content="response")]

7. Returning Full State Instead of Partial

pythonCopy code
# BAD - returns entire state (may reset other fields)
def my_node(state: State) -> State:
    return {
        "counter": state["counter"] + 1,
        "messages": state["messages"],  # Unnecessary!
        "other": state["other"]          # Unnecessary!
    }

# GOOD - return only changed fields
def my_node(state: State) -> dict:
    return {"counter": state["counter"] + 1}

8. Pydantic State Without Annotations

pythonCopy code
# BAD - Pydantic model without reducer loses append behavior
class State(BaseModel):
    messages: list  # No reducer!

# GOOD - use Annotated even with Pydantic
class State(BaseModel):
    messages: Annotated[list, add_messages]

Graph Structure Issues

9. Missing Entry Point

pythonCopy code
# BAD - no edge from START
builder.add_node("process", process_fn)
builder.add_edge("process", END)
graph = builder.compile()  # Error: no entrypoint!

# GOOD - connect START
builder.add_edge(START, "process")

10. Unreachable Nodes

pythonCopy code
# BAD - orphan node
builder.add_node("main", main_fn)
builder.add_node("orphan", orphan_fn)  # Never reached!
builder.add_edge(START, "main")
builder.add_edge("main", END)

# Check with visualization
print(graph.get_graph().draw_mermaid())

11. Conditional Edge Without All Paths

pythonCopy code
# BAD - missing path in conditional
def router(state) -> Literal["a", "b", "c"]:
    ...

builder.add_conditional_edges("node", router, {"a": "a", "b": "b"})
# "c" path missing!

# GOOD - include all possible returns
builder.add_conditional_edges("node", router, {"a": "a", "b": "b", "c": "c"})
# Or omit path_map to use return values as node names

12. Command Without destinations

pythonCopy code
# BAD - Command return without destinations (breaks visualization)
def dynamic(state) -> Command[Literal["next", "__end__"]]:
    return Command(goto="next")

builder.add_node("dynamic", dynamic)  # Graph viz won't show edges

# GOOD - declare destinations
builder.add_node("dynamic", dynamic, destinations=["next", END])

Async Issues

13. Mixing Sync/Async Incorrectly

pythonCopy code
# BAD - async node called with sync invoke
async def my_node(state):
    result = await async_operation()
    return {"result": result}

graph.invoke(input)  # May not await properly!

# GOOD - use ainvoke for async graphs
await graph.ainvoke(input)
# Or provide both sync and async versions

14. Blocking Calls in Async Context

pythonCopy code
# BAD - blocking call in async node
async def my_node(state):
    result = requests.get(url)  # Blocks event loop!
    return {"result": result}

# GOOD - use async HTTP client
async def my_node(state):
    async with httpx.AsyncClient() as client:
        result = await client.get(url)
    return {"result": result}

Tool Integration Issues

15. Tool Calls Without Corresponding ToolMessage

pythonCopy code
# BAD - AI message with tool_calls but no tool execution
messages = [
    HumanMessage(content="search for X"),
    AIMessage(content="", tool_calls=[{"id": "1", "name": "search", ...}])
    # Missing ToolMessage! Next LLM call will fail
]

# GOOD - always pair tool_calls with ToolMessage
messages = [
    HumanMessage(content="search for X"),
    AIMessage(content="", tool_calls=[{"id": "1", "name": "search", ...}]),
    ToolMessage(content="results", tool_call_id="1")
]

16. Parallel Tool Calls Before Interrupt

pythonCopy code
# BAD - model may call multiple tools including interrupt
model = ChatOpenAI().bind_tools([interrupt_tool, other_tool])
# If both called in parallel, interrupt behavior is undefined

# GOOD - disable parallel tool calls before interrupt
model = ChatOpenAI().bind_tools(
    [interrupt_tool, other_tool],
    parallel_tool_calls=False
)

Checkpointing Issues

17. InMemorySaver in Production

pythonCopy code
# BAD - in-memory checkpointer loses state on restart
graph = builder.compile(checkpointer=InMemorySaver())  # Testing only!

# GOOD - use persistent storage in production
from langgraph.checkpoint.postgres import PostgresSaver
checkpointer = PostgresSaver.from_conn_string(conn_string)
graph = builder.compile(checkpointer=checkpointer)

18. Subgraph Checkpointer Confusion

pythonCopy code
# BAD - subgraph with explicit False prevents persistence
subgraph = sub_builder.compile(checkpointer=False)

# GOOD - use None to inherit parent's checkpointer
subgraph = sub_builder.compile(checkpointer=None)  # Inherits from parent
# Or True for independent checkpointing
subgraph = sub_builder.compile(checkpointer=True)

Performance Issues

19. Large State in Every Update

pythonCopy code
# BAD - returning large data in every node
def node(state):
    large_data = fetch_large_data()
    return {"large_field": large_data}  # Checkpointed every step!

# GOOD - use references or store
from langgraph.store.memory import InMemoryStore

def node(state, *, store: BaseStore):
    store.put(namespace, key, large_data)
    return {"data_ref": f"{namespace}/{key}"}

20. Missing Recursion Limit Handling

pythonCopy code
# BAD - no protection against infinite loops
def router(state):
    return "agent"  # Always loops!

# GOOD - check remaining steps or use RemainingSteps
from langgraph.managed import RemainingSteps

class State(TypedDict):
    messages: Annotated[list, add_messages]
    remaining_steps: RemainingSteps

def check_limit(state):
    if state["remaining_steps"] < 2:
        return END
    return "continue"

Code Review Checklist

1. State schema uses Annotated with reducers for collections
2. Nodes return partial state updates, not mutations
3. Conditional edges return valid node names or END
4. Graph has path from START to all nodes
5. Checkpointer provided if using interrupts
6. Thread ID provided in config when using checkpointer
7. Tool calls paired with ToolMessages
8. Async nodes use async operations
9. Production uses persistent checkpointer
10. Recursion limits considered for loops