Skip to content

feat(langchain): Set system instruction attribute #5357

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
alexander-alderman-webb wants to merge 2 commits into
base: master
Choose a base branch
from
+60 −8
Open

feat(langchain): Set system instruction attribute #5357

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
1c87221
feat(langchain): Set system instruction attribute
alexander-alderman-webb Jan 21, 2026
306741e
.
alexander-alderman-webb Jan 22, 2026
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
4 changes: 4 additions & 0 deletions sentry_sdk/_types.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -359,3 +359,7 @@ class SDKInfo(TypedDict):
)

HttpStatusCodeRange = Union[int, Container[int]]

class TextPart(TypedDict):
type: Literal["text"]
content: str
6 changes: 6 additions & 0 deletions sentry_sdk/consts.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -542,6 +542,12 @@ class SPANDATA:
Example: 2048
"""

GEN_AI_SYSTEM_INSTRUCTIONS = "gen_ai.system_instructions"
"""
The system instructions passed to the model.
Example: [{"type": "text", "text": "You are a helpful assistant."},{"type": "text", "text": "Be concise and clear."}]
"""

GEN_AI_REQUEST_MESSAGES = "gen_ai.request.messages"
"""
The messages passed to the model. The "content" can be a string or an array of objects.
Expand Down
36 changes: 36 additions & 0 deletions sentry_sdk/integrations/langchain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -36,6 +36,7 @@
from uuid import UUID

from sentry_sdk.tracing import Span
from sentry_sdk._types import TextPart


try:
Expand Down Expand Up @@ -189,6 +190,29 @@ def _get_current_agent() -> "Optional[str]":
return None


def _get_system_instructions(messages: "List[List[BaseMessage]]") -> "List[TextPart]":
system_instructions = []

for list_ in messages:
for message in list_:
if message.type == "system":
system_instructions.append(message)

return system_instructions


def _transform_system_instructions(
system_instructions: "List[BaseMessage]",
) -> "List[TextPart]":
return [
{
"type": "text",
"content": instruction.content,
}
for instruction in system_instructions
Copy link

@cursor cursor bot Jan 22, 2026

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

System instructions field uses wrong key name

Medium Severity

The _transform_system_instructions function creates dictionaries with "content" as the key, but the docstring example in consts.py and other integrations (anthropic, pydantic_ai) consistently use "text" as the key for text parts. The docstring shows {"type": "text", "text": "You are a helpful assistant."} but the implementation produces {"type": "text", "content": "..."}. This inconsistency will cause the langchain integration to produce data in a different format than other integrations and what's documented.

Additional Locations (2)

Fix in Cursor Fix in Web

]


class LangchainIntegration(Integration):
identifier = "langchain"
origin = f"auto.ai.{identifier}"
Expand Down Expand Up @@ -430,9 +454,21 @@ def on_chat_model_start(
_set_tools_on_span(span, all_params.get("tools"))

if should_send_default_pii() and self.include_prompts:
system_instructions = _get_system_instructions(messages)
if len(system_instructions) > 0:
set_data_normalized(
span,
SPANDATA.GEN_AI_SYSTEM_INSTRUCTIONS,
_transform_system_instructions(system_instructions),
unpack=False,
)

normalized_messages = []
for list_ in messages:
for message in list_:
if message.type == "system":
continue

normalized_messages.append(
self._normalize_langchain_message(message)
)
Expand Down
22 changes: 14 additions & 8 deletions tests/integrations/langchain/test_langchain.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -217,17 +217,21 @@ def test_langchain_agent(
assert chat_spans[1]["data"]["gen_ai.usage.total_tokens"] == 117

if send_default_pii and include_prompts:
assert (
"You are very powerful"
in chat_spans[0]["data"][SPANDATA.GEN_AI_REQUEST_MESSAGES]
)
assert [
{
"type": "text",
"content": "You are very powerful assistant, but don't know current events",
}
] == json.loads(chat_spans[0]["data"][SPANDATA.GEN_AI_SYSTEM_INSTRUCTIONS])
assert "5" in chat_spans[0]["data"][SPANDATA.GEN_AI_RESPONSE_TEXT]
assert "word" in tool_exec_span["data"][SPANDATA.GEN_AI_TOOL_INPUT]
assert 5 == int(tool_exec_span["data"][SPANDATA.GEN_AI_TOOL_OUTPUT])
assert (
"You are very powerful"
in chat_spans[1]["data"][SPANDATA.GEN_AI_REQUEST_MESSAGES]
)
assert [
{
"type": "text",
"content": "You are very powerful assistant, but don't know current events",
}
] == json.loads(chat_spans[1]["data"][SPANDATA.GEN_AI_SYSTEM_INSTRUCTIONS])
assert "5" in chat_spans[1]["data"][SPANDATA.GEN_AI_RESPONSE_TEXT]

# Verify tool calls are recorded when PII is enabled
Expand All @@ -243,8 +247,10 @@ def test_langchain_agent(
tool_call_str = str(tool_calls_data)
assert "get_word_length" in tool_call_str
else:
assert SPANDATA.GEN_AI_SYSTEM_INSTRUCTIONS not in chat_spans[0].get("data", {})
assert SPANDATA.GEN_AI_REQUEST_MESSAGES not in chat_spans[0].get("data", {})
assert SPANDATA.GEN_AI_RESPONSE_TEXT not in chat_spans[0].get("data", {})
assert SPANDATA.GEN_AI_SYSTEM_INSTRUCTIONS not in chat_spans[1].get("data", {})
assert SPANDATA.GEN_AI_REQUEST_MESSAGES not in chat_spans[1].get("data", {})
assert SPANDATA.GEN_AI_RESPONSE_TEXT not in chat_spans[1].get("data", {})
assert SPANDATA.GEN_AI_TOOL_INPUT not in tool_exec_span.get("data", {})
Expand Down
Loading