PR #35799: fix(core): remove v__args sentinel from StructuredTool schema

• Repository: langchain-ai/langchain
• Author: MaybeSam05
• State: closed | merged: False
• Link: https://github.com/langchain-ai/langchain/pull/35799

Description (problem / solution / changelog)

Fixes #35796

StructuredTool.from_function was including Pydantic’s internal v__args sentinel in the generated args schema when the wrapped function had a parameter named args, causing the real args parameter to be dropped and tool schemas to be invalid for some backends. This PR filters out the v__args sentinel while preserving a genuine args parameter and adds a regression test to cover this case.

How did you verify your code works?

• Created a virtual environment and installed langchain-core + test deps.

• Ran:

  source .venv/bin/activate
  pytest libs/core/tests/unit_tests/test_tools.py -k "structured_tool_from_function_with_args_param"

Changed files

• libs/core/langchain_core/tools/base.py (modified, +10/-1)
• libs/core/tests/unit_tests/test_tools.py (modified, +20/-0)

PR #35808: fix(core): restore original param name when function has arg named 'args'

• Repository: langchain-ai/langchain
• Author: giulio-leone
• State: closed | merged: False
• Link: https://github.com/langchain-ai/langchain/pull/35808

Description (problem / solution / changelog)

Summary

Fixes #35796

When StructuredTool.from_function wraps a function with a parameter named args (or kwargs), Pydantic's ValidatedFunction renames it to v__args (v__kwargs) to avoid clashing with its own sentinel fields. The resulting JSON schema exposes the mangled name instead of the original, which:

1. Hides the real parameter from LLMs — the model never learns the parameter exists
2. Breaks Gemini API — the spurious v__args field has {type: array, items: {}}, which Gemini rejects with INVALID_ARGUMENT: missing field items

Root Cause

create_schema_from_function already strips Pydantic's sentinel args/kwargs fields when the function doesn't use *args/**kwargs. However, it didn't handle the reverse rename: when a regular parameter named args exists, Pydantic creates v__args — and that slipped through the filter.

Fix

1. In create_schema_from_function: detect when the function has a regular parameter named args (or kwargs) and build a field_renames map (v__args → args)
2. In _create_subset_model_v2: apply the rename when constructing the output model, so the JSON schema uses the original parameter name

Verification

from langchain_core.tools import StructuredTool

def run_command(working_dir: str, args: list[str], timeout: int = 60) -> str:
    """Run a command."""
    ...

st = StructuredTool.from_function(run_command)
schema = st.args_schema.model_json_schema()
print(list(schema['properties'].keys()))
# Before: ['working_dir', 'timeout', 'v__args']  ← spurious
# After:  ['working_dir', 'timeout', 'args']      ← correct

All existing test_tools.py tests pass (31/31 sync tests — the 1 async failure is pre-existing, missing pytest-asyncio dep).

Changed files

• libs/core/langchain_core/tools/base.py (modified, +22/-1)
• libs/core/langchain_core/utils/pydantic.py (modified, +8/-3)

PR #35815: fix: handle 'args' parameter name conflict in StructuredTool schema generation

• Repository: langchain-ai/langchain
• Author: gambletan
• State: closed | merged: False
• Link: https://github.com/langchain-ai/langchain/pull/35815

Description (problem / solution / changelog)

Summary

• Fixes StructuredTool.from_function injecting a spurious v__args field and dropping the legitimate args parameter when a function parameter is literally named args
• Adds filtering for Pydantic's internal v__args/v__kwargs sentinel fields that leak into the schema when parameter names conflict with Pydantic internals
• Preserves backward compatibility: functions with *args/**kwargs (VAR_POSITIONAL/VAR_KEYWORD) continue to work as before

Root cause

In create_schema_from_function, the check not has_args and field == "args" filters out any field named args when the function has no *args. But when a function has a regular parameter named args (e.g., def run_command(args: list[str])), Pydantic's ValidatedFunction keeps it as args in the model while renaming its own internal sentinel to v__args. The old code:

1. Incorrectly filtered out the legitimate args field
2. Let the spurious v__args sentinel pass through into the schema

Fix

1. Before filtering args/kwargs fields, check whether the function actually has a regular parameter with that name
2. Always filter out v__args and v__kwargs (Pydantic's renamed internal sentinels)

Test plan

• Verify StructuredTool.from_function with a function having args: list[str] produces schema with args and without v__args
• Verify functions with *args (VAR_POSITIONAL) still work correctly
• Verify functions without any args parameter still work correctly
• Verify same behavior for kwargs parameter name

Fixes #35796

🤖 Generated with Claude Code

Changed files

• libs/core/langchain_core/tools/base.py (modified, +14/-2)

PR #35826: fix: Strip v__args/v__kwargs Pydantic v2 internal fields

• Repository: langchain-ai/langchain
• Author: LincolnBurrows2017
• State: closed | merged: False
• Link: https://github.com/langchain-ai/langchain/pull/35826

Description (problem / solution / changelog)

Summary

Fixes issue #35796: StructuredTool.from_function injects spurious v__args field when function has param named args.

Problem

The code checked for "args" and "kwargs" but Pydantic v2 creates "v__args" and "v__kwargs" internal fields.

Fix

Changed to filter the Pydantic v2 internal field names:

-if not has_args and field == "args":
+if not has_args and field == "v__args":  # Pydantic v2 internal field
     continue
-if not has_kwargs and field == "kwargs":
+if not has_kwargs and field == "v__kwargs":
     continue

Testing

Added regression test: test_structured_tool_args_param_not_injected

Changed files

• libs/core/langchain_core/tools/base.py (modified, +3/-2)
• libs/core/tests/unit_tests/test_tools.py (modified, +21/-0)