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

[Ethan T. (gambletan)]

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

[github-actions]

This PR has been automatically closed because you are not assigned to the linked issue.

External contributors must be assigned to an issue before opening a PR for it. Please:

1. Comment on the linked issue to request assignment from a maintainer
2. Once assigned, edit your PR description and the PR will be reopened automatically

[codspeed-hq]

Merging this PR will improve performance by 34.39%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚠️ Different runtime environments detected

Some benchmarks with significant performance changes were compared across different runtime environments,
which may affect the accuracy of the results.

Open the report in CodSpeed to investigate

⚡ 13 improved benchmarks
⏩ 23 skipped benchmarks¹

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	WallTime	test_import_time[tool]	578.6 ms	484.4 ms	+19.46%
⚡	WallTime	test_import_time[RunnableLambda]	534.4 ms	459.2 ms	+16.39%
⚡	WallTime	test_import_time[Runnable]	527.8 ms	455.1 ms	+15.98%
⚡	WallTime	test_import_time[ChatPromptTemplate]	704 ms	568 ms	+23.93%
⚡	WallTime	test_import_time[LangChainTracer]	486.7 ms	432.3 ms	+12.59%
⚡	WallTime	test_import_time[PydanticOutputParser]	567.6 ms	491.2 ms	+15.57%
⚡	WallTime	test_async_callbacks_in_sync	25.1 ms	18.7 ms	+34.39%
⚡	WallTime	test_import_time[InMemoryVectorStore]	659.5 ms	555.9 ms	+18.63%
⚡	WallTime	test_import_time[BaseChatModel]	567.4 ms	495.4 ms	+14.55%
⚡	WallTime	test_import_time[CallbackManager]	331.9 ms	296.6 ms	+11.92%
⚡	WallTime	test_import_time[Document]	196.1 ms	173.4 ms	+13.12%
⚡	WallTime	test_import_time[InMemoryRateLimiter]	186.5 ms	162.6 ms	+14.72%
⚡	WallTime	test_import_time[HumanMessage]	274.8 ms	242 ms	+13.58%

---

_{Comparing gambletan:fix/structured-tool-args-field (9e1c215) with master (307cdca)}

Footnotes

1. 23 benchmarks were skipped, so the baseline results were used instead. If they were deleted from the codebase, click here and archive them to remove them from the performance reports. ↩