Python: fix(workflows): support keyword-only RunContext in functional workflows

[Nithin (nithin42)]

Summary

Fixes #9324.

Resolves an issue where defining a @workflow with a keyword-only RunContext parameter (e.g. async def wf(text: str, *, ctx: RunContext) -> str:) succeeded at decoration time but failed during execution with:

TypeError: wf() takes 1 positional argument but 2 were given

[Copilot]

🟢 Approval recommended

The focused binding fix preserves positional behavior, includes regression coverage, and has no identified blocking issues.

0 open findings

What changed in this PR

Fixes #9324 by respecting keyword-only parameters when invoking functional workflows, while preserving positional argument handling.

Changes:

• Pass keyword-only context and input parameters by name.
• Add regression tests for annotated and unannotated context parameters, keyword-only inputs, and streaming.

File	Description
python/​packages/​core/​tests/​workflow/​test_functional_workflow.py	Adds keyword-only parameter regression tests.
python/​packages/​core/​agent_framework/​_workflows/​_functional.py	Injects keyword-only arguments through keyword bindings.

🧠 Review effort: Balanced

---

💡 Add a code-review agent skill for context-aware, tailored reviews. Learn more in the docs.

[Bicheng (Kenneth) (BichengWang)]

Reviewed 8d8806dab43bbc71b13537c59533aab25a177131. I found no correctness blocker in the argument-binding change: keyword-only values are passed by name, while positional-only and ordinary positional parameters keep their existing ordering. Both streaming and non-streaming runs reach the same _execute path.

Offline validation on macOS arm64 / CPython 3.14.7:

• All 159 tests in python/packages/core/tests/workflow/test_functional_workflow.py passed.
• Replacing only _execute in memory with the parent commit's implementation made all four new regression tests fail with the expected positional-argument TypeError; all four pass with this head.
• An additional 20-case harness passed: eight input/context signature arrangements in both streaming modes, plus four HITL resume cases covering response-only continuation and checkpoint restore into a newly built workflow. The resume cases used an explicit replay key and verified that a completed step ran only once.

Non-blocking coverage suggestion: keep a test for an annotated context named something other than ctx, such as async def wf(text: str, *, context: RunContext), and one keyword-only-context HITL checkpoint-resume test. These worked in the additional harness, but the committed tests currently exercise the ctx name fallback and state access only.

I checked the existing review and inline comments before this review. Validation used cached dependencies and local source; I did not run the full workspace suite, type checks, other Python versions, or service integration tests. The issue-first maintainer agreement for #9324 remains a separate prerequisite for the maintainers to resolve.

[Nithin (nithin42)]

Thanks for testing this so thoroughly Bicheng (Kenneth) (@BichengWang)! Really appreciate the feedback. I'll wait to hear what the maintainers think and will happily add those extra tests if they'd like them.