[fhgffy]

Motivation and Context

A kernel function returning an empty list raises IndexError when called through as_agent_framework_tool(). This is a normal result for tools such as searches with no matches: the wrapper attempts to read results[0] after serializing zero items.

Description

Serialize empty and multi-item lists as JSON arrays, while keeping the existing single-item unwrapping and no-result behavior. Add four regression cases for empty, single-item and multi-item lists, plus None.

Validation on Windows / Python 3.12:

• The empty-list case fails on the original code; the other three cases pass.
• tests/unit/functions: 189 passed after the change.
• Ruff check and format checks pass for both changed files; wheel and sdist builds pass.
• The test runs the real SK invocation path and substitutes only the optional AIFunction constructor. It does not exercise a live Agent Framework integration.

The full test run stops during collection because the optional Bedrock dependency boto3 is absent. The native mypy check reports 27 diagnostics; all 27 match the unmodified baseline. Neither check is reported as passing.

Contribution Checklist

• The code builds clean without any errors or warnings
• The PR follows the SK Contribution Guidelines and the pre-submission formatting script raises no violations
• All unit tests pass, and I have added new tests where possible
• I didn't break anyone 😄

[Copilot]

Copilot review overview

🟢 Approval recommended

The focused fix resolves the reported exception while preserving documented behavior with adequate regression coverage.

Review effort: Balanced
Findings: None

What changed in this PR

Fixes Agent Framework tool serialization for kernel functions returning empty lists.

Changes:

• Serializes empty and multi-item lists as JSON arrays.
• Preserves single-item unwrapping and None behavior.
• Adds regression coverage for all relevant list cardinalities.

File	Description
python/​semantic_kernel/​functions/​kernel_function.py	Handles empty lists without indexing into them.
python/​tests/​unit/​functions/​test_kernel_function_from_method.py	Tests empty, single-item, multi-item, and None results.

---

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

[PRABHU KIRAN VANDRANKI (VANDRANKI)]

Community review. This does not clear the merge gate. I ran the real KernelFunction.as_agent_framework_tool wrapper on upstream main and on this PR head (python 3.13, semantic-kernel python/ from the checkout), with a stub agent_framework module providing AIFunction, and called the returned tool.func(). I did not run the PR's pytest file.

The fix is correct and minimal. On main, a function returning [] makes the wrapper raise IndexError: list index out of range (results is empty and len(results) > 1 is false, so it takes json.dumps(results[0])). With != 1 the empty list is serialized as "[]".

Results per return value (main, then PR head):

function returns	main	PR head
[]	IndexError	'[]'
["one"]	'"one"'	'"one"'
["one", "two"]	'["one", "two"]'	same
[{"a": 1}]	'{"a": 1}'	same
[[1, 2]]	'[1, 2]'	same
[[]]	'[]'	'[]'
None	"The function did not return a result."	same

So nothing regresses and the crash is gone.

Not caused by this PR, but the new test pins it and the table shows it: a one-element list is unwrapped while any other length is not. So the agent cannot tell ["one"] from "one", [[1, 2]] from [1, 2], or [[]] from []; with this PR [] and [[]] now produce the same text. If the wrapper's contract is "return the JSON of what the function returned", the unwrap branch is the odd part. I would not expect this PR to change it, but the added (["one"], '"one"') test case makes the current behavior look intentional, so it is worth a sentence from the maintainers on whether it is.