[Reginald Alfret V (reginaldalfret)]

Motivation and Context

Fixes #14359

When a tool function returns an empty string "", ChatHistory.add_tool_message(content="", tool_call_id="...") previously evaluated content as falsy in _prepare_for_add():

if role == AuthorRole.TOOL and content and not items:

Because content was "", this check failed and fell through without creating a FunctionResultContent item and without attaching tool_call_id. The resulting ChatMessageContent had role tool but an empty items list. Subsequently calling ChatMessageContent.to_dict() failed with IndexError: list index out of range when attempting to access self.items[0] to retrieve the tool call ID.

Description

• In python/semantic_kernel/contents/chat_history.py, updated _prepare_for_add to check content is not None instead of truthy content:

  if role == AuthorRole.TOOL and content is not None and not items:

• In python/tests/unit/contents/test_chat_history.py, added a unit test test_add_tool_message_empty_content_to_dict_succeeds verifying that calling add_tool_message("", tool_call_id="call_123") correctly populates a FunctionResultContent with result="" and id="call_123", and that .to_dict() executes cleanly without raising IndexError.

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 fix and regression test address the empty tool-result serialization failure.

Review effort: Lite
Findings: None

What changed in this PR

Fixes preservation and serialization of empty-string tool results in Python chat history.

Changes:

• Preserves empty tool results and their tool call IDs.
• Adds regression coverage for serialization.

File	Description
python/​semantic_kernel/​contents/​chat_history.py	Preserves empty tool results as FunctionResultContent.
python/​tests/​unit/​contents/​test_chat_history.py	Adds regression coverage for empty-result serialization.

---

💡 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. I am not a maintainer, so this does not clear the merge gate.

I read the full diff and ChatHistory._prepare_for_add / add_tool_message in python/semantic_kernel/contents/chat_history.py, then ran the change (python/ in a venv, Python 3.13).

The bug on main: _prepare_for_add used role == AuthorRole.TOOL and content and not items, so add_tool_message("", tool_call_id=...) skipped the FunctionResultContent branch. I ran it on origin/main: it raises IndexError: list index out of range, which says nothing about the cause. With the PR (content is not None), the same call returns a message with one FunctionResultContent item and to_dict() gives {'role': 'tool', 'content': '', 'tool_call_id': 'call_1'}.

What I ran:

• tests/unit/contents on the PR head: 395 passed.
• tests/unit/contents/test_chat_history.py with chat_history.py taken from origin/main: the new test_add_tool_message_empty_content_to_dict_succeeds fails and the other tests in that file pass.
• A probe of add_tool_message with "", "ok" and "0": the last two behave the same on main and the PR. With [] it raises IndexError on both; with None it raises NotImplementedError on both (the singledispatchmethod has no handler for it). I did not change or test anything else for those two, and the PR does not claim to cover them.
• add_user_message(""), add_assistant_message(""), add_system_message("") and add_developer_message("") on the PR head all succeed and produce a message with no items and content=''. So the tool role was the one role where an empty string failed, which matches what the PR fixes.

Why it matters in practice: a tool or function that legitimately returns an empty string is a normal case (for example an empty search result), and the manual path of adding the result with add_tool_message crashed. I grepped semantic_kernel for callers of add_tool_message outside chat_history.py and found none (the two azure_chat_completion.py hits are a different method), so within the library this only affects callers that add tool results themselves.

I did not run the connector tests, any model call, or CI. I have no blocking concerns.

[Reginald Alfret V (reginaldalfret)]

Thank you PRABHU KIRAN VANDRANKI (@VANDRANKI) for the community review and verifying that the unit tests pass cleanly.

The PR addresses #14359 by ensuring empty string results in �dd_tool_message are preserved as FunctionResultContent when content is not None rather than dropped by truthiness checks.

Requesting maintainer approval to run the GitHub Actions workflows (Python Unit Tests, Python Code Quality Checks, etc.) and for official maintainer review.