Propagate message.AuthorName to the OpenAI chat request name field

[PratikDhanave (PratikDhanave)]

What

buildMessageParam in provider/openaiprovider/chat.go never read msg.AuthorName, so the participant name was silently dropped from Chat Completions requests. System, user, and assistant params are now built as struct literals (rather than the openai.SystemMessage/UserMessage convenience helpers) so the optional name field can be set. A new sanitizeAuthorName helper produces the value; tool messages are left unchanged since they have no name field.

Why

message.Message.AuthorName is populated by the multi-agent / group-chat hosting flows (agentworkflow sets it), and the OpenAI SDK exposes Name on the System/User/Assistant message params to differentiate participants of the same role. The Go port dropped it, losing attribution.

This restores parity with the .NET OpenAIChatClient, which sets ParticipantName = SanitizeAuthorName(AuthorName). sanitizeAuthorName mirrors that behavior: it returns empty for empty/whitespace input, keeps only alphanumeric characters, and caps the result at 64 characters, so blank or fully-disallowed names leave the field unset.

Tests

Added black-box tests to chat_test.go driving the exported Run API through the existing fake-transport harness and asserting the outgoing request body:

• TestChatAuthorNamePropagation_NonStreaming — AuthorName: "Agent One" on system/user/assistant messages surfaces as "name": "AgentOne".
• TestChatAuthorNameSanitizationAndTruncation_NonStreaming — a disallowed char plus a 70-char name is stripped and truncated to 64 runes.
• TestChatAuthorNameEmpty_NonStreaming — empty / whitespace-only names leave name unset.

These fail before the change and pass after. go build ./..., go vet ./provider/openaiprovider/..., and go test ./provider/openaiprovider/... are green.

[Copilot]

Pull request overview

This PR restores propagation of message.Message.AuthorName into the OpenAI Chat Completions request message "name" field for system/user/assistant roles in openaiprovider, aligning Go behavior with the multi-agent/group-chat hosting flows that populate AuthorName.

Changes:

• Build system/user/assistant chat message params as struct literals so the optional name field can be set.
• Add sanitizeAuthorName helper to trim/strip disallowed characters and cap names at 64 characters, leaving the field unset when the result is empty.
• Add black-box request-body tests (non-streaming) asserting name propagation, sanitization/truncation, and empty-name behavior.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
provider/openaiprovider/chat.go	Propagates sanitized AuthorName into OpenAI message params for system/user/assistant and adds the sanitizer helper.
provider/openaiprovider/chat_test.go	Adds request-body assertions verifying name propagation/sanitization through the exported Run API.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[github-actions]

Cross-repo parity review — one issue found

This PR correctly ports the AuthorName → OpenAI Chat Completions name field propagation from .NET and Python. The overall approach (struct literals, conditional Name field, skip tool messages, 64-char cap) is well-aligned with both upstream implementations.

One parity gap: underscore handling in sanitizeAuthorName

SDK	Allowed characters
Python (_chat_completion_client.py)	[a-zA-Z0-9_] (regex [^a-zA-Z0-9_]+)
.NET (SanitizeAuthorName reference)	[a-zA-Z0-9_] (per Python module comment)
Go (this PR)	[a-zA-Z0-9] (unicode.IsLetter || unicode.IsDigit) — underscores stripped

An agent named Agent_One reaches the API as AgentOne in Go but Agent_One in Python/.NET — a silent cross-SDK divergence in multi-agent group-chat participant attribution.

See the inline comment on sanitizeAuthorName for the one-line fix and a suggested test case.

Generated by Go API Consistency Review Agent · sonnet46 · 51.9 AIC · ⌖ 5.88 AIC · ⊞ 6K · ◷

[github-actions]

Generated by Go API Consistency Review Agent · sonnet46 · 51.9 AIC · ⌖ 5.88 AIC · ⊞ 6K

[github-actions]

Parity issue: sanitizeAuthorName strips underscores, but upstream Python and .NET preserve them.

The Go implementation filters with unicode.IsLetter || unicode.IsDigit, which silently drops _.

Both upstream implementations explicitly keep underscores in allowed characters:

• Python (_chat_completion_client.py): _INVALID_AUTHOR_NAME_RE = re.compile(r"[^a-zA-Z0-9_]+") — keeps [a-zA-Z0-9_].
• .NET: The PR description itself cites .NET OpenAIChatClient.SanitizeAuthorName; the Python module's comment describes its character set as [a-zA-Z0-9_], matching the .NET original.

A participant named Agent_One would be sent as AgentOne by Go but as Agent_One by Python/.NET — a silent cross-SDK divergence for multi-agent group-chat flows that use underscore-separated agent names.

Suggested fix:

if unicode.IsLetter(r) || unicode.IsDigit(r) || r == '_' {

The existing test TestChatAuthorNamePropagation_NonStreaming (input "Agent One" → expected "AgentOne") does not cover this case because a space is stripped by all three implementations. Please add a test with AuthorName: "Agent_One" expecting "name": "Agent_One" in the outgoing request.