fix(audio): keep the selected command on voice/STT-dictated messages

[RoyBA]

Problem

When a command is selected in the composer (e.g. /search) and the user dictates their message with the voice/STT feature instead of typing it, the message is sent without the selected command. Typing the exact same message correctly includes the command.

Root cause

The selected command lives in frontend state (persistentCommandState) and is attached to the message only in the text-composer submit path (the client_message event). The audio path (audio_start → audio_chunk →
audio_end) never transmits the command, and the transcribed user message is created on the backend (in the app's on_audio_end handler), so it has no knowledge of the UI selection.

Fix

Carry the selected command through the audio turn and auto-attach it to the user message produced during that turn.

Frontend — send the selected command with audio_start:

• VoiceButton reads persistentCommandState and passes selectedCommand?.id to startConversation.
• useAudio.startConversation(command?) → useChatInteract.startAudioStream(command?) → socket.emit('audio_start', { command }).

Backend — store it for the turn and attach it:

• BaseSession gains a current_command field.
• audio_start stores session.current_command; audio_end clears it, so it only applies to that audio turn.
• Message.__post_init__ auto-attaches current_command to user_messages that don't already carry a command.

Backward compatibility

Fully backward compatible. The audio_start payload is optional (old clients that emit no payload keep working), the new frontend arguments are optional, and the existing Message.command field is reused. Typed messages and assistant messages are unchanged.

Testing

• Added unit tests in backend/tests/test_message.py: auto-attach on user_message, explicit command takes precedence, no-op when no command is set, and assistant messages never inherit the command.
• Full backend suite, mypy, ruff, pnpm type-check, ESLint, and Prettier all pass.
• Manual check: select a command, dictate a message via the mic → the resulting user message carries the command, identical to typing it.

Out of scope

modes has the same gap (also only attached in the text submit path). Left for a follow-up to keep this PR focused on commands.

---

Summary by cubic

Voice/STT-dictated messages now keep the selected command like typed messages; previously they dropped it.

• Frontend sends the selected command id with audio_start; the payload is optional, so old clients keep working.
• Backend stores the command on the session only when the audio start is accepted, and clears it on audio_end.
• Message.__post_init__ auto-attaches the session command to command-less user_messages; from_dict resets from the payload so typed or resumed messages never inherit an active audio turn's command.

^{Written for commit cb0f7fe. Summary will update on new commits.}

[cubic-dev-ai]

All reported issues were addressed across 7 files

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[Copilot]

Pull request overview

Carries selected commands through voice/STT turns so dictated messages match typed-message behavior.

Changes:

• Sends the selected command with audio_start.
• Stores and attaches the command to generated user messages.
• Adds command inheritance and deserialization tests.

Reviewed changes

Copilot reviewed 7 out of 7 changed files in this pull request and generated 1 comment.

Show a summary per file

File	Description
libs/react-client/src/useChatInteract.ts	Emits command with audio start.
libs/react-client/src/useAudio.ts	Forwards command when starting audio.
frontend/src/components/chat/MessageComposer/VoiceButton.tsx	Reads and passes the selected command.
backend/chainlit/socket.py	Manages command state across audio turns.
backend/chainlit/session.py	Adds session command state.
backend/chainlit/message.py	Attaches commands to generated user messages.
backend/tests/test_message.py	Tests command attachment behavior.

Suppressed comments (1)

backend/chainlit/socket.py:492

• This unconditional clear races with a subsequent audio turn. endConversation marks the client off before emitting audio_end, while this handler awaits the app's on_audio_end; a quick new audio_start can therefore overwrite current_command, after which the older handler both exposes the new command to the old callback and clears it here. Associate the command with a turn/context (or serialize audio turns) so overlapping handlers cannot misattribute or erase commands.

        session.current_command = None

---

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

[dokterbob]

@codex review

[cubic-dev-ai]

3 issues found across 8 files

Prompt for AI agents (unresolved issues)

Check if these issues are valid — if so, understand the root cause of each and fix them. If appropriate, use sub-agents to investigate and fix each issue separately.


<file name="backend/chainlit/message.py">

<violation number="1" location="backend/chainlit/message.py:88">
P2: When a client sends a non-string `command` in an audio or message payload, this assignment bypasses `Message.__init__`'s string normalization and stores arbitrary JSON in `Message.command`. Validate or normalize the payload command before restoring it so persistence and callbacks receive the declared string type.</violation>
</file>

<file name="backend/tests/test_socket.py">

<violation number="1" location="backend/tests/test_socket.py:637">
P3: The PR calls out "audio_start payload is optional" as a backward-compat requirement, but every new test passes `payload={"command": "search"}`. The `payload.get("command") if payload else None` branch in `socket.py` is never exercised, so an older frontend sending `audio_start` with no payload (or a payload without a `command` key) after an accepted start would silently lose the command without any test catching it. Add a test asserting `session.current_command is None` when `accepted=True` with `payload=None` and with `payload={}`.</violation>
</file>

<file name="backend/chainlit/socket.py">

<violation number="1" location="backend/chainlit/socket.py:440">
P2: Starting a new dictation before the previous `on_audio_end` finishes clears the new turn's command here. Clear only if the ending handler still owns the active audio turn, or serialize audio turns.</violation>
</file>

<sub>Reply with feedback, questions, or to request a fix.

Re-trigger cubic</sub>

[cubic-dev-ai]

P2: When a client sends a non-string command in an audio or message payload, this assignment bypasses Message.__init__'s string normalization and stores arbitrary JSON in Message.command. Validate or normalize the payload command before restoring it so persistence and callbacks receive the declared string type.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At backend/chainlit/message.py, line 88:

<comment>When a client sends a non-string `command` in an audio or message payload, this assignment bypasses `Message.__init__`'s string normalization and stores arbitrary JSON in `Message.command`. Validate or normalize the payload command before restoring it so persistence and callbacks receive the declared string type.</comment>

<file context>
@@ -73,6 +81,12 @@ def from_dict(self, _dict: StepDict):
+        # the payload so messages rebuilt here (incoming client messages, thread
+        # resume) never inherit an active audio turn's command that
+        # __post_init__ applies to command-less user messages.
+        message.command = _dict.get("command")
+        return message
 
</file context>

Suggested change

	message.command = _dict.get("command")
	payload_command = _dict.get("command")
	message.command = str(payload_command) if payload_command else None

[cubic-dev-ai]

P2: Starting a new dictation before the previous on_audio_end finishes clears the new turn's command here. Clear only if the ending handler still owns the active audio turn, or serialize audio turns.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At backend/chainlit/socket.py, line 440:

<comment>Starting a new dictation before the previous `on_audio_end` finishes clears the new turn's command here. Clear only if the ending handler still owns the active audio turn, or serialize audio turns.</comment>

<file context>
@@ -428,16 +428,22 @@ async def window_message(sid, data):
 
+    # Only keep the UI-selected command when audio is accepted (consumed in
+    # Message.__post_init__), so a declined/disabled start leaves nothing stale.
+    session.current_command = None
+
     if config.features.audio and config.features.audio.enabled:
</file context>

[cubic-dev-ai]

P3: The PR calls out "audio_start payload is optional" as a backward-compat requirement, but every new test passes payload={"command": "search"}. The payload.get("command") if payload else None branch in socket.py is never exercised, so an older frontend sending audio_start with no payload (or a payload without a command key) after an accepted start would silently lose the command without any test catching it. Add a test asserting session.current_command is None when accepted=True with payload=None and with payload={}.

Prompt for AI agents

Check if this issue is valid — if so, understand the root cause and fix it. At backend/tests/test_socket.py, line 637:

<comment>The PR calls out "audio_start payload is optional" as a backward-compat requirement, but every new test passes `payload={"command": "search"}`. The `payload.get("command") if payload else None` branch in `socket.py` is never exercised, so an older frontend sending `audio_start` with no payload (or a payload without a `command` key) after an accepted start would silently lose the command without any test catching it. Add a test asserting `session.current_command is None` when `accepted=True` with `payload=None` and with `payload={}`.</comment>

<file context>
@@ -628,3 +629,43 @@ async def test_on_chat_start_not_duplicated_on_fresh_then_reconnect(
+class TestAudioStartCommand:
+    """audio_start only keeps the UI-selected command when audio is accepted."""
+
+    async def _run(self, *, accepted, enabled=True, payload=None):
+        session = Mock()
+        session.current_command = "stale"
</file context>