Code Review - overall a clean and focused PR.

Code Review

Overall this is a clean, focused PR with a small diff and a clear rationale. Here is my analysis.

Main fix — removal of redundant clearChunks()

Correct. In src/utils/audioUtils.ts, stopPlayback() (~line 690) already does this.sessionId++ and resets chunkCounter, pendingChunks, and allChunksReceived to their initial values. clearChunks() (~line 683) does exactly the same thing. Calling clearChunks() right after stopPlayback() caused a double sessionId++. While this did not produce visible bugs (in-flight addChunk calls capture mySession synchronously before either increment, so they are already invalidated by the first stopPlayback() increment), it was semantically wrong and redundant. Removing it is the right call.

Comment updates

Both changes are accurate:

1. speech_stopped case: the old comment referenced a server gate in voice_bridge.py that no longer exists after the server-side refactor (PR 81 on MyTaskly-server). The new wording correctly attributes turn management to OpenAI VAD.
2. onChunk callback: same fix for the same reason — mic gating moved from voice_bridge.py to OpenAI server-side VAD. Accurate and avoids misleading future maintainers.

Minor suggestions (non-blocking)

• clearChunks() is still exposed on AudioPlayer. Now that stopPlayback() covers all the same resets, it is worth checking whether clearChunks() is called anywhere else in the codebase. If not, removing it in a follow-up would reduce the API surface and eliminate the class of confusion that led to this bug.

• The interrupted handler sets agentEndedRef.current = true, which is correct and intentional (prevents the audio_end callback from transitioning back to processing for the old turn). This is unrelated to this PR, but a short inline comment explaining the why would help future readers.

Summary: bug fix is correct, comment updates are accurate, scope is focused with no unrelated changes, no security concerns. Approved.