VS Code Codex extension chat double-renders output and displays timer drift

What version of the IDE extension are you using?

26.609.30741

What subscription do you have?

Free

Which IDE are you using?

VS Code

What platform is your computer?

Darwin 25.5.0 arm64 arm

What issue are you seeing?

After a session completes and the final report renders in the panel, an identical copy of the report appears below it. The duplicate content is byte-identical to the original, not a revision or correction. The elapsed timer also shows a different value after apparent completion than what was visible at first render, consistently running 4 to 10 seconds longer in the completion event than the live counter showed.

Confirmed across 8 sessions using the VS Code extension v.26.609.30741 two models (GPT-5.4-Mini, GPT-5.5), four reasoning levels (Low, Medium, High, Extra High), and session runtimes from 25 to 233 seconds.

What steps can reproduce the bug?

No specific steps required. The duplicate render appeared in every session tested, including the simplest case: a single-turn session, zero tool calls, 18,002 total tokens, 38 seconds runtime. Complexity does not appear to be a factor.

The ~/.codex/sessions rollout .jsonl files confirm single emission in every case: one final answer event_msg, one matching response_item, one task_complete carrying last_agent_message, zero records after task_complete. The duplication may be downstream of the log writer, in the panel render path.

What is the expected behavior?

The final answer should render once. The elapsed timer displayed in the panel should match the duration_ms value in the task_complete event, or the discrepancy between the two should be documented.

Additional information

I also observed timer drift on the desktop app, but not any output duplication recently. On the desktop app , chat output appeared to be corrected (formatting cleared or duplications deleted), or reasoning was removed, making the chat unreliable as a source for documenting agent activity, forcing users to dig into logs to verify behavior, which complicates how users can interact with the chat. Duplication issues, but on different surfaces and/or include more complicated workflows:

• Issue #15318
• Issue #15912
• Issue #14805