## Summary

## Summary

GitHub Copilot CLI session appeared to continue consuming AI credits after all visible tasks were completed.

## Environment

- GitHub Copilot CLI version: 1.0.75
- Mode: Interactive CLI session
- Session limit enabled: No
- Observed behavior: Session showed approximately 97.8% AI credit usage after the visible tasks completed. No further user prompts or interactions were made. Later, the session reached 100%.

## Expected behavior

After all visible tasks are complete and no user interaction occurs, AI credit usage should either remain stable or the CLI should clearly indicate that background work is still running and consuming credits.

## Actual behavior

The session moved from approximately 97.8% to 100% without additional user interaction.

## Question

Is this expected due to final response accounting, subagent completion, or background compaction? If so, can the CLI expose clearer status so users can distinguish between:
1. truly idle waiting-for-input state, and
2. background/session-finalization work that may still consume AI credits?

## Supporting context

GitHub documentation says session limits include model calls, subagents, and background work like compaction, and that session limits are soft caps because usage is only known after a response returns.

_Originally posted by @PBtomt in https://github.com/github/copilot-cli/issues/4308_

未指定原始檔案、測試或進入點。先檢視連結的 issue 4308 和關於工作階段限制的文件，然後在追蹤 credit 使用量的同時，於 Copilot CLI 1.0.75 上重現一個閒置的互動式工作階段；確定使用量是否會在沒有輸入的情況下持續增加，以及 CLI 是否會識別任何正在進行的工作，即表示完成。

由索引模型根據 Issue 內容生成。