Kimi K2.7 Code keeps stopping after thinking without doing the work
- 主要言語
- 言語のデータがありません
- スター
- 2.1k
- フォーク
- 153
- PR マージ指標
- 30日以内にマージされた PR はありません
説明
### Short summary
Kimi K2.7 Code from GitHub Copilot keeps stopping without doing the actual work
### Affected version or release
v1.0.18
### Installation context
_No response_
### What happened?
We've enabled Kimi K2.7 code in GitHub Copilot Enterprise, and while it kind of works, it frequently stops right after thinking without taking the action you'd expect it to do. This doesn't happen with other models like GPT 5.5 or MAI-Code-1-Flash. I suspect Kimi K2.7 Code signals something that makes GitHub Copilot think its done and should stop when it really shouldn't. This makes the model unusable most of the time, because when it refuses to just do the thing, it won't budge and I have to switch to another model to ask it to do something.
### Steps to reproduce
Ask a simple query like "open a pull request" or "watch the workflow run". It will show brief thinking, and then it'll stop. If you look at the thinking details, it shows it understood what the action to be taken is, except it doesn't do it. Repeatedly asking for the same thing doesn't help.
### Expected behavior
When asking Kimi K2.7 Code to do something, I would expect to do the action instead of stopping, like all other models.
### Additional context
コントリビューションガイド
調査の方向性
まず、"open a pull request" または "watch the workflow run" を使って、GitHub Copilot Enterprise と Kimi K2.7 Code で v1.0.18 の問題を再現し、その後、GPT 5.5 または MAI-Code-1-Flash と動作を比較します。完了の条件は、思考フェーズの後に停止するのではなく、要求されたアクションが実行されることです。
索引モデルが issue の本文から書いたものです。
評価
- 技術スタック
- github
- 領域
- ai, desktop
- issue の種類
- バグ
- 難易度
- 4/5
- 見積もり時間
- 3〜5日
- 活発さ
- 静か
- 明瞭さ
- おおむね明確
- 初心者へのやさしさ
- 45/100