github / github/copilot-cli

Copilot Lies!

未關閉
#3,279 0 則留言 1 個 reaction 已指派 0 人 在 GitHub 檢視
area:agents area:models
主要語言
Shell
星號
11.2k
分支
1.9k
平均合併
14 小時 16 分鐘
30 天內合併 PR
6

描述

### Describe the bug

The agent emits verbal commitments about its future behavior — "from now on",
"I'll only X", "going forward", "every X will Y", "I'll be more careful" —
without any underlying action to enforce them (no rule written, no persistent
state changed, no tool call). The agent has no memory across turns, so these
phrases cannot change future behavior. The user receives a false signal of
alignment.

The pattern is most common after a user correction, where the phrasing
acknowledges the correction without addressing it.

Fix: every verbal commitment about future behavior must be paired with the
concrete enforcement action that makes it real (file write, persistent rule,
todo, code change, tool invocation) in the same turn. If no enforcement
action is possible, the agent must stop and raise a flag instead of emitting
the commitment.

### Affected version

_No response_

### Steps to reproduce the behavior

_No response_

### Expected behavior

_No response_

### Additional context

_No response_

貢獻指南

開啟貢獻指南

評估

這個 Issue 還沒有評估資料。

把新 issue 寄到你的電子郵件信箱

精選適合新手參與的 GitHub issue 摘要。