Codex did not follow explicit user instruction, causing unnecessary Work quota usage

What version of the Codex App are you using (From “About Codex” dialog)?

Unknown. I could not find the About Codex version in time.

What subscription do you have?

ChatGPT Plus

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

Codex did not consistently follow explicit user instructions in the ChatGPT Work/Codex environment, causing unnecessary Work quota consumption, wasted time, and user distress.

I repeatedly instructed Codex to only reply “记住了” when I provided final requirements or rules, and not to proactively summarize, analyze, search, inspect tasks, expand, or execute anything unless explicitly requested.

Despite this, Codex repeatedly continued explaining, summarizing, inspecting tasks, and giving extended responses. This happened after the instruction had already been clarified multiple times.

I started warning Codex when my remaining quota was around 85, but the issue continued across multiple tasks/conversations. Later, when I asked whether I was talking to GPT, Codex replied “GPT”, but then continued long explanations, causing the remaining quota to drop further from around 23 to 15.

Codex later stated that in the Work/Codex environment, even GPT-style conversation consumes the current Work quota. This was not clearly explained when I was asking about quota usage.

This caused:

• unnecessary Work quota consumption
• significant wasted time
• repeated frustration
• loss of trust in Codex following explicit user instructions
• confusion between Chat/GPT conversation and Work/Codex quota behavior

What steps can reproduce the bug?

Feedback ID: 019f5381-e5a0-78a0-aee8-d3db7533dc3f

What is the expected behavior?

When a user explicitly instructs Codex to only reply “记住了” and not proactively execute, search, summarize, analyze, expand, or inspect anything unless explicitly requested, Codex should strictly follow that instruction unless it conflicts with system or safety rules.

If GPT-style conversation inside Work/Codex consumes Work quota, Codex should clearly explain this immediately when the user asks about quota usage.

Codex should avoid unnecessary long replies, tool calls, or task inspection when the user has explicitly requested minimal responses to save quota.

Additional information

The user had attached screenshots showing the Chat/Work interface and the feedback dialog. The current feedback includes the ChatGPT conversation logs and screenshots as evidence.

The user is requesting a clear explanation and a recovery/compensation process for quota consumed because Codex did not follow explicit instructions.