Codex CLI repeatedly interrupts long prompts with “Conversation interrupted”

Description

Codex CLI repeatedly aborts a conversation while submitting a long, structured implementation prompt. The CLI displays:

Conversation interrupted - tell the model what to do differently. Something went wrong? Hit `/feedback` to report the issue

This happened more than once in a fresh Codex session.

Uploaded thread

Thread ID from the generated feedback link:

019fd90d-0df2-7921-b151-52f1949d9f45

Environment

• Codex CLI on a remote Linux HPC login node
• Shell session over SSH
• Model command used:

codex -m gpt-5.6-terra -c 'model_reasoning_effort="high"'

Steps to reproduce

1. Start a fresh Codex CLI session.
2. Submit a long, highly structured prompt containing many sections, constraints, paths, and validation requirements.
3. Wait for Codex to process the prompt.
4. The conversation is interrupted before the task completes, and the CLI shows the error above.
5. Repeating in a new session can produce the same interruption.

Expected behavior

Codex should either process the full prompt, reject it with a clear size/context-limit message, or preserve the session and ask the user to shorten/split the request.

Actual behavior

The conversation terminates with a generic interruption message and no clear indication whether the cause is prompt length, transport failure, model failure, or session instability.

Impact

This makes long, governance-heavy implementation tasks unreliable and forces the user to restart the conversation and manually split prompts into multiple parts.

Suggested improvements

• Show the underlying error category when possible.
• Warn before submission if the prompt is too large.
• Preserve the session and allow retrying the same prompt.
• Provide a built-in way to submit a long prompt in multiple acknowledged parts without starting execution early.

I can provide additional CLI version, logs, or terminal output if needed.