Codex CLI requests consistently cut off at ~150s (stream disconnected); please increase or make timeout configurable #3478
Description
What version of Codex is running?
codex-cli 0.33.0
Which model were you using?
gpt-5
What platform is your computer?
Linux 6.14.0-1011-aws x86_64 x86_64
What steps can reproduce the bug?
Summary:
Codex CLI requests are consistently cut off at around 150 seconds, which makes it difficult to complete complex coding tasks that require longer reasoning and multi-file edits. The stream terminates with the error shown below. The VS Code extension seems to allow a longer timeout and often completes the same tasks, but the exact duration is unclear.
Steps to Reproduce
- Open Codex CLI and start a session with streaming enabled.
- Ask Codex to implement a complex, multi-file refactor or generate substantial code with tests (e.g., multi-step agent flow that applies patches and explains changes).
- Let the model think and stream for a while without interrupting.
After approximately 150 seconds, the stream stops and the CLI reports an error.
Error message
stream error: stream disconnected before completion: Transport error: error decoding response body
What is the expected behavior?
- Long-running requests should be allowed to finish streaming if the model is still producing output.
- The timeout should be longer by default (e.g., 5–10 minutes), or at least configurable via a flag or environment variable.
- The timeout value should be consistent (or clearly documented) across the CLI and the VS Code extension.
What do you see instead?
- Requests consistently terminate near 150s with a transport/stream decoding error.
- Complex tasks (long reasoning, multi-file patches) fail to complete in the CLI.
Additional information
- Network is stable; this occurs across multiple sessions and prompts.
Impact
- High: Complex refactors, larger code generation, and longer reasoning sessions regularly fail in CLI usage.
- This forces users to split work into many smaller prompts or switch tools.