Question 1

How to fix Bigger Context failures with saved chats in Codex Web GPT v6.1.11 (strict mode violation, message verification errors)?

Accepted Answer

Upgrade to Codex Web GPT v6.1.12 or later. Version 6.1.11 had four bugs on saved-chat pages with Bigger Context enabled: (1) `attachFiles` used a page-wide file-input locator that matched both the hidden 'home' composer form and the visible 'thread' form, causing Playwright 'strict mode violation: resolved to 2 elements' errors when attaching images; (2) `readChatGptUserMessage` treated auto-linked URLs (`<a>` elements) as unsupported, so message verification in `reconcileMultipartHistory` / `userMessageMatchesSubmission` always failed with 'ChatGPT changed an acknowledged Bigger Context exchange'; (3) expanding the ~230k-character stage message for verification unmounted the rest of the virtualized conversation, detaching the bound assistant response ('ChatGPT opened another user turn while the bound assistant response was detached'); (4) send confirmation timed out on long multi-part transactions because old turns unmounted as new ones mounted. v6.1.12 fixes all of these by scoping image uploads to the active composer's own form, preserving completed upload identities across virtualization, restoring collapsed messages after verification, and accepting auto-linked URLs while checking exact text. Note: the duplicate `data-turn-key` error ('ChatGPT exposed duplicate conversation turn identities') on project pages was traced to a local patch, not the stock build; if you see it on the unmodified release, file a new issue with browser checkpoint JSON.

Question 2

How to fix "Invalid conversation body" error when approving Codex Native2 tool requests in ChatGPT saved chats?

Accepted Answer

This happens when the tool approval card is clicked while the assistant message that requested the tool is still streaming — ChatGPT rejects the approval, no MCP call reaches Codex, and the turn fails with `chatgpt_response_incomplete`.

**Fix:** Upgrade Codex Web GPT to **v6.1.12 or later**, which waits for the Stop button to disappear (streaming finished) before automatically approving the tool request.

**Workaround for v6.1.10/6.1.11:** Turn off **Automatically approve one-time tool requests**, then manually click **Always allow** only after the Stop button disappears from the composer (roughly 2.4+ seconds after the card appears), within the bridge's 60-second wait window. Clicking earlier — whether Allow once or Always allow — triggers the error.

Question 3

How to fix "ChatGPT personalization preflight exceeded its readiness deadline" error in Codex Web GPT?

Accepted Answer

This error occurs because ChatGPT removed the Personalized/Unpersonalized toggle and plugin access from Temporary Chats (around 2026-10-18/19), so the preflight in v6.1.9/6.1.10 waits for a personalization control that no longer exists and times out after ~30s. Every turn that opens a fresh Temporary Chat fails with "stream disconnected before completion". Verified workaround: set `"useSavedChats": true` in `config.json` so the bridge opens normal (saved) chats instead of Temporary Chats. Saved chats skip the personalization preflight entirely (`saved-chat-navigation-complete` → `connector-mention-triggered` → `connector-selected`), and the @codex connector works normally. Confirmed working on stock 6.1.10: tool-calling turns and multi-part Bigger Context all pass. A permanent fix would require the launcher to detect the missing personalization control/connector in Temporary Chats and automatically fall back to saved chats.

Question 4

How to fix "Compaction continuation environment conflicts with its current Codex rollout (fields: workspace_roots, writable_roots)" error?

Accepted Answer

This error occurs in codex-chatgpt-web version 6.1.9 (and nearby versions) when Codex Desktop is restarted: Codex stops writing `workspace_roots` to the rollout `turn_context`, but the compaction continuation still claims the original multiple workspace roots. The launcher's trusted-environment check then rejects it with a roots mismatch, killing every compaction turn (ordinary turns still work).

**Fix:** Upgrade codex-chatgpt-web to **v6.1.11** or later, which contains the patch: when the current Codex record explicitly grants full access (`danger-full-access` with disabled permission profile) but omits workspace roots, the launcher now keeps the workspace context from the current request instead of falling back to `[cwd]`. Explicit roots, permissions, working directory, and turn identity are still validated.

**Steps:**
1. Update to release v6.1.11: https://github.com/miuuyy/codex-chatgpt-web/releases/tag/v6.1.11
2. Restart the launcher/bridge and Codex Desktop.
3. Retry the compaction continuation in the affected thread.

If it still fails after upgrading, open a new issue with Activity → Export safe log, the failure timestamp, and your launcher/Codex versions.

Question 5

How to fix 'ChatGPT personalization preflight exceeded its readiness deadline' error and Codex connector not found in Temporary Chat?

Accepted Answer

Root cause: ChatGPT removed the Personalized/Unpersonalized toggle and plugin/connector access from Temporary Chats, so `@codex` no longer lists connectors like 'Codex Native2 PC2' there, causing verification to time out with `code=connector_not_found`. Workaround (confirmed in #436 and #859, works on Windows and macOS): 1) In the launcher, enable **Settings → Save chats in ChatGPT** (`useSavedChats: true`). 2) Finish active tasks, then fully quit and reopen both the launcher and Codex. 3) Start a new task — the connector appears instantly in normal saved-chat mode and tools work again. Caveat for v6.1.10: **MCP → Verify runtime** ignores this setting and always checks Temporary Chat, so verification may still fail — test with an actual task instead. A fix making verification respect the selected mode (with saved chats default-on for new setups) is pending release. Note: enabling saved chats means task conversations will remain in your ChatGPT history.

codex-chatgpt-web

Core Features

What It Can't Do

Use Cases

Tags

Getting Started

Similar Projects