Cannot create new local conversation (WSL)

What version of the IDE extension are you using?

v0.5.74 (Pre-Release)

What subscription do you have?

Plus

Which IDE are you using?

VS Code 1.109.3

What platform is your computer?

Microsoft Windows NT 10.0.26200.0 x64

What issue are you seeing?

Codex VS Code Extension: "Start New Agent" fails in WSL (both 0.5.74 and 0.4.74)

Summary

In VS Code Remote (WSL), starting a new Codex agent consistently fails.
I tested both OpenAI extension versions 0.5.74 (pre-release) and 0.4.74 (stable).
Both versions show the same runtime and thread-state failures.

This seemed to coincide with fact that I had two VS Code projects open - one WSL and one Windows. This may or may not be related.

Date / Time

Initially observed on **February 12, 2026

Environment

• VS Code Server (Remote WSL)
• Extension under test: openai.chatgpt
• Tested versions:
• 0.5.74 (openai.chatgpt-0.5.74-linux-x64)
• 0.4.74 (openai.chatgpt-0.4.74-linux-x64)
• Workspace: /home/micah/projects/HA_Visual_Dashboard_Maker

Actual

• Extension host throws:
• PendingMigrationError: navigator is now a global in nodejs
• Codex logs show:
• state db missing rollout path for thread ...
• no rollout found for conversation id ...
• no rollout found for thread id ...
• thread/resume request failures with code=-32600

Key evidence

0.5.74

• ~/.vscode-server/data/logs/20260213T222845/exthost2/remoteexthost.log
• ~/.vscode-server/data/logs/20260213T222845/exthost2/openai.chatgpt/Codex.log

0.4.74 A/B retest

• ~/.vscode-server/data/logs/20260213T222845/exthost3/remoteexthost.log
• ~/.vscode-server/data/logs/20260213T222845/exthost3/openai.chatgpt/Codex.log

A/B result

Downgrading from 0.5.74 to 0.4.74 did not change behavior.
Both versions fail with the same signatures, suggesting an upstream extension/runtime issue rather than a single-version regression.

Extra context

A fresh rollout file exists on disk (example):

• ~/.codex/sessions/2026/02/13/rollout-2026-02-13T22-29-07-019c56c3-06fc-7db3-a928-9c3607e17129.jsonl

But extension logs still report missing rollout path / missing rollout for thread IDs.

Remediation Steps Already Attempted (February 13, 2026)

I attempted the following remediation steps before filing this report:

1. Initial log review

I reviewed the VS Code Remote (WSL) logs under ~/.vscode-server/data/logs and repeatedly observed:

• PendingMigrationError: navigator is now a global in nodejs
• state db missing rollout path for thread ...
• no rollout found for conversation/thread id ...
• thread/resume failures with code=-32600

2. Full local reset

I ran a full reset to clear potentially corrupted local state while preserving auth/config:

• Moved older extension bits out of the active extensions path.
• Reset workspace storage for workspace ID b6cc891f0c8b9078d9d5396cd8460ca5.
• Reset local Codex state:
• ~/.codex/sessions
• ~/.codex/shell_snapshots

Result: the stale-session behavior changed, but the core failures persisted.

3. Extension scan cleanup

I moved backup folders out of ~/.vscode-server/extensions so VS Code would stop trying to scan non-extension directories as extensions.

Result: extension scan noise was reduced, but the core Codex failure persisted.

4. Reinstall / version-state verification

I verified that pre-release openai.chatgpt-0.5.74 was installed and active in WSL extension host state.

Result: issue remained reproducible on 0.5.74.

5. Controlled A/B version test

I performed a controlled downgrade from 0.5.74 to 0.4.74, reloaded VS Code, retried Start New Agent, and captured fresh logs.
I then compared 0.4.74 and 0.5.74 logs side-by-side.

Result: no behavioral improvement. Both versions showed the same failure signatures.

6. Restore to pre-release

After the A/B test, I switched back to openai.chatgpt-0.5.74 to return to the preferred test state.

7. Evidence collection

I prepared a support bundle containing the key logs and A/B comparison:

• Bundle directory: .codex_support_20260213_2236
• Archive: /home/micah/projects/HA_Visual_Dashboard_Maker/codex_vscode_bug_20260213.tar.gz

8. Existing issue check

I checked https://github.com/openai/codex/issues for matching open reports and did not find one that clearly matched this exact error pattern.

---

Current assessment

Despite reset, cleanup, reinstall verification, and A/B version testing, the failure remained reproducible.
Based on the repeated signatures across both extension versions, this appears more likely to be an upstream extension/runtime bug than a local one-time state issue.

Attachments prepared

• .codex_support_20260213_2236/AB_COMPARISON.md
• .codex_support_20260213_2236/remoteexthost_20260213T222845_exthost2.log
• .codex_support_20260213_2236/Codex_20260213T222845_exthost2.log
• .codex_support_20260213_2236/remoteexthost_20260213T222845_exthost3_0.4.74.log
• .codex_support_20260213_2236/Codex_20260213T222845_exthost3_0.4.74.log
• .codex_support_20260213_2236/rollout-2026-02-13T22-29-07-019c56c3-06fc-7db3-a928-9c3607e17129.jsonl

What steps can reproduce the bug?

Reproduction steps

1. Open VS Code in WSL remote session.
2. Activate OpenAI Codex extension.
3. Click Start New Agent.
4. Observe failure/errors in extension host logs.
5. Repeat after full local state reset and after version downgrade/upgrade test.

What is the expected behavior?

Expected

• New Codex agent/session starts successfully.
• Conversation thread is created/resumed without rollout lookup errors.

Additional information

codex_vscode_bug_20260213.tar.gz