The Codex Changelog

This episode digs into a silent OAuth refresh failure that leaves MCP tools disappearing without warning, plus the debugging steps needed to force re-authentication when expired tokens get stuck in cache. It also covers a nasty workspace-root regression that can balloon environment paths across turns, along with a Responses Lite bug that breaks tools through custom gateways and Azure APIM.

We dig into a baffling installer failure caused by unauthenticated GitHub API calls hitting rate limits and silently masking HTTP 403 errors as missing release assets. Then we cover the fix: consolidated metadata lookups, token support for CI pipelines, and a few helpful permission and Windows updater improvements.

Discover why a freshly updated CLI can still reject new subagent models when an old background daemon is holding onto stale code. The episode walks through diagnosing version mismatches, checking running processes, and using a socket-bypassing flag or process kill to restore proper model support.

Show Notes

• Stale app-server survives CLI update and causes 0.147.0 ...: https://github.com/openai/codex/issues/37536

The hosts break down ChatGPT’s native Linux desktop preview, including supported Ubuntu, Debian, and Fedora builds, local Codex execution, and one-command imports from Claude Code, Claude Cowork, and Cursor. They also dig into Wayland input bugs, CLI-to-desktop project mapping quirks, and the migration pressure caused by model sunsets.

Show Notes

• ChatGPT & Codex changelog: https://learn.chatgpt.com/docs/changelog
• Codex in ChatGPT desktop app for Linux is now in preview 🐧: https://community.openai.com/t/codex-in-chatgpt-desktop-app-for-linux-is-now-in-preview/1390027

This episode breaks down how the latest macOS ChatGPT app update turns desktop activity into local Markdown memories using accessibility event streams and an ephemeral Codex session. It also covers opt-in privacy controls, plain-text storage risks, prompt injection concerns, and the looming deadline to migrate authenticated sessions off older GPT models.

This episode breaks down Codex 0.147.0’s new automated approval flow, showing how approve for me replaces full auto with policy-driven sandbox reviews that speed up routine edits while still blocking risky actions. It also covers platform caveats like Windows ACL latency, plus release polish such as improved macOS notarization, terminal rendering fixes, and stricter trust checks.

Show Notes

• Releases · openai/codex: https://github.com/openai/codex/releases

This episode covers major Codex 0.147.0 upgrades, from persistent sidebar sections and paginated transcript loading to stricter SQLite writer locks and seamless legacy session migration. It also explains new automatic secret redaction that masks bearer tokens and API keys in commands and replay logs.

Show Notes

• Releases · openai/codex: https://github.com/openai/codex/releases

We break down how Codex CLI 0.147.0 finally stops duplicate session history by syncing Claude and Cursor imports in place, while preserving titles, bindings, and working-directory context. The episode also covers Cursor rule imports into native host skills, plus security and stability improvements like secret redaction and Ghostty TUI fixes.

Show Notes

• ChatGPT & Codex changelog: https://learn.chatgpt.com/docs/changelog

We break down how Codex CLI 0.147.0 brings native Amazon Bedrock search caching and server-side conversation compaction to cut latency, reduce token waste, and keep long-running agent sessions snappy. The episode also covers setup tips, IAM and cross-region caveats, plus quality-of-life fixes like bearer token redaction, frozen terminal input resolution, and stronger project trust prompts.

This episode covers Codex 0.147.0’s move to the new MCP 2026 07 28 spec, including non-blocking server startup, cached tool definitions, paginated schema discovery, and multi-round tool elicitation. It also touches on legacy compatibility pitfalls and a few welcome terminal UI polish fixes.