feat(release): prepare v4.0.4 model discovery and runtime compatibility - #50
Conversation
- Read the daily CLI catalog without sending prompts or resuming sessions. - Cache by account, directory and CLI build, with bounded probe cleanup. - Use native models in Web, BTW and TUI pickers and refresh on open. - Preserve native Opus aliases and historical explicit model selections.
- Refresh SDK dependency locks and exact runtime validation together. - Keep the daily CLI and existing native session lifecycle unchanged. - Update version guidance and cover both runtime dependency pins.
- Align backend and Web versions and refresh installation examples. - Document native Claude models, the SDK upgrade and inline code colors. - Explain the SDK migration requirement and desktop release scope.
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: d3a1ad50f2
ℹ️ About Codex in GitHub
Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
- Open a pull request for review
- Mark a draft as ready
- Comment "@codex review".
If Codex has suggestions, it will comment; otherwise it will react with 👍.
When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".
- Match Work discovery to its runtime policy and keep Code defaults isolated. - Inherit native Claude model and effort defaults for new sessions. - Refresh explicit Codex model reads while coalescing requests per account. - Preserve account-local catalogs on failure and cover reload and cancellation paths.
- Accept only the official account-derived alias and owner-private listener. - Fence app-server replacement by the physical socket identity. - Reuse the same validation in runtime readiness and installation receipts.
- Expand Claude reasoning by default while preserving manual disclosure state. - Focus the composer on an unfocused slash key without inserting it. - Respect editors, IME composition and dialogs and lazy-load the question sheet.
- Document native defaults, Codex catalog refresh and protected socket support. - Describe Claude thinking expansion and the composer focus shortcut in both languages.
|
Codex Review: Didn't find any major issues. Hooray! Reviewed commit: ℹ️ About Codex in GitHubCodex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you
If Codex has suggestions, it will comment; otherwise it will react with 👍. When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback". |
Model pickers can hide newly available models after a CLI upgrade or browser reload. v4.0.4 reads Claude models and supported effort levels from the daily CLI and refreshes explicit Codex catalog requests instead of retaining a stale ten-minute list. Discovery creates no model turn. Work uses its own runtime policy, account caches remain isolated, and new Claude sessions inherit native defaults.
The release also validates Codex 0.156 protected socket aliases for shared app-server readiness, upgrades the verified Claude Agent SDK to 0.2.157, renders inline code in softer harness colors, expands Claude thinking text by default, and adds an unfocused
/shortcut that focuses the composer without typing. Existing sessions keep their explicit settings; protocol remains v72.Upgrade limitation: official v4.0.3 Wrappers need the documented one-time installer upgrade because the SDK/service contract changes. Drain Claude work and migrate any independent Claude service before Wrapper activation. Native Claude/Codex CLIs are updated separately. Electron, DSH and MCP computer use remain outside this release.
Validation:
Merge and tag v4.0.4 only after CI and the latest-head AI review complete. The release workflow then builds, smoke-tests and publishes six platform bundles.