Repository navigation
Conversation
Ship the Desktop Statsig wrap and Settings switch on by default as v0.13.2-1, warn when CODEXHOST_CPA_LIVE_VOICE=0 needs a relaunch, and retarget npm/GitHub identity to @karlorz/codexhost-cli.
feat: add default-on live-voice override for the karlorz fork
Keep GitHub latest discovery, Settings commands, and console NOTICE on this fork so hyphen versions are not treated as an upgrade from BytePioneer.
Fork tags like 0.13.2-2 must publish as latest so in-app Updates can discover them. Named prereleases stay on next/test.
Hide the native out-of-usage card on every Composer from Settings → Appearance (default on), including asides that omit status ARIA. Project this Composer's account and reserve usage gates so Send is not blocked by exhausted ChatGPT quota, including Codex threads and same-atom owner or descendant subscriptions.
This reverts commit ff7fe5b.
Keep karlorz live-voice, update lock, and numeric fork-counter identity. Drop the overlapping fork usage-banner/send-gate rewrite in favor of upstream Desktop 26.1002 gate projection and external-Harness banner hide.
Record shipped v0.13.2-3 and the next unused counter v0.13.3-0 after syncing BytePioneer 0.13.3.
Keep 0.13.3-cpa.1 on the named-prerelease path. First fork of upstream 0.13.3 stays 0.13.3-0; later counters on that version are 0.13.3-0-1.
Route native Desktop realtime start/stop for external Threads through Host, report externalSandbox, and acknowledge thread/settings/update without applying Harness permissions. Carry the Composer Harness selection into overlay voice launch, isolate leftover overlay notifications, and install a Grok-only Desktop session rule.
Stop installing ~/.grok/rules/codexhost-desktop.md. Pass the rules on Grok session/new and session/load, and retire leftover Host-owned copies.
When a previous npm start or crash leaves the Controller listening with no Desktop, acquire launcher ownership and continue a clean launch instead of blocking on the control endpoint.
The linux "not installed" case was matching /opt/homebrew/bin/claude on this machine. Force isExecutable false so the test does not depend on the host PATH.
Keep a single numeric counter after the upstream version. The next fork release after v0.13.3-0 is v0.13.3-1.
Emit Host generatedImage items from Grok ImageGen/ImageEdit ACP output and project them to Desktop imageGeneration with savedPath so the official gallery can render session files.
Keep exactOptionalPropertyTypes: do not write undefined onto the optional field.
Record Try again recovery and the Desktop 26.1007 image-completion notification contract.
Re-apply the karlorz Settings toggle dropped during the upstream v0.13.3 usage-gate sync. Hide the out-of-usage card on every Composer by default, and lift the Renderer usage gate so Send and Enter still work on new drafts, existing Threads, and Codex Threads.
Park Host/Renderer queue leftover work off main so Desktop 26.1021 audit can continue on a clean tree.
Normalize queued type:text items and input_text to include text_elements so Desktop 26.1021 can refresh the server queue. Keep empty-queue turn/start as -32602 after interrupt.
Add the 26.1021.31345 native Chat Instant Connect profile with the relocated composer START binding, Work capability snapshot recovery, provider-only Voice Settings explanation, and document-language locale fallback for English Grok permission menus. Detect the installed Desktop version for the live-voice hook, pin compat-daily to the Chat assets module, and document the live voice-hook debug path. Preserve native eligibility, credentials, Statsig, drafts, and live Desktop state. Microphone and audible conversation acceptance remain unverified.
…rofiles MSIX Identity Version is a deployment number, not the About/product version used to select audited voice assets. Keep native eligibility checks while binding the Windows Chat and Work profiles after official login.
Closing the attachment socket is not proof the Controller process exited. Snapshot matching PIDs before STOP so Linux restart still escalates when cmdline rescans miss the fixture.
Add Live voice chat, Multi-select questions, Subagents, Computer Use, and Generated images rows to all three README matrices, plus a legend and per-row notes grounded in adapter code. Record two live findings: Desktop has no multi-select question card, so multi-select renders as single-choice, and Grok's ask_user_question fails because the ACP client only implements _x.ai/mcp/elicit. Note that Grok-specific behavior depends on the patched grokgod build (0011, 0025, 0026). Align the Chinese and Korean matrices with the English one, which also restores the missing Kimi Code and Copilot columns.
Author
|
Wrong target: this is fork-only documentation (live voice is not an upstream capability). Reopening against karlorz/codex-host. |
|
Caution Review failedThe pull request is closed. Finishing Touches
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Purpose
Extend the Harness feature matrix in all three READMEs with capabilities the fork's maintained build actually has, and record the two question gaps discovered by live test.
New rows: Live voice chat, Multi-select questions, Subagents, Computer Use / in-app browser, Generated images. New legend marks: ◐ partial/degraded/saved intent, ? unverified.
Findings recorded
AskUserQuestion({multiSelect: true})and Alpha/Beta/Gamma rendered as single-choice and returned"Alpha"with no Host error. The Host contract keepsmultiple; adapters that preserve it are marked ◐.ask_user_questionfails over ACP withMethod not found: _x.ai/ask_user_questionbecause the plugin only implements_x.ai/mcp/elicit. Cancellation and MCP consent approvals still work, so the Questions row is ◐ for Grok.Matrix consistency
Chinese was missing the Kimi Code column; Korean was missing Kimi Code and Copilot. All three matrices are now 18 capability rows × 14 Harness columns with identical cells.
Validation
npx prettier --check README.md docs/project/README.zh-CN.md docs/project/README.ko.md docs/harnesses/capability-boundaries.mdgit diff --checkNo source code changed. Live voice: Claude Code ✅ attended Desktop call, Grok ◐ (real gateway call plus isolated Host trial), other Harnesses unverified (?).