Repository navigation
feat(providers): clearer Claude start failures, Codex gateway and security notice - #914
Merged
Merged
Conversation
…urity notice Picks up Claude Code 2.1.319-2.1.325 and Codex 0.168-0.169. Claude - A refused start (a settings file that is not valid JSON, a bad proxy address, a CLI too old) showed four stacked errors and a false "could not restore this thread's previous Claude session" warning: Claude's reason, the SDK's "returned an error result" copy of it, then both again when the send recovered the dead session. Chat sessions now ask Claude Code to write the refusal as a result (CLAUDE_CODE_STARTUP_FAILURE_RESULTS), the adapter reads `startup_failure_reason`, and the thread hears Claude's words once. A fresh session that was refused holds nothing to restore, so the next start skips the lost-context warning. - Since 2.1.321, subagent reports that wait together are answered in one model call, and every other queued report or task notice gets an empty zero-turn result of its own. Each one ended a turn: a message sent while reports landed was marked finished with no answer, and the suggested next message was dropped. Those results are now skipped, ahead of the append-only counter. - Agent SDK 0.3.319 -> 0.3.322, the newest release past the 3-day wait. Codex - Bindings regenerated from 0.169. - When Codex's default model provider is not OpenAI (`modelProvider/list`), the row says "Runs on <name>" and Account & Sign-in has a Model provider line. A ChatGPT sign-in beside it keeps its email, workspace and Sign out. With no account the badge reads "No sign-in needed" instead of "Checking". - OpenAI's account security notice (`account/securitySetup/read`) shows on the Codex card. Its button only opens https pages on openai.com and chatgpt.com, its text is flattened and capped, and it is shown only to the user and workspace it is bound to. A check that cannot read the account forgets it. The gateway's own model is not added to the model list: `model/list` returns OpenAI's catalog for every provider, and Codex text generation runs with --ignore-user-config, so that model would be sent to OpenAI.
|
The latest updates on your projects. Learn more about Vercel for GitHub.
|
This branch was successfully deployed
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.
Picks up what changed in Claude Code 2.1.319 to 2.1.325 and Codex 0.168 to 0.169.
Claude
A refused start says why, once. When Claude Code refuses to start (a settings file that is not valid JSON, a bad proxy address, a CLI too old), the chat showed four stacked errors and a false "could not restore this thread's previous Claude session" warning. The send recovered the dead session and started Claude a second time, and each start reported the reason plus the SDK's own copy of it. Chat sessions now ask Claude Code to write the refusal as a result, the adapter reads
startup_failure_reason, and the thread hears Claude's words once. A fresh session that was refused holds nothing to restore, so the next start skips the lost-context warning.Batched subagent reports stop ending turns. Since 2.1.321, subagent reports that wait together are answered in one model call, and every other queued report or task notice gets an empty zero-turn result of its own. Each one ended a turn: a message sent while reports landed was marked finished with no answer, and the suggested next message was dropped. Those results are now skipped, ahead of the append-only counter.
Agent SDK 0.3.319 to 0.3.322, the newest release past the 3-day wait.
Codex
Bindings regenerated from 0.169.
Model provider. When Codex's default model provider is not OpenAI, the row says "Runs on" and its name, and Account & Sign-in has a Model provider line. A ChatGPT sign-in beside it keeps its email, workspace and Sign out. With no account, the badge reads "No sign-in needed" instead of a "Checking" that never finished.
Security notice. OpenAI's account security notice shows on the Codex card. Its button only opens https pages on openai.com and chatgpt.com, its text is flattened and capped, and it is shown only to the user and workspace it is bound to. A check that cannot read the account forgets it.
Not in this PR
model/listreturns OpenAI's catalog for every provider, and Codex text generation runs with--ignore-user-config, so that model would be sent to OpenAI.Checked
vp fmt,vp lint,vp run typecheck, full unit suite, browser suite (two unrelated specs failed under load and pass alone:ProviderUpdateProgress,steppedLoops).