fix(host-transfer): remove auto-resolve for single clients to restore backward compat#29462
fix(host-transfer): remove auto-resolve for single clients to restore backward compat#29462credence-the-bot[bot] wants to merge 2 commits into
Conversation
…hen disabled - Add 'analyze-conversation' assistant-scope flag to feature-flag-registry.json (defaultEnabled: false) - ConversationTitleActionsControl + ConversationActionsMenuContent: make onAnalyzeConversation optional ((() -> Void)? = nil); gate VMenuItem on presence of non-nil closure - ConversationTitleOverlay: make onAnalyzeConversation optional to propagate nil cleanly through the chain - TopBarView: inject AssistantFeatureFlagStore via @Environment; pass nil for onAnalyzeConversation when flag is off - SidebarView: gate onAnalyze on assistantFeatureFlagStore.isEnabled("analyze-conversation") - ConversationSwitcherDrawer: inject AssistantFeatureFlagStore via @Environment; gate onAnalyze the same way
… backward compat Phase 3 (PR #29440) added auto-resolution of targetClientId when exactly one host_file-capable client is connected. This causes the content routes to require the x-vellum-client-id header on all GET/PUT transfer requests, even for untargeted calls. Pre-Phase 3 macOS clients (e.g. 0.7.1-staging.5, both Vellum.app and Vellum Staging.app still at May 1 builds) do not send this header. Result: the content GET returns 400, the Swift client throws rather than calling post-transfer-result, and the pending interaction times out after 120 seconds. Fix: remove the auto-resolve block from both requestToHost() and requestToSandbox(). Untargeted transfers broadcast without targetClientId — backward-compatible with all existing macOS clients. Explicit target_client_id from tool input still routes correctly (Phase 3 feature intact). Tests updated: auto-resolve assertions flipped to expect undefined targetClientId for untargeted single-client calls.
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
![chatgpt-codex-connector[bot]](https://avatars.githubusercontent.com/in/1144995?s=60&v=4){kind=small}
There was a problem hiding this comment.
💡 Codex Review
Here are some automated review suggestions for this pull request.
Reviewed commit: 3284903916
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".
| "key": "analyze-conversation", | ||
| "label": "Analyze Conversation", | ||
| "description": "Show the 'Analyze' / 'Analyze conversation' option in conversation context menus and the conversation title actions dropdown.", | ||
| "defaultEnabled": false |
There was a problem hiding this comment.
This change gates every manual Analyze entry point behind analyze-conversation, but the newly introduced registry default is false, so existing users lose the Analyze action immediately after upgrade unless a per-assistant override is already present. Because this commit does not add any bootstrap/migration path that seeds the flag to true, it introduces a user-facing behavior regression unrelated to the host-transfer fix.
Useful? React with 👍 / 👎.
Problem
Phase 3 (PR #29440) added auto-resolution of
targetClientIdwhen exactly onehost_file-capable client is connected. This makes the content routes require thex-vellum-client-idheader on all GET/PUT transfer requests, even for untargeted calls.Pre-Phase 3 macOS clients (
0.7.1-staging.5; bothVellum.appandVellum Staging.appstill at May 1 builds) do not send this header. Result: the content GET returns 400, the Swift client throws rather than callingpost-transfer-result, and the pending interaction times out after 120 seconds — every time, consistently.Diagnosis
From feedback bundle
feedback-019df0a7-99a9-7446-8bdf-6228ec86ad08:/Applications/Vellum.appand/Applications/Vellum Staging.apphavevellum-daemonbinaries timestamped May 1 (pre-Phase 3)if (capable.length === 1) resolvedTargetClientId = capable[0].clientId→ makesx-vellum-client-idmandatory on content routesHostProxyClientdoes not sendX-Vellum-Client-Id→ 400 onGET /v1/transfers/:transferId/content→TransferError.pullFailed(400)thrown →host-transfer-resultnever posted → 120 s hangFix
Remove the auto-resolve block from both
requestToHost()andrequestToSandbox(). Untargeted transfers broadcast withouttargetClientId— backward-compatible with all existing macOS clients. Explicittarget_client_idfrom the tool input still routes correctly (Phase 3 cross-client targeting feature intact).Changes
assistant/src/daemon/host-transfer-proxy.ts— remove auto-resolveelseblocks from both methodsassistant/src/__tests__/host-transfer-proxy-targeted.test.ts— flip the two auto-resolve test cases to asserttargetClientIdisundefinedfor untargeted single-client callsAll 47 proxy tests pass.