Area
apps/desktop (provider maintenance / backend startup)
Steps to reproduce
- Have a CLI-backed provider enabled that is slightly out of date (in my case OpenCode / Codex).
- Launch the desktop app (cold start).
- Inspect the backend traces under
~/.t3/userdata/logs/server.trace.ndjson and desktop.trace.ndjson.
Expected behavior
Provider auto-update / maintenance should not block or significantly delay the backend becoming ready. The window/backend readiness should not wait on a network npm install, and any provider update should be deferred, backgrounded, and cancellable.
Actual behavior
On startup the backend runs a provider update that executes npm install <package>@latest and dominates the cold start. Slowest startup-phase spans from a single launch:
+3.75s 47729ms ProviderMaintenanceRunner.updateProvider
+3.75s 47727ms ProviderMaintenanceRunner.runProviderUpdate
+3.75s 47727ms ProviderMaintenanceRunner.runCommandAndVerify
+3.75s 47727ms ProviderMaintenanceRunner.runCommand
+3.75s 47061ms ProviderMaintenanceRunner.collectCommandResult
+0.36s 3705ms checkOpenCodeProviderStatus
+1.87s 1190ms opencode.provider.list
desktop.backendManager.waitForHttpReady was ~12s; the provider update (npm install @latest) ran ~47s on top of that. The update command (${definition.npmPackageName}@latest) appears to run automatically as part of maintenance rather than only when the user accepts an update notification.
Impact
Major degradation or frequent failure (very slow / heavy cold start, especially on slow networks).
Version or commit
0.0.27 (Linux AppImage, packaged as the t3code-bin AUR package)
Environment
Arch Linux, Hyprland (Wayland), zsh. Providers enabled: Claude Agent, Codex. OpenCode present.
Logs or stack traces
updateProvider span attributes are empty, so the exact provider isn't recorded in the span; the surrounding activity (checkOpenCodeProviderStatus, opencode.provider.list) points to the OpenCode/Codex CLI being updated via npm.
Workaround
None found that is clean. There is a T3CODE_DISABLE_AUTO_UPDATE env var, but it appears to gate the Electron app updater, not provider maintenance. A documented way to disable provider auto-update at startup (or to make it strictly opt-in / background) would help.
Area
apps/desktop (provider maintenance / backend startup)
Steps to reproduce
~/.t3/userdata/logs/server.trace.ndjsonanddesktop.trace.ndjson.Expected behavior
Provider auto-update / maintenance should not block or significantly delay the backend becoming ready. The window/backend readiness should not wait on a network
npm install, and any provider update should be deferred, backgrounded, and cancellable.Actual behavior
On startup the backend runs a provider update that executes
npm install <package>@latestand dominates the cold start. Slowest startup-phase spans from a single launch:desktop.backendManager.waitForHttpReadywas ~12s; the provider update (npm install @latest) ran ~47s on top of that. The update command (${definition.npmPackageName}@latest) appears to run automatically as part of maintenance rather than only when the user accepts an update notification.Impact
Major degradation or frequent failure (very slow / heavy cold start, especially on slow networks).
Version or commit
0.0.27 (Linux AppImage, packaged as the
t3code-binAUR package)Environment
Arch Linux, Hyprland (Wayland), zsh. Providers enabled: Claude Agent, Codex. OpenCode present.
Logs or stack traces
updateProviderspan attributes are empty, so the exact provider isn't recorded in the span; the surrounding activity (checkOpenCodeProviderStatus,opencode.provider.list) points to the OpenCode/Codex CLI being updated via npm.Workaround
None found that is clean. There is a
T3CODE_DISABLE_AUTO_UPDATEenv var, but it appears to gate the Electron app updater, not provider maintenance. A documented way to disable provider auto-update at startup (or to make it strictly opt-in / background) would help.