Make server settings authoritative for runtime behavior#1421
Merged
juliusmarminge merged 30 commits intomainfrom Mar 26, 2026
Merged
Make server settings authoritative for runtime behavior#1421juliusmarminge merged 30 commits intomainfrom
juliusmarminge merged 30 commits intomainfrom
Conversation
- Move settings persistence and validation into the server - Route runtime/provider behavior through server settings - Update client and tests to use the unified settings flow
github-actions
bot
added
size:XL
|
Important Review skippedAuto reviews are disabled on this repository. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Repository UI Review profile: CHILL Plan: Pro Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
✨ Finishing Touches🧪 Generate unit tests (beta)
Comment |
![cursor[bot]](https://avatars.githubusercontent.com/in/1210556?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
Autofix Details
Bugbot Autofix prepared fixes for both issues found in the latest run.
- ✅ Fixed: Decider test asserts removed event field
- Removed the assertion for
assistantDeliveryModein the test since the decider no longer emits that field in thethread.turn-start-requestedevent payload.
- Removed the assertion for
- ✅ Fixed: Stale dependency in useCallback after removing usage
- Removed
settings.enableAssistantStreamingfrom theonSendOrRetryuseCallback dependency array since it is no longer referenced in the callback body.
- Removed
Or push these changes by commenting:
@cursor push e35da91f49
Preview (e35da91f49)
diff --git a/apps/server/src/orchestration/decider.projectScripts.test.ts b/apps/server/src/orchestration/decider.projectScripts.test.ts
--- a/apps/server/src/orchestration/decider.projectScripts.test.ts
+++ b/apps/server/src/orchestration/decider.projectScripts.test.ts
@@ -188,7 +188,6 @@
if (turnStartEvent?.type !== "thread.turn-start-requested") {
return;
}
- expect(turnStartEvent.payload.assistantDeliveryMode).toBe("buffered");
expect(turnStartEvent.payload).toMatchObject({
threadId: ThreadId.makeUnsafe("thread-1"),
messageId: asMessageId("message-user-1"),
diff --git a/apps/web/src/components/ChatView.tsx b/apps/web/src/components/ChatView.tsx
--- a/apps/web/src/components/ChatView.tsx
+++ b/apps/web/src/components/ChatView.tsx
@@ -2972,7 +2972,6 @@
selectedProvider,
setComposerDraftInteractionMode,
setThreadError,
- settings.enableAssistantStreaming,
selectedModel,
],
);![macroscopeapp[bot]](https://avatars.githubusercontent.com/in/900172?s=60&v=4){kind=small}
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 2 potential issues.
Autofix Details
Bugbot Autofix prepared fixes for both issues found in the latest run.
- ✅ Fixed: Empty string binary path breaks CLI command spawning
- Fixed by guarding deriveProviderStartOptions on settings.claude.binaryPath being truthy (not just the object) and switching from ?? to || for binaryPath fallbacks in ClaudeTextGeneration and CodexTextGeneration so empty strings correctly fall back to default CLI names.
- ✅ Fixed: Migration skips old flat binary path keys
- Added explicit migration of old flat localStorage keys (claudeBinaryPath, codexBinaryPath, codexHomePath) into the new nested codex/claude objects in migrateLocalSettingsToServer.
Or push these changes by commenting:
@cursor push f8c9d71187
Preview (f8c9d71187)
diff --git a/apps/server/src/git/Layers/ClaudeTextGeneration.ts b/apps/server/src/git/Layers/ClaudeTextGeneration.ts
--- a/apps/server/src/git/Layers/ClaudeTextGeneration.ts
+++ b/apps/server/src/git/Layers/ClaudeTextGeneration.ts
@@ -95,7 +95,7 @@
const runClaudeCommand = Effect.gen(function* () {
const command = ChildProcess.make(
- claudeSettings?.binaryPath ?? "claude",
+ claudeSettings?.binaryPath || "claude",
[
"-p",
"--output-format",
diff --git a/apps/server/src/git/Layers/CodexTextGeneration.ts b/apps/server/src/git/Layers/CodexTextGeneration.ts
--- a/apps/server/src/git/Layers/CodexTextGeneration.ts
+++ b/apps/server/src/git/Layers/CodexTextGeneration.ts
@@ -153,7 +153,7 @@
const reasoningEffort =
modelSelection.options?.reasoningEffort ?? CODEX_GIT_TEXT_GENERATION_REASONING_EFFORT;
const command = ChildProcess.make(
- codexSettings?.binaryPath ?? "codex",
+ codexSettings?.binaryPath || "codex",
[
"exec",
"--ephemeral",
diff --git a/apps/server/src/serverSettings.ts b/apps/server/src/serverSettings.ts
--- a/apps/server/src/serverSettings.ts
+++ b/apps/server/src/serverSettings.ts
@@ -97,7 +97,7 @@
},
}
: {}),
- ...(settings.claude
+ ...(settings.claude?.binaryPath
? {
claudeAgent: {
binaryPath: settings.claude.binaryPath,
diff --git a/apps/web/src/hooks/useSettings.ts b/apps/web/src/hooks/useSettings.ts
--- a/apps/web/src/hooks/useSettings.ts
+++ b/apps/web/src/hooks/useSettings.ts
@@ -159,6 +159,33 @@
}
}
+ // Migrate old flat binary-path keys into their new nested shape
+ const codexBinaryPath = old["codexBinaryPath"];
+ const codexHomePath = old["codexHomePath"];
+ const claudeBinaryPath = old["claudeBinaryPath"];
+
+ if (
+ (typeof codexBinaryPath === "string" && codexBinaryPath) ||
+ (typeof codexHomePath === "string" && codexHomePath)
+ ) {
+ const existing = (serverPatch["codex"] as Record<string, unknown> | undefined) ?? {};
+ serverPatch["codex"] = {
+ ...existing,
+ ...(typeof codexBinaryPath === "string" && codexBinaryPath
+ ? { binaryPath: codexBinaryPath }
+ : {}),
+ ...(typeof codexHomePath === "string" && codexHomePath ? { homePath: codexHomePath } : {}),
+ };
+ }
+
+ if (typeof claudeBinaryPath === "string" && claudeBinaryPath) {
+ const existing = (serverPatch["claude"] as Record<string, unknown> | undefined) ?? {};
+ serverPatch["claude"] = {
+ ...existing,
+ binaryPath: claudeBinaryPath,
+ };
+ }
+
if (Object.keys(serverPatch).length > 0) {
void ensureNativeApi()
.server.updateSettings(serverPatch as ServerSettingsPatch)- Move Codex and Claude config lookups to `settings.providers` - Update web composer, settings UI, and model selection to use unified settings - Refresh browser fixtures and contracts for the new settings shape
github-actions
bot
added
size:XXL
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 3 total unresolved issues (including 2 from previous reviews).
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Empty string binary path not handled by nullish coalescing
- Replaced
??with||in both ClaudeTextGeneration.ts and CodexTextGeneration.ts so that an empty string binaryPath correctly falls back to the default CLI name.
- Replaced
Or push these changes by commenting:
@cursor push f74ee02af1
Preview (f74ee02af1)
diff --git a/apps/server/src/git/Layers/ClaudeTextGeneration.ts b/apps/server/src/git/Layers/ClaudeTextGeneration.ts
--- a/apps/server/src/git/Layers/ClaudeTextGeneration.ts
+++ b/apps/server/src/git/Layers/ClaudeTextGeneration.ts
@@ -95,7 +95,7 @@
const runClaudeCommand = Effect.gen(function* () {
const command = ChildProcess.make(
- claudeSettings?.binaryPath ?? "claude",
+ claudeSettings?.binaryPath || "claude",
[
"-p",
"--output-format",
diff --git a/apps/server/src/git/Layers/CodexTextGeneration.ts b/apps/server/src/git/Layers/CodexTextGeneration.ts
--- a/apps/server/src/git/Layers/CodexTextGeneration.ts
+++ b/apps/server/src/git/Layers/CodexTextGeneration.ts
@@ -153,7 +153,7 @@
const reasoningEffort =
modelSelection.options?.reasoningEffort ?? CODEX_GIT_TEXT_GENERATION_REASONING_EFFORT;
const command = ChildProcess.make(
- codexSettings?.binaryPath ?? "codex",
+ codexSettings?.binaryPath || "codex",
[
"exec",
"--ephemeral",- Allow server settings to decode without an explicit providers block
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 3 total unresolved issues (including 2 from previous reviews).
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Shallow merge in updateSettings overwrites nested providers object
- Deep-merged the
providersobject inupdateSettingsso that when a patch includesproviders, each sub-provider (codex, claudeAgent) is merged with the current values instead of being replaced wholesale.
- Deep-merged the
Or push these changes by commenting:
@cursor push bba0a8a3d0
Preview (bba0a8a3d0)
diff --git a/apps/server/src/serverSettings.ts b/apps/server/src/serverSettings.ts
--- a/apps/server/src/serverSettings.ts
+++ b/apps/server/src/serverSettings.ts
@@ -271,7 +271,19 @@
writeSemaphore.withPermits(1)(
Effect.gen(function* () {
const current = yield* getSettingsFromCache;
- const next: ServerSettings = { ...current, ...patch };
+ const next: ServerSettings = {
+ ...current,
+ ...patch,
+ providers: patch.providers
+ ? {
+ codex: { ...current.providers.codex, ...patch.providers.codex },
+ claudeAgent: {
+ ...current.providers.claudeAgent,
+ ...patch.providers.claudeAgent,
+ },
+ }
+ : current.providers,
+ };
yield* writeSettingsAtomically(next);
yield* Cache.set(settingsCache, cacheKey, next);
yield* emitChange(next);
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 4 total unresolved issues (including 3 from previous reviews).
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Empty string binary path breaks nullish coalescing fallback
- Replaced
??with||in both CodexTextGeneration.ts and ClaudeTextGeneration.ts so the empty string default fromSchema.withDecodingDefault(() => "")correctly falls back to "codex" and "claude" respectively.
- Replaced
Or push these changes by commenting:
@cursor push 98e6f51693
Preview (98e6f51693)
diff --git a/apps/server/src/git/Layers/ClaudeTextGeneration.ts b/apps/server/src/git/Layers/ClaudeTextGeneration.ts
--- a/apps/server/src/git/Layers/ClaudeTextGeneration.ts
+++ b/apps/server/src/git/Layers/ClaudeTextGeneration.ts
@@ -95,7 +95,7 @@
const runClaudeCommand = Effect.gen(function* () {
const command = ChildProcess.make(
- claudeSettings?.binaryPath ?? "claude",
+ claudeSettings?.binaryPath || "claude",
[
"-p",
"--output-format",
diff --git a/apps/server/src/git/Layers/CodexTextGeneration.ts b/apps/server/src/git/Layers/CodexTextGeneration.ts
--- a/apps/server/src/git/Layers/CodexTextGeneration.ts
+++ b/apps/server/src/git/Layers/CodexTextGeneration.ts
@@ -153,7 +153,7 @@
const reasoningEffort =
modelSelection.options?.reasoningEffort ?? CODEX_GIT_TEXT_GENERATION_REASONING_EFFORT;
const command = ChildProcess.make(
- codexSettings?.binaryPath ?? "codex",
+ codexSettings?.binaryPath || "codex",
[
"exec",
"--ephemeral",
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: deepMerge corrupts arrays by merging them as objects
- Added Array.isArray guards at both the top-level early return and the per-key recursive branch in deepMerge so arrays are replaced wholesale instead of being spread into plain objects and merged index-by-index.
Or push these changes by commenting:
@cursor push 2388e7d43f
Preview (2388e7d43f)
diff --git a/packages/shared/src/Struct.ts b/packages/shared/src/Struct.ts
--- a/packages/shared/src/Struct.ts
+++ b/packages/shared/src/Struct.ts
@@ -1,7 +1,12 @@
import * as P from "effect/Predicate";
export function deepMerge<T>(current: T, patch: unknown): T {
- if (!P.isObject(current) || !P.isObject(patch)) {
+ if (
+ !P.isObject(current) ||
+ !P.isObject(patch) ||
+ Array.isArray(current) ||
+ Array.isArray(patch)
+ ) {
return patch as T;
}
@@ -10,7 +15,10 @@
if (value === undefined) continue;
const existing = next[key];
- next[key] = P.isObject(existing) && P.isObject(value) ? deepMerge(existing, value) : value;
+ next[key] =
+ P.isObject(existing) && P.isObject(value) && !Array.isArray(existing) && !Array.isArray(value)
+ ? deepMerge(existing, value)
+ : value;
}
return next as T;
- Read Codex and Claude CLI paths from server settings - Add provider registry and status layers for health checks - Update session startup, model selection, and UI status handling
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Child process env missing inherited process.env variables
- Spread process.env into the env option before the CODEX_HOME override, matching the pattern used in codexAppServerManager.ts and CodexProvider.ts.
Or push these changes by commenting:
@cursor push 42afbb2266
Preview (42afbb2266)
diff --git a/apps/server/src/git/Layers/CodexTextGeneration.ts b/apps/server/src/git/Layers/CodexTextGeneration.ts
--- a/apps/server/src/git/Layers/CodexTextGeneration.ts
+++ b/apps/server/src/git/Layers/CodexTextGeneration.ts
@@ -172,7 +172,10 @@
"-",
],
{
- env: codexSettings?.homePath ? { CODEX_HOME: codexSettings.homePath } : {},
+ env: {
+ ...process.env,
+ ...(codexSettings?.homePath ? { CODEX_HOME: codexSettings.homePath } : {}),
+ },
cwd,
shell: process.platform === "win32",
stdin: {
Comment on lines
+46
to
+48
| yield* Stream.runForEach(input.streamSettings, (nextSettings) => | ||
| Effect.asVoid(reconcileSnapshot(nextSettings)), | ||
| ).pipe(Effect.forkScoped); |
Contributor
There was a problem hiding this comment.
🟠 High provider/makeManagedServerProvider.ts:46
The settings stream fiber crashes on reconcileSnapshot failure and stops processing settings changes. Unlike the periodic refresh fiber (lines 50-55) which catches and logs errors with Effect.ignoreCause, the stream handler on lines 46-48 propagates errors unhandled, causing the fiber to terminate. Consider wrapping reconcileSnapshot with Effect.ignoreCause({ log: true }) so stream processing continues after errors.
- yield* Stream.runForEach(input.streamSettings, (nextSettings) =>
- Effect.asVoid(reconcileSnapshot(nextSettings)),
- ).pipe(Effect.forkScoped);
+ yield* Stream.runForEach(input.streamSettings, (nextSettings) =>
+ reconcileSnapshot(nextSettings).pipe(Effect.ignoreCause({ log: true })),
+ ).pipe(Effect.forkScoped);Also found in 1 other location(s)
apps/server/src/orchestration/Layers/ProviderCommandReactor.ts:432
The
yield* Effect.map(serverSettingsService.getSettings, ...)on line 432-435 can fail withServerSettingsError, but this error is not caught by the.pipe(Effect.catch(...), Effect.catchCause(...))handlers on lines 437-468. Theyield*forgetSettingsexecutes before thegenerateBranchNamecall, so if it fails, the error propagates up through the generator and bypasses the error handlers that were intended to log warnings for failures in this function. Before this change,modelSelectionwas a static object that couldn't fail. The fix should either move thegetSettingscall outside the object literal and wrap the whole block in error handling, or useEffect.flatMapto sequence the operations so failures are captured by the existingcatchCause.
🚀 Reply "fix it for me" or copy this AI Prompt for your agent:
In file apps/server/src/provider/makeManagedServerProvider.ts around lines 46-48:
The settings stream fiber crashes on `reconcileSnapshot` failure and stops processing settings changes. Unlike the periodic refresh fiber (lines 50-55) which catches and logs errors with `Effect.ignoreCause`, the stream handler on lines 46-48 propagates errors unhandled, causing the fiber to terminate. Consider wrapping `reconcileSnapshot` with `Effect.ignoreCause({ log: true })` so stream processing continues after errors.
Evidence trail:
- `apps/server/src/provider/makeManagedServerProvider.ts` lines 46-48 (settings stream without error handling) vs lines 50-55 (periodic refresh with `Effect.ignoreCause({ log: true })`)
- `apps/server/src/serverSettings.ts` lines 267-271 shows correct pattern: `Stream.runForEach(...).pipe(Effect.ignoreCause({ log: true }), Effect.forkIn(...))`
- `apps/server/src/keybindings.ts` lines 836-840 uses same correct pattern
- Commit: REVIEWED_COMMIT
Also found in 1 other location(s):
- apps/server/src/orchestration/Layers/ProviderCommandReactor.ts:432 -- The `yield* Effect.map(serverSettingsService.getSettings, ...)` on line 432-435 can fail with `ServerSettingsError`, but this error is not caught by the `.pipe(Effect.catch(...), Effect.catchCause(...))` handlers on lines 437-468. The `yield*` for `getSettings` executes *before* the `generateBranchName` call, so if it fails, the error propagates up through the generator and bypasses the error handlers that were intended to log warnings for failures in this function. Before this change, `modelSelection` was a static object that couldn't fail. The fix should either move the `getSettings` call outside the object literal and wrap the whole block in error handling, or use `Effect.flatMap` to sequence the operations so failures are captured by the existing `catchCause`.
- Track settings updates in the test service - Point Codex tests at the fake binary via server settings instead of env vars
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
There are 2 total unresolved issues (including 1 from previous review).
Autofix Details
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Duplicate server settings layer creates inconsistent state
- Replaced the duplicate
ServerSettingsService.layerTest()call at line 327 with the existingserverSettingsLayervariable so all layers share the same Ref state.
- Replaced the duplicate
Or push these changes by commenting:
@cursor push a70e7d7e49
Preview (a70e7d7e49)
diff --git a/apps/server/integration/OrchestrationEngineHarness.integration.ts b/apps/server/integration/OrchestrationEngineHarness.integration.ts
--- a/apps/server/integration/OrchestrationEngineHarness.integration.ts
+++ b/apps/server/integration/OrchestrationEngineHarness.integration.ts
@@ -324,7 +324,7 @@
);
const layer = orchestrationReactorLayer.pipe(
Layer.provide(persistenceLayer),
- Layer.provideMerge(ServerSettingsService.layerTest()),
+ Layer.provideMerge(serverSettingsLayer),
Layer.provideMerge(ServerConfig.layerTest(workspaceDir, rootDir)),
Layer.provideMerge(NodeServices.layer),
);- Route Codex and Claude health checks through ServerSettingsService - Reuse layerTest helpers in provider tests and orchestration tests - Update provider snapshots and structs for the new settings authority
- Pass the full process environment to the Codex child process - Keep CODEX_HOME override when a custom settings home is configured
- Guard `start()` with an atomic started flag - Await the existing deferred when startup has already begun
- compare provider snapshots before broadcasting registry updates - add regression coverage for equal provider lists
Contributor
There was a problem hiding this comment.
Cursor Bugbot has reviewed your changes and found 1 potential issue.
Bugbot Autofix prepared a fix for the issue found in the latest run.
- ✅ Fixed: Settings error escapes error handlers in branch rename
- Moved the serverSettingsService.getSettings call into the Effect chain via Effect.flatMap so that settings failures are now caught by the downstream Effect.catch/Effect.catchCause handlers instead of escaping the generator.
Or push these changes by commenting:
@cursor push 361b51678a
Preview (361b51678a)
diff --git a/apps/server/src/orchestration/Layers/ProviderCommandReactor.ts b/apps/server/src/orchestration/Layers/ProviderCommandReactor.ts
--- a/apps/server/src/orchestration/Layers/ProviderCommandReactor.ts
+++ b/apps/server/src/orchestration/Layers/ProviderCommandReactor.ts
@@ -424,48 +424,52 @@
const oldBranch = input.branch;
const cwd = input.worktreePath;
const attachments = input.attachments ?? [];
- yield* textGeneration
- .generateBranchName({
- cwd,
- message: input.messageText,
- ...(attachments.length > 0 ? { attachments } : {}),
- modelSelection: yield* Effect.map(
- serverSettingsService.getSettings,
- (settings) => settings.textGenerationModelSelection,
+ yield* Effect.flatMap(
+ Effect.map(
+ serverSettingsService.getSettings,
+ (settings) => settings.textGenerationModelSelection,
+ ),
+ (modelSelection) =>
+ textGeneration.generateBranchName({
+ cwd,
+ message: input.messageText,
+ ...(attachments.length > 0 ? { attachments } : {}),
+ modelSelection,
+ }),
+ ).pipe(
+ Effect.catch((error) =>
+ Effect.logWarning(
+ "provider command reactor failed to generate worktree branch name; skipping rename",
+ { threadId: input.threadId, cwd, oldBranch, reason: error.message },
),
- })
- .pipe(
- Effect.catch((error) =>
- Effect.logWarning(
- "provider command reactor failed to generate worktree branch name; skipping rename",
- { threadId: input.threadId, cwd, oldBranch, reason: error.message },
- ),
- ),
- Effect.flatMap((generated) => {
- if (!generated) return Effect.void;
+ ),
+ Effect.flatMap((generated) => {
+ if (!generated) return Effect.void;
- const targetBranch = buildGeneratedWorktreeBranchName(generated.branch);
- if (targetBranch === oldBranch) return Effect.void;
+ const targetBranch = buildGeneratedWorktreeBranchName(generated.branch);
+ if (targetBranch === oldBranch) return Effect.void;
- return Effect.flatMap(
- git.renameBranch({ cwd, oldBranch, newBranch: targetBranch }),
- (renamed) =>
- orchestrationEngine.dispatch({
- type: "thread.meta.update",
- commandId: serverCommandId("worktree-branch-rename"),
- threadId: input.threadId,
- branch: renamed.branch,
- worktreePath: cwd,
- }),
- );
+ return Effect.flatMap(
+ git.renameBranch({ cwd, oldBranch, newBranch: targetBranch }),
+ (renamed) =>
+ orchestrationEngine.dispatch({
+ type: "thread.meta.update",
+ commandId: serverCommandId("worktree-branch-rename"),
+ threadId: input.threadId,
+ branch: renamed.branch,
+ worktreePath: cwd,
+ }),
+ );
+ }),
+ Effect.catchCause((cause) =>
+ Effect.logWarning("provider command reactor failed to generate or rename worktree branch", {
+ threadId: input.threadId,
+ cwd,
+ oldBranch,
+ cause: Cause.pretty(cause),
}),
- Effect.catchCause((cause) =>
- Effect.logWarning(
- "provider command reactor failed to generate or rename worktree branch",
- { threadId: input.threadId, cwd, oldBranch, cause: Cause.pretty(cause) },
- ),
- ),
- );
+ ),
+ );
});
const processTurnStartRequested = Effect.fnUntraced(function* (- Expose `server.refreshProviders` over WebSocket and NativeApi - Add a settings UI refresh button with last-checked relative time - Refresh provider cache and invalidate server config queries
- Read text generation model selection from server settings before branch generation - Simplify branch rename flow and keep failures logged at the reactor boundary
- Remove leftover temp files after renaming keybindings and server settings - Keep atomic write flow while avoiding stale temporary artifacts
aaditagrawal
pushed a commit
to aaditagrawal/t3code
that referenced
this pull request
Mar 26, 2026
6 tasks
bulbulogludemir
added a commit
to bulbulogludemir/krabbycode
that referenced
this pull request
Mar 26, 2026
…ings architecture Major upstream changes merged: - Server-authoritative settings (pingdotgg#1421): appSettings.ts deleted, settings moved to serverSettings.ts + contracts/settings.ts + useSettings hook - Provider-aware model selections (pingdotgg#1371): model→modelSelection refactor, new composerDraftStore, unified TraitsPicker, migration 016 - Context window usage UI (pingdotgg#1351): ContextWindowMeter component - Claude as TextGenerationProvider (pingdotgg#1323): git text generation via Claude - Bootstrap FD for security (pingdotgg#1398): sensitive config sent over file descriptor - Sidebar recency sorting (pingdotgg#1372): projects/threads sorted by recency - ProviderHealth replaced by ProviderRegistry + per-provider services - Word wrapping, terminal history preservation, various fixes Krabby-specific adaptations: - Migration 016 renumbered to 024 (Krabby 016-023 preserved) - 20+ Krabby settings ported to new ClientSettingsSchema - KrabbyProviderOptions schema for chrome/enterprise/budget fields - All @t3tools→@kRaBby, t3code→krabbycode, service keys t3/→krabby/ Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
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
2 participants
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.
Closes #1033
Closes #1289
Summary
serverSettingsservice with disk-backed load/save, change notifications, and WebSocket exposure.Testing
ProviderRuntimeIngestion,ProviderCommandReactor, andmainwiring.Note
Medium Risk
Changes core runtime behavior by moving provider startup parameters, text-generation model selection, and assistant streaming decisions into persisted server settings; misconfiguration or settings migration issues could affect provider launch and message delivery mode.
Overview
Makes server-owned settings the source of truth for runtime behavior. Introduces
ServerSettingsService(disk-backedsettings.json, validation, change stream, debounced file watching) and wires it into server startup and test harnesses.Provider and git text-generation flows now read from server settings instead of client-supplied fields or hard-coded defaults: Codex/Claude CLI
binaryPath/homePathare pulled from settings,GitManager/ProviderCommandReactoruse the settings’textGenerationModelSelection, and assistant delta delivery switches between buffered/streaming based onenableAssistantStreaming.Replaces the old startup-time provider health layer with a live
ProviderRegistrybacked by newCodexProvider/ClaudeProvidermanaged-provider layers (status/version/auth probing + model capability/normalization helpers), and updates keybindings/config writing to be more robust (lenient JSON parse, atomic write cleanup, debounced watch events).Written by Cursor Bugbot for commit 21838dc. This will update automatically on new commits. Configure here.
Note
Make server settings authoritative for provider config, model selection, and assistant delivery mode
ServerSettingsServicewith persistent JSON storage, deep-merge patch semantics, and a change stream; settings are now the authoritative source for provider binary paths, model selection, and assistant streaming mode.ProviderRegistryaggregating Codex and Claude snapshots (installation status, version, auth, models) with periodic refresh and change broadcasting; replaces the oldProviderHealthservice.GitManager), branch name generation (ProviderCommandReactor), and assistant delivery mode (ProviderRuntimeIngestion) now read configuration from server settings instead of request payloads.providerOptionsandassistantDeliveryModefrom orchestration contracts (ProviderSessionStartInput,ThreadTurnStartCommand,GitRunStackedActionInput); callers can no longer override these per-request.useSettings/useUpdateSettingshook pair in the web client that merges server and client settings and routes patches to the correct backing store.resolveModelSlugno longer falls back to a provider default when the slug is unknown; it returns the normalized slug unconditionally, which may surface unresolved slugs where callers previously received a safe default.Macroscope summarized 21838dc.