diff --git a/CHANGELOG.md b/CHANGELOG.md
index 453e9377..f8e157f1 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -5,6 +5,23 @@ All notable changes to this project will be documented in this file.
 The format is based on [Keep a Changelog](https://keepachangelog.com/en/1.1.0/),
 and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0.html).
 
+## [0.9.20] - 2026-08-01
+
+### Added
+
+- 💬 **Chat settings have their own Admin page.** Admins can now turn chat title generation on or off, choose the models used for titles, summaries, and approval reviews, and adjust when long chats are compacted.
+- 🗂️ **Git settings have their own Admin page.** Admins can now choose which model drafts commit messages.
+- 🧠 **Memory and skills reviews can use a chosen model.** Background reviews can now use a specific model, or keep following the current chat model when left unset.
+
+### Changed
+
+- 🧭 **Model choices are clearer for helper features.** Model pickers now show when a feature is using the current or default model instead of forcing a saved choice.
+- ⚙️ **Admin settings are easier to scan.** Chat and Git options moved out of Models into dedicated pages, leaving model setup focused on model setup.
+
+### Fixed
+
+- 🤖 **Auto mode asks at smarter moments.** Computer can now approve safe, expected tool use in Auto mode and still asks when an action looks risky or unclear.
+
 ## [0.9.19] - 2026-07-31
 
 ### Changed
diff --git a/cptr/frontend/src/lib/apis/memory.ts b/cptr/frontend/src/lib/apis/memory.ts
index fb9bbcd8..a660aac6 100644
--- a/cptr/frontend/src/lib/apis/memory.ts
+++ b/cptr/frontend/src/lib/apis/memory.ts
@@ -18,6 +18,7 @@ export type MemorySettings = {
 	enabled: boolean;
 	tool_enabled: boolean;
 	background_review_enabled: boolean;
+	'background_review.model'?: string | null;
 	review_interval_turns: number;
 	user_char_limit: number;
 	workspace_char_limit: number;
diff --git a/cptr/frontend/src/lib/components/Admin/Chat.svelte b/cptr/frontend/src/lib/components/Admin/Chat.svelte
new file mode 100644
index 00000000..2f8f1369
--- /dev/null
+++ b/cptr/frontend/src/lib/components/Admin/Chat.svelte
@@ -0,0 +1,190 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+	import { toast } from 'svelte-sonner';
+	import { getAdminConfig, updateConfig } from '$lib/apis/admin';
+	import { t } from '$lib/i18n';
+	import Spinner from '$lib/components/common/Spinner.svelte';
+	import ToggleSwitch from '$lib/components/common/ToggleSwitch.svelte';
+	import ModelSelector from '$lib/components/common/ModelSelector.svelte';
+
+	let loading = $state(true);
+	let saving = $state(false);
+	let titleGenerationEnabled = $state(true);
+	let titleGenerationModel = $state<string | null>(null);
+	let contextCompactionModel = $state<string | null>(null);
+	let toolApprovalReviewModel = $state<string | null>(null);
+	let compactTokenThreshold = $state(80000);
+
+	onMount(async () => {
+		try {
+			const config = await getAdminConfig();
+			titleGenerationEnabled =
+				config['chat.title_generation.enabled'] !== false &&
+				config['chat.title_generation.enabled'] !== 'false';
+			titleGenerationModel =
+				typeof config['chat.title_generation.model'] === 'string'
+					? config['chat.title_generation.model']
+					: null;
+			contextCompactionModel =
+				typeof config['chat.context_compaction.model'] === 'string'
+					? config['chat.context_compaction.model']
+					: null;
+			toolApprovalReviewModel =
+				typeof config['tool_approval.review.model'] === 'string'
+					? config['tool_approval.review.model']
+					: null;
+			compactTokenThreshold = Number(config['chat.compact_token_threshold']) || 80000;
+		} catch {
+			toast.error($t('admin.failedToLoadConfig'));
+		} finally {
+			loading = false;
+		}
+	});
+
+	async function save() {
+		saving = true;
+		try {
+			await updateConfig({
+				'chat.title_generation.enabled': titleGenerationEnabled,
+				'chat.title_generation.model': titleGenerationModel,
+				'chat.context_compaction.model': contextCompactionModel,
+				'tool_approval.review.model': toolApprovalReviewModel,
+				'chat.compact_token_threshold': Math.max(10000, Number(compactTokenThreshold) || 80000)
+			});
+			toast.success($t('settings.saved'));
+		} catch {
+			toast.error($t('admin.failedToSave'));
+		} finally {
+			saving = false;
+		}
+	}
+</script>
+
+<div class="flex flex-col h-full">
+	{#if loading}
+		<div class="flex justify-center py-8"><Spinner size={16} /></div>
+	{:else}
+		<div class="flex-1 min-h-0 overflow-y-auto scrollbar-hover pr-1.5 -mr-1.5">
+			<h2 class="text-sm font-medium text-gray-900 dark:text-white mb-4">{$t('admin.chat')}</h2>
+
+			<h3 class="text-xs text-gray-400 dark:text-gray-600 mb-2">
+				{$t('admin.chatTitles')}
+			</h3>
+			<div class="flex flex-col gap-2.5">
+				<label class="flex items-center justify-between cursor-pointer">
+					<span class="text-xs text-gray-600 dark:text-gray-400">
+						{$t('admin.chatTitleGeneration')}
+					</span>
+					<ToggleSwitch
+						value={titleGenerationEnabled}
+						onchange={(value) => {
+							titleGenerationEnabled = value;
+						}}
+					/>
+				</label>
+				<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+					{$t('admin.chatTitleGenerationHint')}
+				</p>
+
+				{#if titleGenerationEnabled}
+					<div>
+						<div class="flex items-center justify-between gap-3">
+							<span class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+								{$t('admin.chatTitleModel')}
+							</span>
+							<div class="shrink-0">
+								<ModelSelector
+									bind:selectedModel={titleGenerationModel}
+									nullable
+									nullLabel={$t('modelSelector.currentModel')}
+									preferAbove={false}
+								/>
+							</div>
+						</div>
+						<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+							{$t('admin.chatTitleModelHint')}
+						</p>
+					</div>
+				{/if}
+			</div>
+
+			<h3 class="text-xs text-gray-400 dark:text-gray-600 mb-2 mt-5">
+				{$t('admin.contextCompaction')}
+			</h3>
+			<div class="flex flex-col gap-2.5">
+				<div>
+					<div class="flex items-center justify-between gap-3">
+						<span class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+							{$t('admin.contextSummaryModel')}
+						</span>
+						<div class="shrink-0">
+							<ModelSelector
+								bind:selectedModel={contextCompactionModel}
+								nullable
+								nullLabel={$t('modelSelector.currentModel')}
+								preferAbove={false}
+							/>
+						</div>
+					</div>
+					<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+						{$t('admin.contextSummaryModelHint')}
+					</p>
+				</div>
+				<div>
+					<label class="text-xs text-gray-600 dark:text-gray-400" for="compact-threshold">
+						{$t('admin.compactTokenThreshold')}
+					</label>
+					<div class="flex items-center gap-1.5 mt-1">
+						<input
+							id="compact-threshold"
+							type="number"
+							bind:value={compactTokenThreshold}
+							min="10000"
+							max="1000000"
+							step="10000"
+							class="w-24 h-7 px-2 rounded-lg text-xs bg-gray-100 dark:bg-white/6 text-gray-700 dark:text-gray-300 border border-gray-200 dark:border-white/8 outline-none focus:border-blue-400 dark:focus:border-blue-500 transition-colors"
+						/>
+						<span class="text-[0.6875rem] text-gray-400 dark:text-gray-600">
+							{$t('admin.compactTokenThresholdUnit')}
+						</span>
+					</div>
+					<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 mt-0.5">
+						{$t('admin.compactTokenThresholdHint')}
+					</p>
+				</div>
+			</div>
+
+			<h3 class="text-xs text-gray-400 dark:text-gray-600 mb-2 mt-5">
+				{$t('admin.toolApproval')}
+			</h3>
+			<div>
+				<div class="flex items-center justify-between gap-3">
+					<span class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+						{$t('admin.toolApprovalReviewModel')}
+					</span>
+					<div class="shrink-0">
+						<ModelSelector
+							bind:selectedModel={toolApprovalReviewModel}
+							nullable
+							nullLabel={$t('modelSelector.currentModel')}
+							preferAbove={false}
+						/>
+					</div>
+				</div>
+				<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+					{$t('admin.toolApprovalReviewModelHint')}
+				</p>
+			</div>
+		</div>
+
+		<div class="shrink-0 pt-3 flex justify-end">
+			<button
+				class="text-[0.8125rem] text-gray-600 dark:text-gray-400 hover:text-gray-900 dark:hover:text-white transition-colors duration-100 disabled:opacity-50"
+				disabled={saving}
+				onclick={save}
+			>
+				{saving ? $t('settings.saving') : $t('settings.save')}
+			</button>
+		</div>
+	{/if}
+</div>
diff --git a/cptr/frontend/src/lib/components/Admin/Git.svelte b/cptr/frontend/src/lib/components/Admin/Git.svelte
new file mode 100644
index 00000000..3f3b954e
--- /dev/null
+++ b/cptr/frontend/src/lib/components/Admin/Git.svelte
@@ -0,0 +1,77 @@
+<script lang="ts">
+	import { onMount } from 'svelte';
+	import { toast } from 'svelte-sonner';
+	import { getAdminConfig, updateConfig } from '$lib/apis/admin';
+	import { t } from '$lib/i18n';
+	import Spinner from '$lib/components/common/Spinner.svelte';
+	import ModelSelector from '$lib/components/common/ModelSelector.svelte';
+
+	let loading = $state(true);
+	let saving = $state(false);
+	let commitMessageModel = $state<string | null>(null);
+
+	onMount(async () => {
+		try {
+			const config = await getAdminConfig();
+			commitMessageModel =
+				typeof config['git.commit_message_generation.model'] === 'string'
+					? config['git.commit_message_generation.model']
+					: null;
+		} catch {
+			toast.error($t('admin.failedToLoadConfig'));
+		} finally {
+			loading = false;
+		}
+	});
+
+	async function save() {
+		saving = true;
+		try {
+			await updateConfig({ 'git.commit_message_generation.model': commitMessageModel });
+			toast.success($t('settings.saved'));
+		} catch {
+			toast.error($t('admin.failedToSave'));
+		} finally {
+			saving = false;
+		}
+	}
+</script>
+
+<div class="flex flex-col h-full">
+	{#if loading}
+		<div class="flex justify-center py-8"><Spinner size={16} /></div>
+	{:else}
+		<div class="flex-1 min-h-0 overflow-y-auto scrollbar-hover pr-1.5 -mr-1.5">
+			<h2 class="text-sm font-medium text-gray-900 dark:text-white mb-4">{$t('admin.git')}</h2>
+
+			<div>
+				<div class="flex items-center justify-between gap-3">
+					<span class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+						{$t('admin.gitCommitMessageModel')}
+					</span>
+					<div class="shrink-0">
+						<ModelSelector
+							bind:selectedModel={commitMessageModel}
+							nullable
+							nullLabel={$t('modelSelector.defaultModel')}
+							preferAbove={false}
+						/>
+					</div>
+				</div>
+				<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+					{$t('admin.gitCommitMessageModelHint')}
+				</p>
+			</div>
+		</div>
+
+		<div class="shrink-0 pt-3 flex justify-end">
+			<button
+				class="text-[0.8125rem] text-gray-600 dark:text-gray-400 hover:text-gray-900 dark:hover:text-white transition-colors duration-100 disabled:opacity-50"
+				disabled={saving}
+				onclick={save}
+			>
+				{saving ? $t('settings.saving') : $t('settings.save')}
+			</button>
+		</div>
+	{/if}
+</div>
diff --git a/cptr/frontend/src/lib/components/Admin/Models.svelte b/cptr/frontend/src/lib/components/Admin/Models.svelte
index 45cba5b1..0b1a6c9d 100644
--- a/cptr/frontend/src/lib/components/Admin/Models.svelte
+++ b/cptr/frontend/src/lib/components/Admin/Models.svelte
@@ -48,10 +48,6 @@
 	// Default model
 	let defaultModelId = $state('');
 
-	// Context compaction
-	let compactTokenThreshold = $state(80000);
-	let compactDirty = $state(false);
-
 	const TEMPLATE_VARIABLES = [
 		{ name: 'CPTR_CONTEXT', desc: 'Runtime, machine, workspace, and tool context' },
 		{ name: 'RUNTIME_ENV', desc: 'Runtime environment (host or container)' },
@@ -117,7 +113,7 @@ Workspace: {{WORKSPACE_NAME}}
 Files:
 {{FILE_TREE}}`;
 
-	let hasDirty = $derived(globalDirty || compactDirty || models.some((m) => m.dirty));
+	let hasDirty = $derived(globalDirty || models.some((m) => m.dirty));
 
 	function parseRows(config: ModelConfigEntry | undefined): ParamRow[] {
 		const rp = config?.params?.request_params;
@@ -256,10 +252,9 @@ Files:
 	onMount(async () => {
 		await loadModelConfig();
 
-		// Load admin config (default model, context compaction)
+		// Load default model
 		try {
 			const adminCfg = await getAdminConfig();
-			compactTokenThreshold = Number(adminCfg['chat.compact_token_threshold']) || 80000;
 			defaultModelId =
 				typeof adminCfg['chat.default_model'] === 'string' ? adminCfg['chat.default_model'] : '';
 		} catch {}
@@ -349,12 +344,7 @@ Files:
 			globalDirty = false;
 			models.forEach((m) => (m.dirty = false));
 
-			// Save default model
-			await updateConfig({
-				'chat.compact_token_threshold': compactTokenThreshold,
-				'chat.default_model': defaultModelId
-			});
-			compactDirty = false;
+			await updateConfig({ 'chat.default_model': defaultModelId });
 
 			toast.success($t('settings.saved'));
 		} catch {
@@ -552,44 +542,18 @@ Files:
 				</button>
 			</div>
 
-			<!-- Default model -->
-			<h3 class="text-xs text-gray-400 dark:text-gray-600 mb-2">
-				{$t('models.defaultModel')}
-			</h3>
 			<div class="mb-1">
-				<ModelSelector bind:selectedModel={defaultModelId} preferAbove={false} align="start" />
-			</div>
-			<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 mb-5">
-				{$t('models.defaultModelHint')}
-			</p>
-
-			<!-- Context compaction -->
-			<h3 class="text-xs text-gray-400 dark:text-gray-600 mb-2">{$t('admin.contextCompaction')}</h3>
-
-			<div class="flex flex-col gap-2.5 mb-5">
-				<div>
-					<label class="text-xs text-gray-600 dark:text-gray-400" for="compact-threshold"
-						>{$t('admin.compactTokenThreshold')}</label
-					>
-					<div class="flex items-center gap-1.5 mt-1">
-						<input
-							id="compact-threshold"
-							type="number"
-							bind:value={compactTokenThreshold}
-							oninput={() => (compactDirty = true)}
-							min="10000"
-							max="1000000"
-							step="10000"
-							class="w-24 h-7 px-2 rounded-lg text-xs bg-gray-100 dark:bg-white/6 text-gray-700 dark:text-gray-300 border border-gray-200 dark:border-white/8 outline-none focus:border-blue-400 dark:focus:border-blue-500 transition-colors"
-						/>
-						<span class="text-[0.6875rem] text-gray-400 dark:text-gray-600"
-							>{$t('admin.compactTokenThresholdUnit')}</span
-						>
+				<div class="flex items-center justify-between gap-3">
+					<h3 class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+						{$t('models.defaultModel')}
+					</h3>
+					<div class="shrink-0">
+						<ModelSelector bind:selectedModel={defaultModelId} preferAbove={false} />
 					</div>
-					<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 mt-0.5">
-						{$t('admin.compactTokenThresholdHint')}
-					</p>
 				</div>
+				<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+					{$t('models.defaultModelHint')}
+				</p>
 			</div>
 
 			<!-- Global defaults -->
diff --git a/cptr/frontend/src/lib/components/Admin/Skills.svelte b/cptr/frontend/src/lib/components/Admin/Skills.svelte
index f6d9600a..742d5d90 100644
--- a/cptr/frontend/src/lib/components/Admin/Skills.svelte
+++ b/cptr/frontend/src/lib/components/Admin/Skills.svelte
@@ -5,12 +5,14 @@
 	import { t } from '$lib/i18n';
 	import ToggleSwitch from '$lib/components/common/ToggleSwitch.svelte';
 	import Spinner from '$lib/components/common/Spinner.svelte';
+	import ModelSelector from '$lib/components/common/ModelSelector.svelte';
 
 	let loading = $state(true);
 	let saving = $state(false);
 	let enabled = $state(true);
 	let toolEnabled = $state(true);
 	let backgroundReview = $state(true);
+	let backgroundReviewModel = $state<string | null>(null);
 	let reviewInterval = $state(10);
 
 	onMount(async () => {
@@ -22,6 +24,10 @@
 			backgroundReview =
 				config['skills.background_review_enabled'] !== false &&
 				config['skills.background_review_enabled'] !== 'false';
+			backgroundReviewModel =
+				typeof config['skills.background_review.model'] === 'string'
+					? config['skills.background_review.model']
+					: null;
 			reviewInterval = Number(config['skills.review_interval_turns']) || 10;
 		} catch {
 			toast.error($t('admin.failedToLoadConfig'));
@@ -36,6 +42,7 @@
 				'skills.enabled': enabled,
 				'skills.tool_enabled': toolEnabled,
 				'skills.background_review_enabled': backgroundReview,
+				'skills.background_review.model': backgroundReviewModel,
 				'skills.review_interval_turns': Math.max(1, Number(reviewInterval) || 10)
 			});
 			toast.success($t('settings.saved'));
@@ -94,6 +101,25 @@
 							}}
 						/>
 					</label>
+
+					<div>
+						<div class="flex items-center justify-between gap-3">
+							<span class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+								{$t('admin.skillsBackgroundReviewModel')}
+							</span>
+							<div class="shrink-0">
+								<ModelSelector
+									bind:selectedModel={backgroundReviewModel}
+									nullable
+									nullLabel={$t('modelSelector.currentModel')}
+									preferAbove={false}
+								/>
+							</div>
+						</div>
+						<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+							{$t('admin.skillsBackgroundReviewModelHint')}
+						</p>
+					</div>
 				{/if}
 			</div>
 
diff --git a/cptr/frontend/src/lib/components/GitBar.svelte b/cptr/frontend/src/lib/components/GitBar.svelte
index 16c58721..d4fe2c8e 100644
--- a/cptr/frontend/src/lib/components/GitBar.svelte
+++ b/cptr/frontend/src/lib/components/GitBar.svelte
@@ -1,7 +1,6 @@
 <script lang="ts">
 	import { goto } from '$app/navigation';
 	import { activeWorkspace, addWorkspace, openFileTab } from '$lib/stores';
-	import { defaultModel } from '$lib/stores/chat';
 	import {
 		getGitLog,
 		getGitDiff,
@@ -383,7 +382,7 @@
 		if (!stagedFiles.length || generatingCommitMessage) return;
 		generatingCommitMessage = true;
 		try {
-			const message = await generateGitCommitMessage(workspacePath, $defaultModel || undefined);
+			const message = await generateGitCommitMessage(workspacePath);
 			commitSummary = message.summary;
 			commitDescription = message.description;
 		} catch (e) {
@@ -1499,8 +1498,8 @@
 									{#each fileDiff as df}
 										{#if fileDiff.length > 1}
 											<div
-											class="px-2 py-1 text-[0.625rem] text-gray-500 dark:text-gray-400 border-b border-gray-100 dark:border-white/4 sticky top-0 z-10 font-medium"
-											style="background: var(--app-bg); border-color: var(--app-border);"
+												class="px-2 py-1 text-[0.625rem] text-gray-500 dark:text-gray-400 border-b border-gray-100 dark:border-white/4 sticky top-0 z-10 font-medium"
+												style="background: var(--app-bg); border-color: var(--app-border);"
 											>
 												{df.path}
 											</div>
diff --git a/cptr/frontend/src/lib/components/Settings/Memory.svelte b/cptr/frontend/src/lib/components/Settings/Memory.svelte
index 71ca6891..67cb3aa9 100644
--- a/cptr/frontend/src/lib/components/Settings/Memory.svelte
+++ b/cptr/frontend/src/lib/components/Settings/Memory.svelte
@@ -3,6 +3,7 @@
 	import { toast } from 'svelte-sonner';
 	import ToggleSwitch from '$lib/components/common/ToggleSwitch.svelte';
 	import Spinner from '$lib/components/common/Spinner.svelte';
+	import ModelSelector from '$lib/components/common/ModelSelector.svelte';
 	import { t } from '$lib/i18n';
 	import { currentWorkspace } from '$lib/stores';
 	import {
@@ -101,6 +102,27 @@
 							onchange={(v) => (settings = { ...settings!, background_review_enabled: v })}
 						/>
 					</label>
+
+					<div>
+						<div class="flex items-center justify-between gap-3">
+							<span class="min-w-0 text-xs text-gray-600 dark:text-gray-400">
+								{$t('memory.backgroundReviewModel')}
+							</span>
+							<div class="shrink-0">
+								<ModelSelector
+									selectedModel={settings['background_review.model'] ?? null}
+									nullable
+									nullLabel={$t('modelSelector.currentModel')}
+									preferAbove={false}
+									onchange={(model) =>
+										(settings = { ...settings!, 'background_review.model': model })}
+								/>
+							</div>
+						</div>
+						<p class="text-[0.6875rem] text-gray-400 dark:text-gray-600 -mt-1">
+							{$t('memory.backgroundReviewModelHint')}
+						</p>
+					</div>
 				{/if}
 			</div>
 
diff --git a/cptr/frontend/src/lib/components/SettingsModal.svelte b/cptr/frontend/src/lib/components/SettingsModal.svelte
index 852349bd..07280d01 100644
--- a/cptr/frontend/src/lib/components/SettingsModal.svelte
+++ b/cptr/frontend/src/lib/components/SettingsModal.svelte
@@ -13,6 +13,8 @@
 	import Connections from './Admin/Connections.svelte';
 	import Agents from './Admin/Agents.svelte';
 	import Models from './Admin/Models.svelte';
+	import Chat from './Admin/Chat.svelte';
+	import Git from './Admin/Git.svelte';
 	import Skills from './Admin/Skills.svelte';
 	import Messaging from './Admin/Messaging.svelte';
 	import Gateway from './Admin/Gateway.svelte';
@@ -37,6 +39,8 @@
 		| 'connections'
 		| 'agents'
 		| 'models'
+		| 'chat'
+		| 'git'
 		| 'skills'
 		| 'messaging'
 		| 'gateway'
@@ -66,6 +70,8 @@
 		'connections',
 		'agents',
 		'models',
+		'chat',
+		'git',
 		'messaging',
 		'gateway',
 		'audio',
@@ -95,6 +101,8 @@
 		{ id: 'connections', label: $t('admin.connections'), icon: 'plug' },
 		{ id: 'agents', label: $t('admin.agents'), icon: 'terminal' },
 		{ id: 'models', label: $t('admin.models'), icon: 'cube' },
+		{ id: 'chat', label: $t('admin.chat'), icon: 'chat-bubble' },
+		{ id: 'git', label: $t('admin.git'), icon: 'git-branch' },
 		{ id: 'messaging', label: $t('admin.messaging'), icon: 'chat-bubble' },
 		{ id: 'gateway', label: $t('admin.gateway.tab'), icon: 'gateway' },
 		{ id: 'audio', label: $t('admin.audio.title'), icon: 'microphone' },
@@ -206,6 +214,10 @@
 			<Agents />
 		{:else if activeTab === 'models'}
 			<Models />
+		{:else if activeTab === 'chat'}
+			<Chat />
+		{:else if activeTab === 'git'}
+			<Git />
 		{:else if activeTab === 'skills'}
 			<Skills />
 		{:else if activeTab === 'messaging'}
diff --git a/cptr/frontend/src/lib/components/common/ModelSelector.svelte b/cptr/frontend/src/lib/components/common/ModelSelector.svelte
index f196ae00..e6eb0a3c 100644
--- a/cptr/frontend/src/lib/components/common/ModelSelector.svelte
+++ b/cptr/frontend/src/lib/components/common/ModelSelector.svelte
@@ -5,15 +5,19 @@
 	import { t } from '$lib/i18n';
 
 	interface Props {
-		selectedModel: string;
+		selectedModel: string | null;
 		preferAbove?: boolean;
 		align?: 'start' | 'end';
-		onchange?: (model: string) => void;
+		nullable?: boolean;
+		nullLabel?: string;
+		onchange?: (model: string | null) => void;
 	}
 	let {
 		selectedModel = $bindable(),
 		preferAbove = true,
 		align = 'end',
+		nullable = false,
+		nullLabel = 'Current model',
 		onchange
 	}: Props = $props();
 
@@ -31,8 +35,22 @@
 			: $chatModels
 	);
 
-	const menuItems = $derived(
-		filtered.map((m) => ({
+	const menuItems = $derived([
+		...(nullable
+			? [
+					{
+						label: nullLabel,
+						tooltip: nullLabel,
+						active: selectedModel === null || selectedModel === '',
+						check: true,
+						onclick: () => {
+							selectedModel = null;
+							onchange?.(null);
+						}
+					}
+				]
+			: []),
+		...filtered.map((m) => ({
 			label: m.name,
 			tooltip: m.name,
 			active: m.id === selectedModel,
@@ -42,7 +60,7 @@
 				onchange?.(m.id);
 			}
 		}))
-	);
+	]);
 
 	function updateViewportSize() {
 		isSmallViewport = (window.visualViewport?.width ?? window.innerWidth) < 640;
@@ -59,7 +77,7 @@
 	});
 
 	async function toggle() {
-		if ($chatModels.length === 0) return;
+		if ($chatModels.length === 0 && !nullable) return;
 		if (open) {
 			open = false;
 			return;
@@ -80,12 +98,14 @@
 		onclick={toggle}
 	>
 		<span class="truncate max-w-[10rem]"
-			>{$chatModels.length === 0
-				? $t('modelSelector.noModels')
-				: $chatModels.find((m) => m.id === selectedModel)?.name ||
-					$t('modelSelector.selectModel')}</span
+			>{selectedModel === null || selectedModel === ''
+				? nullLabel
+				: $chatModels.length === 0
+					? $t('modelSelector.noModels')
+					: $chatModels.find((m) => m.id === selectedModel)?.name ||
+						$t('modelSelector.selectModel')}</span
 		>
-		{#if $chatModels.length > 0}
+		{#if $chatModels.length > 0 || nullable}
 			<svg
 				class="w-3 h-3 opacity-50"
 				viewBox="0 0 24 24"
@@ -100,7 +120,7 @@
 		{/if}
 	</button>
 
-	{#if open && btnEl && $chatModels.length > 0}
+	{#if open && btnEl && ($chatModels.length > 0 || nullable)}
 		<DropdownMenu
 			items={menuItems}
 			anchor={btnEl}
diff --git a/cptr/frontend/src/lib/i18n/locales/en.json b/cptr/frontend/src/lib/i18n/locales/en.json
index 389b4d27..c8794d55 100644
--- a/cptr/frontend/src/lib/i18n/locales/en.json
+++ b/cptr/frontend/src/lib/i18n/locales/en.json
@@ -159,6 +159,8 @@
 	"memory.enable": "Enable memory",
 	"memory.assistantCanSave": "Assistant can save memories",
 	"memory.backgroundReview": "Background review",
+	"memory.backgroundReviewModel": "Background review model",
+	"memory.backgroundReviewModelHint": "Uses the current chat model when unset.",
 	"memory.limits": "Limits",
 	"memory.reviewEvery": "Review every",
 	"memory.userLimit": "User limit",
@@ -463,6 +465,20 @@
 	"admin.browserBaseUrl": "Base URL",
 	"admin.browserFirecrawlBaseUrlHint": "Change for self-hosted Firecrawl instances",
 	"admin.models": "Models",
+	"admin.chat": "Chat",
+	"admin.chatTitles": "Titles",
+	"admin.chatTitleGeneration": "Chat title generation",
+	"admin.chatTitleGenerationHint": "Name new chats after the first response.",
+	"admin.chatTitleModel": "Chat title model",
+	"admin.chatTitleModelHint": "Uses the current chat model when unset.",
+	"admin.contextSummaryModel": "Context summary model",
+	"admin.contextSummaryModelHint": "Summarizes older messages during compaction.",
+	"admin.toolApproval": "Tool approval",
+	"admin.toolApprovalReviewModel": "Review model",
+	"admin.toolApprovalReviewModelHint": "Reviews pending tool calls in Auto mode.",
+	"admin.git": "Git",
+	"admin.gitCommitMessageModel": "Commit message model",
+	"admin.gitCommitMessageModelHint": "Drafts messages from staged changes.",
 	"models.defaultModel": "Default model",
 	"models.defaultModelNone": "None (use first available)",
 	"models.defaultModelHint": "Pre-selected model for new chats and gateway requests.",
@@ -555,6 +571,8 @@
 	"changelog.done": "Done",
 	"modelSelector.noModels": "No models available",
 	"modelSelector.selectModel": "Select model",
+	"modelSelector.currentModel": "Current model",
+	"modelSelector.defaultModel": "Default model",
 	"modelSelector.noMatches": "No matches",
 	"modelSelector.search": "Search...",
 	"about.licenseName": "Open Use License",
@@ -1199,6 +1217,8 @@
 	"admin.skillsEnable": "Enable skills",
 	"admin.skillsAssistantCanManage": "Assistant can manage skills",
 	"admin.skillsBackgroundReview": "Background review",
+	"admin.skillsBackgroundReviewModel": "Background review model",
+	"admin.skillsBackgroundReviewModelHint": "Uses the current chat model when unset.",
 	"admin.skillsLimits": "Limits",
 	"admin.skillsReviewEvery": "Review every",
 	"admin.chars": "chars",
diff --git a/cptr/routers/chat.py b/cptr/routers/chat.py
index a4cf1e8e..bf0ce828 100644
--- a/cptr/routers/chat.py
+++ b/cptr/routers/chat.py
@@ -910,23 +910,11 @@ async def compact_chat(chat_id: str, body: CompactRequest, request: Request):
         usage = await _get_chat_context_usage(chat, model_id)
         return {"ok": True, "compacted": False, "reason": "too_short", "context_usage": usage}
 
-    connection = target.connection
-    runtime_model = target.runtime_model
-    provider = connection["provider"]
-    api_key = decrypt_key(connection.get("api_key", ""), _get_jwt_secret())
-    from cptr.utils.chat_task import _default_base_url
-
-    base_url = connection.get("base_url") or _default_base_url(provider)
-    api_type = connection.get("api_type", "chat_completions")
-
     summary = await summarize_messages(
         compacted_messages,
         existing_summary,
-        provider,
-        base_url,
-        api_key,
-        runtime_model,
-        api_type=api_type,
+        target.connection,
+        target.runtime_model,
     )
     checkpoint_message_id = _summary_checkpoint_message_id(keep_zone, message_id)
     await ChatMessage.update(checkpoint_message_id, chat_summary=summary)
diff --git a/cptr/routers/git.py b/cptr/routers/git.py
index 23d2b716..397f85c6 100644
--- a/cptr/routers/git.py
+++ b/cptr/routers/git.py
@@ -11,6 +11,7 @@
 from fastapi import APIRouter, HTTPException, Request
 from pydantic import BaseModel
 
+from cptr.utils.ai import generate_text
 from cptr.utils.json_parser import extract_json
 from cptr.utils.git import (
     GitError,
@@ -274,34 +275,24 @@ async def generate_commit_message(body: CommitMessageRequest, request: Request):
     if not patch:
         raise HTTPException(status_code=400, detail="No staged changes")
 
-    from cptr.utils.ai import chat_completion
-    from cptr.utils.chat_task import _default_base_url
-    from cptr.utils.config import _get_jwt_secret
-    from cptr.utils.crypto import decrypt_key
-    from cptr.utils.model_targets import ApiModelTarget, resolve_model_target
     from cptr.models import Config
 
-    model_id = (body.model_id or await Config.get("chat.default_model") or "").strip()
-    if not model_id:
-        raise HTTPException(status_code=400, detail="No default chat model configured")
-    target = await resolve_model_target(model_id, request.app.state)
-    if not isinstance(target, ApiModelTarget):
-        raise HTTPException(status_code=400, detail="Commit messages require an API model")
-    connection = target.connection
-    base_url = (connection.get("base_url") or _default_base_url(connection["provider"])).rstrip("/")
-    try:
-        text = await chat_completion(
-            provider=connection["provider"],
-            base_url=base_url,
-            api_key=decrypt_key(connection.get("api_key", ""), _get_jwt_secret()),
-            model=target.runtime_model,
-            messages=[{"role": "user", "content": patch}],
-            system=COMMIT_MESSAGE_PROMPT,
-            max_tokens=180,
-            api_type=connection.get("api_type", "chat_completions"),
-        )
-    except Exception as e:
-        raise HTTPException(status_code=502, detail="Could not generate a commit message") from e
+    message_model = body.model_id.strip() if isinstance(body.model_id, str) else None
+    if message_model is None:
+        configured_model = await Config.get("git.commit_message_generation.model")
+        message_model = configured_model.strip() if isinstance(configured_model, str) else None
+    if not message_model:
+        default_model = await Config.get("chat.default_model")
+        message_model = default_model.strip() if isinstance(default_model, str) else None
+    text = await generate_text(
+        request,
+        model_id=message_model,
+        messages=[{"role": "user", "content": patch}],
+        system=COMMIT_MESSAGE_PROMPT,
+        max_tokens=180,
+    )
+    if text is None:
+        raise HTTPException(status_code=502, detail="Could not generate a commit message")
     summary, description = _parse_commit_message_response(text)
     if not summary:
         raise HTTPException(status_code=502, detail="Could not generate a commit message")
diff --git a/cptr/utils/ai.py b/cptr/utils/ai.py
index aff87826..f40e69b4 100644
--- a/cptr/utils/ai.py
+++ b/cptr/utils/ai.py
@@ -3,8 +3,9 @@
 Each streaming function takes a ChatCompletionForm + url + key.
 All yield normalized events: text_delta, output, tool_call, usage, done.
 
-chat_completion() provides a simple non-streaming call for lightweight tasks
-(title generation, summarization, etc.).
+chat_completion() is the low-level non-streaming provider call.
+generate_text()/generate_json() add model-id resolution and inheritance for
+small internal calls.
 """
 
 from __future__ import annotations
@@ -223,6 +224,109 @@ async def chat_completion(
     return data.get("choices", [{}])[0].get("message", {}).get("content", "")
 
 
+async def generate_text(
+    request=None,
+    *,
+    model_id: str | None,
+    messages: list[dict],
+    system: str = "",
+    max_tokens: int = 100,
+    active_connection: dict | None = None,
+    active_model: str = "",
+    request_params: dict | None = None,
+    timeout_seconds: float = 20,
+) -> str | None:
+    """Generate text; model_id None inherits active/default."""
+    from cptr.models import Config
+    from cptr.utils.config import _get_jwt_secret
+    from cptr.utils.crypto import decrypt_key
+    from cptr.utils.model_targets import first_api_model_target, resolve_api_model_target
+
+    app_state = getattr(getattr(request, "app", None), "state", None)
+    connection = active_connection
+    runtime_model = active_model
+    selected = model_id.strip() if isinstance(model_id, str) else ""
+    if selected:
+        try:
+            target = await resolve_api_model_target(selected, app_state)
+            connection = target.connection
+            runtime_model = target.runtime_model
+        except Exception:
+            logger.debug("[generate_text] model not available: %s", selected, exc_info=True)
+
+    if not connection or not runtime_model:
+        default_model = await Config.get("chat.default_model")
+        target = None
+        if isinstance(default_model, str) and default_model.strip():
+            try:
+                target = await resolve_api_model_target(default_model.strip(), app_state)
+            except Exception:
+                logger.debug("[generate_text] default model not available", exc_info=True)
+        if target is None:
+            try:
+                target = await first_api_model_target(app_state)
+            except Exception:
+                logger.debug("[generate_text] no fallback API model available", exc_info=True)
+                return None
+        connection = target.connection
+        runtime_model = target.runtime_model
+
+    provider = connection["provider"]
+    base_url = connection.get("base_url") or {
+        "anthropic": "https://api.anthropic.com/v1",
+        "openai": "https://api.openai.com/v1",
+    }.get(provider, "https://api.openai.com/v1")
+    try:
+        return await asyncio.wait_for(
+            chat_completion(
+                provider=provider,
+                base_url=base_url,
+                api_key=decrypt_key(connection.get("api_key", ""), _get_jwt_secret()),
+                model=runtime_model,
+                messages=messages,
+                system=system,
+                max_tokens=max_tokens,
+                api_type=connection.get("api_type", "chat_completions"),
+                request_params=request_params,
+            ),
+            timeout=timeout_seconds,
+        )
+    except Exception:
+        logger.debug("[generate_text] generation failed", exc_info=True)
+        return None
+
+
+async def generate_json(
+    request=None,
+    *,
+    model_id: str | None,
+    messages: list[dict],
+    system: str = "",
+    max_tokens: int = 100,
+    active_connection: dict | None = None,
+    active_model: str = "",
+    request_params: dict | None = None,
+    timeout_seconds: float = 20,
+) -> dict | None:
+    from cptr.utils.json_parser import extract_json
+
+    text = await generate_text(
+        request,
+        model_id=model_id,
+        messages=messages,
+        system=system,
+        max_tokens=max_tokens,
+        active_connection=active_connection,
+        active_model=active_model,
+        request_params=request_params,
+        timeout_seconds=timeout_seconds,
+    )
+    if not text:
+        return None
+    parsed = extract_json(text)
+    return parsed if isinstance(parsed, dict) else None
+
+
 # ── Anthropic ────────────────────────────────────────────────
 
 
@@ -409,7 +513,11 @@ async def stream_anthropic(
                             yield {"type": "done"}
             return
         except _STREAM_RETRY_ERRORS as exc:
-            if emitted or attempt == _STREAM_RETRY_ATTEMPTS - 1 or not _is_retryable_stream_error(exc):
+            if (
+                emitted
+                or attempt == _STREAM_RETRY_ATTEMPTS - 1
+                or not _is_retryable_stream_error(exc)
+            ):
                 raise
             logger.warning(
                 "[stream] anthropic transient stream failure before first event; retrying (%s/%s)",
@@ -482,8 +590,7 @@ def _to_openai_messages(
             new_m.pop("reasoning_items", None)
             if new_m.get("tool_calls"):
                 new_m["tool_calls"] = [
-                    {k: v for k, v in tc.items() if k != "fc_id"}
-                    for tc in new_m["tool_calls"]
+                    {k: v for k, v in tc.items() if k != "fc_id"} for tc in new_m["tool_calls"]
                 ]
             new_m["content"] = formatted_content
             ri = m.get("reasoning_items")
@@ -689,7 +796,11 @@ def complete_reasoning_item() -> dict | None:
                     yield {"type": "done"}
             return
         except _STREAM_RETRY_ERRORS as exc:
-            if emitted or attempt == _STREAM_RETRY_ATTEMPTS - 1 or not _is_retryable_stream_error(exc):
+            if (
+                emitted
+                or attempt == _STREAM_RETRY_ATTEMPTS - 1
+                or not _is_retryable_stream_error(exc)
+            ):
                 raise
             logger.warning(
                 "[stream] openai completions transient stream failure before first event; retrying (%s/%s)",
@@ -942,7 +1053,9 @@ def get_reasoning_item(event: dict) -> dict:
                         elif etype == "response.reasoning_summary_text.delta":
                             item = get_reasoning_item(event)
                             delta = event.get("delta", "")
-                            summary = item.setdefault("summary", [{"type": "summary_text", "text": ""}])
+                            summary = item.setdefault(
+                                "summary", [{"type": "summary_text", "text": ""}]
+                            )
                             if not summary:
                                 summary.append({"type": "summary_text", "text": ""})
                             if isinstance(summary[0], dict):
@@ -974,10 +1087,9 @@ def get_reasoning_item(event: dict) -> dict:
                                 }
                             elif item["type"] == "reasoning":
                                 # Reasoning items must be round-tripped for reasoning models
-                                if (
-                                    active_reasoning_item is not None
-                                    and active_reasoning_item.get("id") == item.get("id")
-                                ):
+                                if active_reasoning_item is not None and active_reasoning_item.get(
+                                    "id"
+                                ) == item.get("id"):
                                     active_reasoning_item = None
                                 emitted = True
                                 yield {
@@ -991,13 +1103,15 @@ def get_reasoning_item(event: dict) -> dict:
                             raise RuntimeError(f"Responses API error: {msg}")
 
                         elif etype == "response.completed":
-                            if (
-                                active_reasoning_item is not None
-                                and _reasoning_items_to_content([active_reasoning_item])
+                            if active_reasoning_item is not None and _reasoning_items_to_content(
+                                [active_reasoning_item]
                             ):
                                 active_reasoning_item["status"] = "completed"
                                 emitted = True
-                                yield {"type": "output", "item": copy.deepcopy(active_reasoning_item)}
+                                yield {
+                                    "type": "output",
+                                    "item": copy.deepcopy(active_reasoning_item),
+                                }
                             usage = event.get("response", {}).get("usage", {})
                             if usage:
                                 emitted = True
@@ -1006,7 +1120,11 @@ def get_reasoning_item(event: dict) -> dict:
                             yield {"type": "done"}
             return
         except _STREAM_RETRY_ERRORS as exc:
-            if emitted or attempt == _STREAM_RETRY_ATTEMPTS - 1 or not _is_retryable_stream_error(exc):
+            if (
+                emitted
+                or attempt == _STREAM_RETRY_ATTEMPTS - 1
+                or not _is_retryable_stream_error(exc)
+            ):
                 raise
             logger.warning(
                 "[stream] openai responses transient stream failure before first event; retrying (%s/%s)",
diff --git a/cptr/utils/chat_task.py b/cptr/utils/chat_task.py
index 15c580c6..1c683703 100644
--- a/cptr/utils/chat_task.py
+++ b/cptr/utils/chat_task.py
@@ -42,7 +42,7 @@
 from cptr.socket.main import emit_to_user
 from cptr.utils.ai import (
     ChatCompletionForm,
-    chat_completion,
+    generate_json,
     stream_anthropic,
     stream_openai_completions,
     stream_openai_responses,
@@ -59,7 +59,6 @@
     _fn_to_schema,
 )
 from cptr.utils.chat_export import export_chat_to_file
-from cptr.utils.json_parser import extract_json
 from cptr.utils.prompt_templates import load_system_prompt as _load_system_prompt
 from cptr.utils.agents.events import (
     AgentDone,
@@ -725,6 +724,73 @@ async def _apply_voice_mode_system_prompt(system: str, chat_params: dict) -> str
     "Do not include quotes around the JSON. Do not explain."
 )
 
+TOOL_APPROVAL_REVIEW_PROMPT = (
+    "Decide whether Computer may run the proposed tool call without asking the user. "
+    'Return ONLY JSON: {"decision":"approve" or "deny","reason":"short reason"}. '
+    "Approve when the call is a reasonable next step for the user's request and the arguments "
+    "are not destructive, surprising, credential-seeking, or unrelated. Approve harmless "
+    "read-only shell commands such as checking the current time/date, current directory, file "
+    "listings, or git status when they directly answer the request. Deny when uncertain."
+)
+
+
+async def review_tool_approval(
+    *,
+    messages: list[dict],
+    assistant_text: str,
+    connection: dict,
+    model: str,
+    tool_name: str,
+    arguments: dict,
+) -> bool:
+    """Return True when Auto mode can run a pending tool without prompting."""
+    latest_user = ""
+    for message in reversed(messages):
+        if message.get("role") == "user":
+            latest_user = _plain_message_text(message.get("content"))
+            break
+    args_text = json.dumps(arguments, ensure_ascii=False, default=str)
+    if len(args_text) > 4000:
+        args_text = args_text[:3500] + "\n...(truncated)"
+    configured_model = await Config.get("tool_approval.review.model")
+    logger.info(
+        "[tool-approval] auto review start tool=%s active_model=%s review_model=%s args=%s",
+        tool_name,
+        model,
+        configured_model or "<current>",
+        args_text[:1000],
+    )
+    parsed = await generate_json(
+        model_id=configured_model,
+        active_connection=connection,
+        active_model=model,
+        messages=[
+            {
+                "role": "user",
+                "content": (
+                    f"Latest user request:\n{latest_user[:2000]}\n\n"
+                    f"Assistant text before tool call:\n{assistant_text[-2000:]}\n\n"
+                    f"Tool: {tool_name}\nArguments:\n{args_text}"
+                ),
+            }
+        ],
+        system=TOOL_APPROVAL_REVIEW_PROMPT,
+        max_tokens=150,
+        timeout_seconds=20,
+    )
+    decision = str((parsed or {}).get("decision") or "").strip().lower()
+    reason = str((parsed or {}).get("reason") or "").strip()
+    approved = decision == "approve"
+    logger.info(
+        "[tool-approval] auto review decision tool=%s decision=%s approved=%s reason=%s raw=%s",
+        tool_name,
+        decision or "<invalid>",
+        approved,
+        reason[:500],
+        parsed,
+    )
+    return approved
+
 
 async def generate_chat_title(
     chat_id: str,
@@ -735,14 +801,13 @@ async def generate_chat_title(
 ):
     """Generate a proper title for a new chat via a lightweight LLM call.
 
-    Uses the shared chat_completion() helper from ai.py.
+    Uses the configured title model, inheriting the active/default model when unset.
     On success, updates the DB and emits a socket event so the frontend
     updates the tab label in real time.
     """
-    provider = connection["provider"]
-    api_key = decrypt_key(connection.get("api_key", ""), _get_jwt_secret())
-    base_url = connection.get("base_url") or _default_base_url(provider)
-    api_type = connection.get("api_type", "chat_completions")
+    title_generation_enabled = await Config.get("chat.title_generation.enabled")
+    if title_generation_enabled is False or title_generation_enabled == "false":
+        return
 
     # Truncate very long messages to keep the prompt small
     truncated = user_message[:200].strip()
@@ -750,19 +815,14 @@ async def generate_chat_title(
         truncated += "..."
 
     try:
-        text = await chat_completion(
-            provider=provider,
-            base_url=base_url,
-            api_key=api_key,
-            model=model,
+        parsed = await generate_json(
+            model_id=await Config.get("chat.title_generation.model"),
+            active_connection=connection,
+            active_model=model,
             messages=[{"role": "user", "content": truncated}],
             system=TITLE_PROMPT,
             max_tokens=50,
-            api_type=api_type,
         )
-
-        # Parse the JSON title
-        parsed = extract_json(text)
         title = (parsed.get("title", "") if isinstance(parsed, dict) else "").strip()
         if not title:
             return
@@ -1865,7 +1925,7 @@ async def _finish_reasoning_item():
         # Plan mode: strip write tools, inject prompt as user message (not system, to preserve cache)
         plan_mode = chat_params.get("plan_mode", False)
         if plan_mode:
-            tools = [t for t in tools if ALL_TOOLS.get(t["name"], {}).get("auto")]
+            tools = [t for t in tools if not ALL_TOOLS.get(t["name"], {}).get("ask", True)]
             tools = [t for t in tools if t["name"] not in {"delegate_task", "update_memory"}]
             tools.append(_fn_to_schema("create_artifact", create_artifact))
             tools.append(ASK_USER_SCHEMA)
@@ -1893,7 +1953,7 @@ async def _finish_reasoning_item():
 
         # Tool approval mode: 'ask' | 'auto' | 'full'
         #   ask  = require approval for ALL tools (including reads)
-        #   auto = auto-approve tools marked auto=True, ask for others
+        #   auto = run ask:false tools; review ask:true tools before prompting
         #   full = auto-approve everything
         approval_mode = chat_params.get("tool_approval_mode", "auto")
         # Legacy compat: old boolean auto_approve_tools
@@ -1924,9 +1984,26 @@ async def run_queued_tool_calls(tool_ctx: dict) -> str:
 
                 name = item.get("name", "")
                 tool = ALL_TOOLS.get(name)
-                needs_approval = not (
-                    approval_mode == "full" or (approval_mode == "auto" and tool and tool["auto"])
+                should_auto = approval_mode == "full" or (
+                    approval_mode == "auto" and tool and not tool.get("ask", True)
                 )
+                if (
+                    not should_auto
+                    and approval_mode == "auto"
+                    and not item.get("approved")
+                    and tool
+                    and await review_tool_approval(
+                        messages=messages,
+                        assistant_text=content,
+                        connection=connection,
+                        model=model,
+                        tool_name=name,
+                        arguments=item.get("arguments") or {},
+                    )
+                ):
+                    item["approved"] = True
+                    should_auto = True
+                needs_approval = not should_auto
                 if needs_approval and not item.get("approved"):
                     item["status"] = "pending"
                     await _save_message(
@@ -2056,11 +2133,8 @@ async def run_queued_tool_calls(tool_ctx: dict) -> str:
                 summary = await summarize_messages(
                     drop_zone,
                     loaded_summary,
-                    provider,
-                    base_url,
-                    api_key,
+                    connection,
                     model,
-                    api_type=api_type,
                 )
 
                 checkpoint_message_id = _summary_checkpoint_message_id(
@@ -2419,7 +2493,7 @@ async def auto_answer():
                             loaded_skill_names.add(skill_name)
                     tool = ALL_TOOLS.get(name)
                     should_auto = approval_mode == "full" or (
-                        approval_mode == "auto" and tool and tool["auto"]
+                        approval_mode == "auto" and tool and not tool.get("ask", True)
                     )
                     if not should_auto:
                         needs_approval = tc
@@ -2747,7 +2821,11 @@ async def auto_answer():
                     fallback = first_user.content[:50].strip() or "New Chat"
                     if chat_obj.title == fallback:
                         await generate_chat_title(
-                            chat_id, user_id, connection, model, first_user.content
+                            chat_id,
+                            user_id,
+                            connection,
+                            model,
+                            first_user.content,
                         )
         except Exception:
             logger.debug(
diff --git a/cptr/utils/memory.py b/cptr/utils/memory.py
index 5751adaa..b95b60d5 100644
--- a/cptr/utils/memory.py
+++ b/cptr/utils/memory.py
@@ -24,6 +24,7 @@
     "enabled": True,
     "tool_enabled": True,
     "background_review_enabled": True,
+    "background_review.model": None,
     "review_interval_turns": 10,
     "user_char_limit": 2000,
     "workspace_char_limit": 3000,
@@ -279,15 +280,19 @@ def _split_markdown_sections(text: str, fallback_heading: str = "") -> list[dict
             headings.append((index, len(match.group(1)), match.group(2).strip()))
     if not headings:
         content = "\n".join(lines).strip()
-        return [
-            {
-                "heading": fallback_heading,
-                "memory_id": _extract_memory_id(content),
-                "content": content,
-                "start": 0,
-                "end": len(lines),
-            }
-        ] if content else []
+        return (
+            [
+                {
+                    "heading": fallback_heading,
+                    "memory_id": _extract_memory_id(content),
+                    "content": content,
+                    "start": 0,
+                    "end": len(lines),
+                }
+            ]
+            if content
+            else []
+        )
 
     sections: list[dict[str, Any]] = []
     for pos, (start, level, heading) in enumerate(headings):
@@ -720,7 +725,9 @@ def _operation_uses_markdown(operation: dict[str, Any]) -> bool:
     )
 
 
-def _find_section_bounds(text: str, *, memory_id: str = "", heading: str = "") -> tuple[int, int] | None:
+def _find_section_bounds(
+    text: str, *, memory_id: str = "", heading: str = ""
+) -> tuple[int, int] | None:
     sections = _split_markdown_sections(text)
     for section in sections:
         if memory_id and section.get("memory_id") == memory_id:
@@ -751,7 +758,11 @@ def _write_if_unchanged(
     current_hash = _file_hash(path)
     if current_hash != original_hash:
         backup = _backup_to_trash(path, "conflict", root)
-        return False, f"refusing to overwrite changed memory file; backup saved to {backup}", str(backup)
+        return (
+            False,
+            f"refusing to overwrite changed memory file; backup saved to {backup}",
+            str(backup),
+        )
     _atomic_write_text(path, content)
     return True, "written", ""
 
@@ -785,7 +796,9 @@ def _add_related_link(section_text: str, link: str) -> str:
     return section_text.rstrip() + f"\n\nRelated: {rendered}\n"
 
 
-def apply_markdown_memory_batch(root: MemoryRoot, operations: list[dict[str, Any]]) -> dict[str, Any]:
+def apply_markdown_memory_batch(
+    root: MemoryRoot, operations: list[dict[str, Any]]
+) -> dict[str, Any]:
     if not operations:
         return {"success": False, "error": "operations list is empty"}
 
@@ -798,7 +811,10 @@ def read_target(operation: dict[str, Any]) -> Path:
 
     def load(path: Path) -> str:
         if path not in touched:
-            touched[path] = (_file_hash(path), path.read_text(errors="replace") if path.exists() else "")
+            touched[path] = (
+                _file_hash(path),
+                path.read_text(errors="replace") if path.exists() else "",
+            )
         return touched[path][1]
 
     def store(path: Path, content: str) -> None:
@@ -849,7 +865,9 @@ def store(path: Path, content: str) -> None:
                 start, end = bounds
                 heading = str(operation.get("heading") or operation.get("title") or "Memory")
                 marker = str(operation.get("memory_id") or "").strip()
-                replacement = _append_memory_section("", {**operation, "heading": heading, "memory_id": marker}, heading)
+                replacement = _append_memory_section(
+                    "", {**operation, "heading": heading, "memory_id": marker}, heading
+                )
                 store(target, _replace_lines(existing, start, end, replacement))
             elif old_text and old_text in existing:
                 store(target, existing.replace(old_text, content, 1))
@@ -877,7 +895,9 @@ def store(path: Path, content: str) -> None:
             messages.append(f"removed memory from {_relative_to_root(target, root.root)}")
 
         elif action == "link":
-            link = str(operation.get("link") or operation.get("target") or operation.get("content") or "").strip()
+            link = str(
+                operation.get("link") or operation.get("target") or operation.get("content") or ""
+            ).strip()
             if not link:
                 return {"success": False, "error": f"{operation_name}: link is required"}
             target = read_target(operation)
@@ -891,7 +911,10 @@ def store(path: Path, content: str) -> None:
                 start, end = bounds
                 lines = existing.splitlines()
                 section_text = "\n".join(lines[start:end])
-                store(target, _replace_lines(existing, start, end, _add_related_link(section_text, link)))
+                store(
+                    target,
+                    _replace_lines(existing, start, end, _add_related_link(section_text, link)),
+                )
             else:
                 store(target, _add_related_link(existing, link))
             messages.append(f"linked memory in {_relative_to_root(target, root.root)}")
@@ -905,14 +928,23 @@ def store(path: Path, content: str) -> None:
                 return {"success": False, "error": f"{operation_name}: destination already exists"}
             destination.parent.mkdir(parents=True, exist_ok=True)
             shutil.move(str(source), str(destination))
-            messages.append(f"moved {_relative_to_root(source, root.root)} to {_relative_to_root(destination, root.root)}")
+            messages.append(
+                f"moved {_relative_to_root(source, root.root)} to {_relative_to_root(destination, root.root)}"
+            )
 
         elif action == "split":
             target = read_target(operation)
             existing = load(target)
-            sections = [section for section in _split_markdown_sections(existing, target.stem) if section.get("content")]
+            sections = [
+                section
+                for section in _split_markdown_sections(existing, target.stem)
+                if section.get("content")
+            ]
             if len(sections) < 2:
-                return {"success": False, "error": f"{operation_name}: file does not have multiple sections"}
+                return {
+                    "success": False,
+                    "error": f"{operation_name}: file does not have multiple sections",
+                }
             folder = target.with_suffix("")
             folder.mkdir(parents=True, exist_ok=True)
             _backup_to_trash(target, "split", root.root)
@@ -925,19 +957,28 @@ def store(path: Path, content: str) -> None:
                     _atomic_write_text(child, child_text)
                 map_lines.append(f"- [[{heading}]]")
             store(target, "\n".join(map_lines).rstrip() + "\n")
-            messages.append(f"split {_relative_to_root(target, root.root)} into {len(sections)} files")
+            messages.append(
+                f"split {_relative_to_root(target, root.root)} into {len(sections)} files"
+            )
 
         elif action == "merge":
-            source = _safe_relative_path(root.root, str(operation.get("source_path") or operation.get("path") or ""))
+            source = _safe_relative_path(
+                root.root, str(operation.get("source_path") or operation.get("path") or "")
+            )
             target = _safe_relative_path(root.root, str(operation.get("target_path") or ""))
             if not source.exists() or not target.exists():
-                return {"success": False, "error": f"{operation_name}: source and target must exist"}
+                return {
+                    "success": False,
+                    "error": f"{operation_name}: source and target must exist",
+                }
             source_text = source.read_text(errors="replace")
             target_text = load(target)
             _backup_to_trash(source, "merge", root.root)
             store(target, target_text.rstrip() + "\n\n" + source_text.strip() + "\n")
             source.unlink()
-            messages.append(f"merged {_relative_to_root(source, root.root)} into {_relative_to_root(target, root.root)}")
+            messages.append(
+                f"merged {_relative_to_root(source, root.root)} into {_relative_to_root(target, root.root)}"
+            )
 
         else:
             return {
@@ -950,7 +991,11 @@ def store(path: Path, content: str) -> None:
         if not success:
             return {"success": False, "error": message, "backup": backup, "path": str(path)}
 
-    return {"success": True, "message": "; ".join(messages) or "No changes.", "path": str(root.root)}
+    return {
+        "success": True,
+        "message": "; ".join(messages) or "No changes.",
+        "path": str(root.root),
+    }
 
 
 async def get_memory_settings() -> dict[str, Any]:
@@ -999,7 +1044,12 @@ def apply_memory_batch(
     next_entries = list(current_entries)
     for index, operation in enumerate(operations):
         if not isinstance(operation, dict):
-            return False, f"Operation {index + 1}: must be an object", current_entries, current_usage
+            return (
+                False,
+                f"Operation {index + 1}: must be an object",
+                current_entries,
+                current_usage,
+            )
         action = operation.get("action")
         content = normalize_memory_text(str(operation.get("content") or ""))
         old_text = normalize_memory_text(str(operation.get("old_text") or ""))
@@ -1008,21 +1058,46 @@ def apply_memory_batch(
         if action in {"add", "replace"}:
             validation_error = memory_text_error(content)
             if validation_error:
-                return False, f"{operation_name}: {validation_error}", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: {validation_error}",
+                    current_entries,
+                    current_usage,
+                )
 
         if action == "add":
             if not content:
-                return False, f"{operation_name}: content is required", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: content is required",
+                    current_entries,
+                    current_usage,
+                )
             if content not in next_entries:
                 next_entries.append(content)
         elif action == "replace":
             if not old_text:
-                return False, f"{operation_name}: old_text is required", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: old_text is required",
+                    current_entries,
+                    current_usage,
+                )
             if not content:
-                return False, f"{operation_name}: content is required", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: content is required",
+                    current_entries,
+                    current_usage,
+                )
             matches = [i for i, entry in enumerate(next_entries) if old_text in entry]
             if not matches:
-                return False, f"{operation_name}: no entry matched '{old_text}'", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: no entry matched '{old_text}'",
+                    current_entries,
+                    current_usage,
+                )
             if len({next_entries[i] for i in matches}) > 1:
                 return (
                     False,
@@ -1033,10 +1108,20 @@ def apply_memory_batch(
             next_entries[matches[0]] = content
         elif action == "remove":
             if not old_text:
-                return False, f"{operation_name}: old_text is required", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: old_text is required",
+                    current_entries,
+                    current_usage,
+                )
             matches = [i for i, entry in enumerate(next_entries) if old_text in entry]
             if not matches:
-                return False, f"{operation_name}: no entry matched '{old_text}'", current_entries, current_usage
+                return (
+                    False,
+                    f"{operation_name}: no entry matched '{old_text}'",
+                    current_entries,
+                    current_usage,
+                )
             if len({next_entries[i] for i in matches}) > 1:
                 return (
                     False,
@@ -1061,7 +1146,12 @@ def apply_memory_batch(
             current_entries,
             current_usage,
         )
-    return True, f"Applied {len(operations)} operation(s).", next_entries, f"{next_usage_count}/{character_limit}"
+    return (
+        True,
+        f"Applied {len(operations)} operation(s).",
+        next_entries,
+        f"{next_usage_count}/{character_limit}",
+    )
 
 
 async def write_memory(
@@ -1138,7 +1228,9 @@ async def read_memory_state(user_id: str, workspace: str) -> dict[str, Any]:
     if workspace:
         workspace_memory = await resolve_memory_file(user_id, workspace, "workspace")
         workspace_entries = await asyncio.to_thread(read_memory_entries, workspace_memory.path)
-        workspace_usage = f"{measure_memory_entries(workspace_entries)}/{workspace_memory.character_limit}"
+        workspace_usage = (
+            f"{measure_memory_entries(workspace_entries)}/{workspace_memory.character_limit}"
+        )
         workspace_path_value = str(workspace_memory.path)
     return {
         "settings": settings,
@@ -1175,7 +1267,9 @@ async def recall_memory_context(
     snippets_by_scope: dict[str, list[MemorySnippet]] = {"user": [], "workspace": []}
 
     for root in roots:
-        baseline_text = root.baseline_path.read_text(errors="replace") if root.baseline_path.is_file() else ""
+        baseline_text = (
+            root.baseline_path.read_text(errors="replace") if root.baseline_path.is_file() else ""
+        )
         contextual_query = query
         if baseline_text:
             baseline_links = _extract_baseline_links(baseline_text)
@@ -1192,7 +1286,8 @@ async def recall_memory_context(
             snippets_by_scope[root.scope].extend(
                 snippet
                 for snippet in snippets
-                if snippet.path != root.baseline_path.name and not PROMPT_THREAT_RE.search(snippet.snippet)
+                if snippet.path != root.baseline_path.name
+                and not PROMPT_THREAT_RE.search(snippet.snippet)
             )
     return MemoryContext(
         user=snippets_by_scope["user"],
@@ -1317,7 +1412,9 @@ def collect() -> list[dict[str, Any]]:
                         "path": rel,
                         "size": stat.st_size,
                         "modified_at": stat.st_mtime,
-                        "headings": [section["heading"] for section in sections if section.get("heading")][:12],
+                        "headings": [
+                            section["heading"] for section in sections if section.get("heading")
+                        ][:12],
                         "baseline": path == root.baseline_path,
                         "trash": _is_trash_path(_relative_to_root(path, root.root)),
                     }
@@ -1423,7 +1520,9 @@ def summarize_recent_conversation(messages: list[dict[str, Any]], assistant_repl
     return "\n\n".join(lines)
 
 
-def build_memory_review_prompt(memory_state: dict[str, Any], workspace: str, transcript: str) -> str:
+def build_memory_review_prompt(
+    memory_state: dict[str, Any], workspace: str, transcript: str
+) -> str:
     return (
         "Review the completed conversation and decide whether cptr should remember "
         "stable facts. Return ONLY JSON with this shape:\n"
@@ -1494,29 +1593,19 @@ async def run_memory_review(
     model: str,
 ) -> None:
     try:
-        from cptr.utils.ai import chat_completion
-        from cptr.utils.chat_task import _default_base_url
-        from cptr.utils.config import _get_jwt_secret
-        from cptr.utils.crypto import decrypt_key
-        from cptr.utils.json_parser import extract_json
+        from cptr.utils.ai import generate_json
 
         memory_state = await read_memory_state(user_id, workspace)
         transcript = summarize_recent_conversation(conversation_messages, assistant_reply)
         prompt = build_memory_review_prompt(memory_state, workspace, transcript)
-        provider = model_connection["provider"]
-        api_key = decrypt_key(model_connection.get("api_key", ""), _get_jwt_secret())
-        base_url = model_connection.get("base_url") or _default_base_url(provider)
-        text = await chat_completion(
-            provider=provider,
-            base_url=base_url,
-            api_key=api_key,
-            model=model,
+        parsed = await generate_json(
+            model_id=await Config.get("memory.background_review.model"),
+            active_connection=model_connection,
+            active_model=model,
             messages=[{"role": "user", "content": prompt}],
             system="You are cptr's private memory reviewer. Return only valid JSON.",
             max_tokens=700,
-            api_type=model_connection.get("api_type", "chat_completions"),
         )
-        parsed = extract_json(text)
         if not isinstance(parsed, dict):
             return
         for scope in ("user", "workspace"):
diff --git a/cptr/utils/skills.py b/cptr/utils/skills.py
index 7b6aa851..10238192 100644
--- a/cptr/utils/skills.py
+++ b/cptr/utils/skills.py
@@ -192,11 +192,15 @@ def write_usage(workspace: str, source: str, data: dict[str, Any]) -> None:
     path = _usage_path(workspace, source)
     path.parent.mkdir(parents=True, exist_ok=True)
     tmp = path.with_name(f".{path.name}.{os.getpid()}.{time.time_ns()}.tmp")
-    tmp.write_text(json.dumps(data, ensure_ascii=False, indent=2, sort_keys=True) + "\n", encoding="utf-8")
+    tmp.write_text(
+        json.dumps(data, ensure_ascii=False, indent=2, sort_keys=True) + "\n", encoding="utf-8"
+    )
     os.replace(tmp, path)
 
 
-def bump_skill_view(workspace: str, name: str, source: str = "workspace", *, used: bool = True) -> None:
+def bump_skill_view(
+    workspace: str, name: str, source: str = "workspace", *, used: bool = True
+) -> None:
     try:
         data = read_usage(workspace, source)
         skills = data.setdefault("skills", {})
@@ -595,7 +599,11 @@ def update_managed_skill(workspace: str, name: str, content: str) -> dict:
     content = _with_skill_metadata(content, preserve_created=old_fm)
     _validate_skill_content(name, content)
     _write_text_atomic(skill_md, content.strip() + "\n")
-    source = "global" if skill_dir.is_relative_to(MANAGED_GLOBAL_SKILL_DIR.expanduser().resolve()) else "workspace"
+    source = (
+        "global"
+        if skill_dir.is_relative_to(MANAGED_GLOBAL_SKILL_DIR.expanduser().resolve())
+        else "workspace"
+    )
     bump_skill_update(workspace, name, source)
     return {"success": True, "name": name, "location": str(skill_md.resolve())}
 
@@ -635,7 +643,11 @@ def write_managed_skill_file(
     if not target.is_relative_to(skill_dir):
         raise ValueError("file_path escapes the skill")
     _write_text_atomic(target, file_content)
-    source = "global" if skill_dir.is_relative_to(MANAGED_GLOBAL_SKILL_DIR.expanduser().resolve()) else "workspace"
+    source = (
+        "global"
+        if skill_dir.is_relative_to(MANAGED_GLOBAL_SKILL_DIR.expanduser().resolve())
+        else "workspace"
+    )
     bump_skill_update(workspace, name, source)
     return {"success": True, "name": name, "location": str(target)}
 
@@ -787,19 +799,19 @@ async def run_skill_review(
     skill_create_requested: bool,
 ) -> None:
     try:
-        from cptr.utils.ai import chat_completion
-        from cptr.utils.chat_task import _default_base_url
-        from cptr.utils.config import _get_jwt_secret
-        from cptr.utils.crypto import decrypt_key
-        from cptr.utils.json_parser import extract_json
+        from cptr.models import Config
+        from cptr.utils.ai import generate_json
         from cptr.utils.memory import summarize_recent_conversation
 
         skills = discover_skills(workspace)
         transcript = summarize_recent_conversation(conversation_messages, assistant_reply)
-        catalog = "\n".join(
-            f"- {skill.name} ({skill.source}, {'managed' if skill.managed else 'read-only'}): {skill.description}"
-            for skill in skills[:80]
-        ) or "- none"
+        catalog = (
+            "\n".join(
+                f"- {skill.name} ({skill.source}, {'managed' if skill.managed else 'read-only'}): {skill.description}"
+                for skill in skills[:80]
+            )
+            or "- none"
+        )
         loaded_blocks = []
         for name in sorted(loaded_skill_names):
             skill = load_skill(workspace, name)
@@ -830,20 +842,14 @@ async def run_skill_review(
             f"Loaded managed skill contents:\n{loaded_context}\n\n"
             f"Conversation:\n{transcript}"
         )
-        provider = model_connection["provider"]
-        api_key = decrypt_key(model_connection.get("api_key", ""), _get_jwt_secret())
-        base_url = model_connection.get("base_url") or _default_base_url(provider)
-        text = await chat_completion(
-            provider=provider,
-            base_url=base_url,
-            api_key=api_key,
-            model=model,
+        parsed = await generate_json(
+            model_id=await Config.get("skills.background_review.model"),
+            active_connection=model_connection,
+            active_model=model,
             messages=[{"role": "user", "content": prompt}],
             system="You are Computer's private skill reviewer. Return only valid JSON.",
             max_tokens=1800,
-            api_type=model_connection.get("api_type", "chat_completions"),
         )
-        parsed = extract_json(text)
         if not isinstance(parsed, dict):
             return
         actions = parsed.get("actions")
diff --git a/cptr/utils/summarize.py b/cptr/utils/summarize.py
index c5fb6991..e64a471c 100644
--- a/cptr/utils/summarize.py
+++ b/cptr/utils/summarize.py
@@ -27,9 +27,7 @@ def _get_summarize_prompt() -> str:
         from cptr.utils.config import load_config
 
         config = load_config()
-        return config.get("chat", {}).get(
-            "compact_summary_prompt", DEFAULT_SUMMARIZE_PROMPT
-        )
+        return config.get("chat", {}).get("compact_summary_prompt", DEFAULT_SUMMARIZE_PROMPT)
     except Exception:
         return DEFAULT_SUMMARIZE_PROMPT
 
@@ -37,19 +35,14 @@ def _get_summarize_prompt() -> str:
 async def summarize_messages(
     messages: list[dict],
     existing_summary: str | None,
-    provider: str,
-    base_url: str,
-    api_key: str,
+    connection: dict,
     model: str,
-    api_type: str = "chat_completions",
 ) -> str:
     """Summarize messages into a compact rolling summary.
 
     If existing_summary is provided, it's included so the new summary
     incorporates everything before it.
     """
-    from cptr.utils.ai import chat_completion
-
     parts: list[str] = []
     if existing_summary:
         parts.append(f"[Previous summary]\n{existing_summary}\n")
@@ -58,9 +51,7 @@ async def summarize_messages(
         role = m.get("role", "unknown")
         content = m.get("content", "")
         if isinstance(content, list):
-            content = " ".join(
-                b.get("text", "") for b in content if b.get("type") == "text"
-            )
+            content = " ".join(b.get("text", "") for b in content if b.get("type") == "text")
         # Truncate very long messages (e.g. tool outputs)
         if len(content) > 2000:
             content = content[:1000] + "\n...(truncated)...\n" + content[-500:]
@@ -72,16 +63,19 @@ async def summarize_messages(
         text = text[:15_000] + "\n...\n" + text[-10_000:]
 
     try:
-        result = await chat_completion(
-            provider=provider,
-            base_url=base_url,
-            api_key=api_key,
-            model=model,
+        from cptr.models import Config
+        from cptr.utils.ai import generate_text
+
+        result = await generate_text(
+            model_id=await Config.get("chat.context_compaction.model"),
+            active_connection=connection,
+            active_model=model,
             messages=[{"role": "user", "content": text}],
             system=_get_summarize_prompt(),
             max_tokens=1000,
-            api_type=api_type,
         )
+        if result is None:
+            raise RuntimeError("summary generation failed")
         logger.info("[summarize] LLM summary: %d chars", len(result))
         return result
     except Exception:
diff --git a/cptr/utils/tools.py b/cptr/utils/tools.py
index 0fc7debf..2e1b62c2 100644
--- a/cptr/utils/tools.py
+++ b/cptr/utils/tools.py
@@ -1429,6 +1429,7 @@ def _resolve_path(path: str, workspace: str) -> Path:
         raise ValueError(f"Path traversal rejected: {path}")
     return full
 
+
 async def create_automation(
     name: str,
     prompt: str,
@@ -1709,11 +1710,7 @@ def _read_skill_file():
                 content = target.read_text(errors="strict")
             except (UnicodeDecodeError, ValueError):
                 return f"Error: binary skill file ({target.suffix}), cannot read as text"
-            return (
-                f'<skill_file name="{skill.name}" path="{file_path}">\n'
-                f"{content}\n"
-                "</skill_file>"
-            )
+            return f'<skill_file name="{skill.name}" path="{file_path}">\n{content}\n</skill_file>'
 
         return await asyncio.to_thread(_read_skill_file)
 
@@ -2281,44 +2278,44 @@ async def notify(message: str, target: str = "", title: str = "", *, __context__
 # ── Registry ────────────────────────────────────────────────
 
 TOOLS: dict[str, dict] = {
-    # Read-only (auto-approve)
-    "read_file": {"fn": read_file, "auto": True},
-    "list_directory": {"fn": list_directory, "auto": True},
-    "search_files": {"fn": search_files, "auto": True},
-    "check_task": {"fn": check_task, "auto": True},
-    "web_search": {"fn": web_search, "auto": True},
-    "read_url": {"fn": read_url, "auto": True},
-    "search_chats": {"fn": search_chats, "auto": True},
-    "list_automations": {"fn": list_automations, "auto": True},
-    "view_skill": {"fn": view_skill, "auto": True},
-    "update_tasks": {"fn": update_tasks, "auto": True},
-    # Write / mutate (require approval unless auto_approve_all)
-    "create_file": {"fn": create_file, "auto": False},
-    "display_file": {"fn": display_file, "auto": False},
-    "edit_file": {"fn": edit_file, "auto": False},
-    "multi_edit_file": {"fn": multi_edit_file, "auto": False},
-    "write_file": {"fn": write_file, "auto": False},
-    "run_command": {"fn": run_command, "auto": False},
-    "send_input": {"fn": send_input, "auto": False},
-    "kill_task": {"fn": kill_task, "auto": False},
-    "create_automation": {"fn": create_automation, "auto": False},
-    "update_automation": {"fn": update_automation, "auto": False},
-    "toggle_automation": {"fn": toggle_automation, "auto": False},
-    "delete_automation": {"fn": delete_automation, "auto": False},
-    "notify": {"fn": notify, "auto": False},
-    "image_generate": {"fn": image_generate, "auto": False},
-    "manage_skill": {"fn": manage_skill, "auto": False},
-    "update_memory": {"fn": update_memory, "auto": True},
+    # Auto mode runs these without asking.
+    "read_file": {"fn": read_file, "ask": False},
+    "list_directory": {"fn": list_directory, "ask": False},
+    "search_files": {"fn": search_files, "ask": False},
+    "check_task": {"fn": check_task, "ask": False},
+    "web_search": {"fn": web_search, "ask": False},
+    "read_url": {"fn": read_url, "ask": False},
+    "search_chats": {"fn": search_chats, "ask": False},
+    "list_automations": {"fn": list_automations, "ask": False},
+    "view_skill": {"fn": view_skill, "ask": False},
+    "update_tasks": {"fn": update_tasks, "ask": False},
+    # Auto mode reviews these first, then asks if denied or unclear.
+    "create_file": {"fn": create_file, "ask": True},
+    "display_file": {"fn": display_file, "ask": True},
+    "edit_file": {"fn": edit_file, "ask": True},
+    "multi_edit_file": {"fn": multi_edit_file, "ask": True},
+    "write_file": {"fn": write_file, "ask": True},
+    "run_command": {"fn": run_command, "ask": True},
+    "send_input": {"fn": send_input, "ask": True},
+    "kill_task": {"fn": kill_task, "ask": True},
+    "create_automation": {"fn": create_automation, "ask": True},
+    "update_automation": {"fn": update_automation, "ask": True},
+    "toggle_automation": {"fn": toggle_automation, "ask": True},
+    "delete_automation": {"fn": delete_automation, "ask": True},
+    "notify": {"fn": notify, "ask": True},
+    "image_generate": {"fn": image_generate, "ask": True},
+    "manage_skill": {"fn": manage_skill, "ask": True},
+    "update_memory": {"fn": update_memory, "ask": False},
 }
 
 # Browser tools — conditionally included in schemas based on browser.enabled
 BROWSER_TOOLS: dict[str, dict] = {
-    "browser_navigate": {"fn": browser_navigate, "auto": False},
-    "browser_snapshot": {"fn": browser_snapshot, "auto": True},
-    "browser_click": {"fn": browser_click, "auto": False},
-    "browser_type": {"fn": browser_type, "auto": False},
-    "browser_screenshot": {"fn": browser_screenshot, "auto": True},
-    "browser_evaluate": {"fn": browser_evaluate, "auto": False},
+    "browser_navigate": {"fn": browser_navigate, "ask": True},
+    "browser_snapshot": {"fn": browser_snapshot, "ask": False},
+    "browser_click": {"fn": browser_click, "ask": True},
+    "browser_type": {"fn": browser_type, "ask": True},
+    "browser_screenshot": {"fn": browser_screenshot, "ask": False},
+    "browser_evaluate": {"fn": browser_evaluate, "ask": True},
 }
 
 
@@ -2671,13 +2668,14 @@ async def _run_subagent_chat(
 
 
 SUBAGENT_TOOLS: dict[str, dict] = {
-    "delegate_task": {"fn": delegate_task, "auto": True},
-    "timer": {"fn": timer, "auto": False},
+    "delegate_task": {"fn": delegate_task, "ask": False},
+    "timer": {"fn": timer, "ask": True},
 }
 
 # Combined lookup for execution and approval (always available regardless of config)
 ALL_TOOLS: dict[str, dict] = {**TOOLS, **BROWSER_TOOLS, **SUBAGENT_TOOLS}
 
+
 BUILTIN_TOOL_GROUPS: dict[str, tuple[str, ...]] = {
     "files": (
         "read_file",
diff --git a/pyproject.toml b/pyproject.toml
index 0e395f06..e9bb75f4 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -1,6 +1,6 @@
 [project]
 name = "cptr"
-version = "0.9.19"
+version = "0.9.20"
 description = "Your computer, from anywhere. Code, manage, and control your machine from the web."
 license = {file = "LICENSE"}
 readme = "README.md"
diff --git a/uv.lock b/uv.lock
index a40c5cef..75c879a6 100644
--- a/uv.lock
+++ b/uv.lock
@@ -284,7 +284,7 @@ wheels = [
 
 [[package]]
 name = "cptr"
-version = "0.9.19"
+version = "0.9.20"
 source = { editable = "." }
 dependencies = [
     { name = "aiosqlite" },