fix: /fast uses native Opus 4.6 fast mode via settings file, not model switch

- Fast mode is speed-optimized API config for Opus 4.6 (2.5x faster, same quality)
- Writes fastMode to .claude/settings.local.json for CLI to pick up
- Adds 'local' to settingSources so CLI reads local settings
- Removes incorrect fastModel/Sonnet-switch logic
- Updates help text with accurate fast mode description

Author: zebbern

claude/client.ts

@@ -191,7 +191,7 @@ export async function sendToClaudeCode(
           // Use Claude Code's system prompt + optional append
           systemPrompt: systemPromptConfig,
           // Load project CLAUDE.md files
-          settingSources: ['project' as const],
+          settingSources: ['project' as const, 'local' as const],
           // Native thinking config (replaces MAX_THINKING_TOKENS env var hack)
           ...(modelOptions?.thinking && { thinking: modelOptions.thinking }),
           // Effort level

claude/info-commands.ts

@@ -75,7 +75,7 @@ export const infoCommands = [
 
   new SlashCommandBuilder()
     .setName('fast')
-    .setDescription('Toggle fast mode — switch between default model and a fast/cheap model (Sonnet)'),
+    .setDescription('Toggle fast mode — 2.5x faster Opus 4.6 responses (higher cost, same quality)'),
 ];
 
 // ================================
@@ -437,7 +437,7 @@ export function createInfoCommandHandlers(deps: InfoCommandHandlerDeps) {
     },
 
     // ================================
-    // /fast — Toggle fast mode
+    // /fast — Toggle fast mode (Opus 4.6 speed-optimized API config)
     // ================================
     // deno-lint-ignore no-explicit-any
     async onFast(ctx: any): Promise<void> {
@@ -450,36 +450,26 @@ export function createInfoCommandHandlers(deps: InfoCommandHandlerDeps) {
       const newFastMode = !current.fastMode;
       deps.updateUnifiedSettings({ fastMode: newFastMode });
 
-      // If there's an active query, switch the model mid-session
-      const activeQuery = getActiveQuery();
-      if (activeQuery) {
-        try {
-          if (newFastMode) {
-            await setActiveModel(current.fastModel || 'claude-sonnet-4-6');
-          } else {
-            // Revert to default model (undefined = SDK default)
-            await setActiveModel(current.defaultModel || undefined);
-          }
-        } catch {
-          // Mid-session switch failed — setting still applies to next query
-        }
+      // Write fastMode to .claude/settings.local.json so the CLI subprocess picks it up
+      try {
+        await writeFastModeToLocalSettings(workDir, newFastMode);
+      } catch (err) {
+        console.error('[/fast] Failed to write local settings:', err);
       }
 
-      const modelName = newFastMode
-        ? (current.fastModel || 'claude-sonnet-4-6')
-        : (current.defaultModel || 'default (Opus)');
-      const midSessionNote = activeQuery
-        ? '\nModel switched on active session.'
+      const activeQuery = getActiveQuery();
+      const sessionNote = activeQuery
+        ? '\n⚠️ Takes effect on **next query** (cannot toggle mid-session via SDK).'
         : '';
 
       await ctx.reply({
         embeds: [{
           color: newFastMode ? 0xffaa00 : 0x5865f2,
           title: newFastMode ? '⚡ Fast Mode ON' : '🧠 Fast Mode OFF',
           description: newFastMode
-            ? `Switched to **${modelName}** — faster responses, lower cost.${midSessionNote}`
-            : `Switched to **${modelName}** — full reasoning power.${midSessionNote}`,
-          footer: { text: 'Use /fast again to toggle back' },
+            ? `Opus 4.6 fast mode enabled — **2.5x faster** responses, higher per-token cost, same quality.${sessionNote}`
+            : `Standard Opus 4.6 mode — normal speed and pricing.${sessionNote}`,
+          footer: { text: 'Use /fast again to toggle' },
           timestamp: new Date().toISOString()
         }]
       });
@@ -491,6 +481,41 @@ export function createInfoCommandHandlers(deps: InfoCommandHandlerDeps) {
 // Helper Functions
 // ================================
 
+import * as path from "https://deno.land/std@0.208.0/path/mod.ts";
+
+/**
+ * Writes fastMode to .claude/settings.local.json in workDir.
+ * The SDK loads this file via settingSources: ['local'].
+ * Merges with existing settings to avoid clobbering other local config.
+ */
+async function writeFastModeToLocalSettings(workDir: string, fastMode: boolean): Promise<void> {
+  const settingsDir = path.join(workDir, ".claude");
+  const settingsPath = path.join(settingsDir, "settings.local.json");
+
+  // Read existing settings (if any)
+  // deno-lint-ignore no-explicit-any
+  let existing: Record<string, any> = {};
+  try {
+    const raw = await Deno.readTextFile(settingsPath);
+    existing = JSON.parse(raw);
+  } catch {
+    // File doesn't exist or is invalid — start fresh
+  }
+
+  // Merge fastMode
+  existing.fastMode = fastMode;
+
+  // Ensure .claude/ directory exists
+  try {
+    await Deno.mkdir(settingsDir, { recursive: true });
+  } catch {
+    // Already exists
+  }
+
+  await Deno.writeTextFile(settingsPath, JSON.stringify(existing, null, 2) + "\n");
+  console.log(`[/fast] Wrote fastMode=${fastMode} to ${settingsPath}`);
+}
+
 // deno-lint-ignore no-explicit-any
 async function sendFullInfoEmbed(ctx: any, account: any, models: any[], mcpServers: any[]): Promise<void> {
   const modelList = models.slice(0, 15).map(m => `• **${m.value}** — ${m.displayName}`).join('\n');

core/handler-registry.ts

@@ -383,10 +383,8 @@ export function createAllHandlers(
     const s = settings.getSettings().unified;
     const opts: ClaudeModelOptions = {};
 
-    // Model — fast mode overrides default model
-    if (s.fastMode && s.fastModel) {
-      opts.model = s.fastModel;
-    } else if (s.defaultModel) {
+    // Model — fast mode is handled by CLI via .claude/settings.local.json (not a model switch)
+    if (s.defaultModel) {
       opts.model = s.defaultModel;
     }
 

help/commands.ts

@@ -786,18 +786,18 @@ export const COMMAND_HELP = {
   },
   "fast": {
     title: "⚡ Fast Mode Toggle",
-    description: "Toggle between the default model (Opus) and a fast/cheap model (Sonnet)",
+    description: "Toggle Opus 4.6 fast mode — 2.5x faster responses at higher per-token cost, same quality",
     usage: "/fast",
     examples: [
       "/fast"
     ],
     parameters: [],
     notes: [
       "Toggles fast mode on/off",
-      "Fast mode uses Sonnet for faster, cheaper responses",
-      "Works mid-session — switches model on an active query",
-      "Configure the fast model via /settings",
-      "Use /fast again to toggle back to your default model"
+      "Fast mode uses the same Opus 4.6 with a speed-optimized API config",
+      "2.5x faster responses, higher per-token cost ($30/$150 MTok)",
+      "Takes effect on the next query (persists across sessions)",
+      "Use /fast again to toggle back to standard mode"
     ]
   }
 };
@@ -877,7 +877,7 @@ export function createHelpHandlers(deps: HelpHandlerDeps) {
             fields: [
               {
                 name: "🤖 Claude Code Commands",
-                value: "`/claude` - Send prompts to Claude Code\n`/claude-enhanced` - Advanced Claude with options\n`/resume` - Resume conversation\n`/claude-cancel` - Cancel running operation\n`/fast` - Toggle fast mode (Opus ↔ Sonnet)",
+                value: "`/claude` - Send prompts to Claude Code\n`/claude-enhanced` - Advanced Claude with options\n`/resume` - Resume conversation\n`/claude-cancel` - Cancel running operation\n`/fast` - Toggle fast mode (2.5x speed)",
                 inline: false
               },
               {

settings/unified-settings.ts

@@ -23,9 +23,8 @@ export interface UnifiedBotSettings {
   // Mode settings
   operationMode: 'normal' | 'plan' | 'auto-accept' | 'danger' | 'dont-ask';
 
-  // Fast mode — toggles between default model and a fast/cheap model
+  // Fast mode — Opus 4.6 speed-optimized API config (2.5x faster, higher cost, same quality)
   fastMode: boolean;
-  fastModel: string;
 
   // Budget settings
   maxBudgetUsd: number | null;
@@ -87,9 +86,8 @@ export const UNIFIED_DEFAULT_SETTINGS: UnifiedBotSettings = {
   // Operation mode
   operationMode: 'normal',
 
-  // Fast mode — off by default, uses Sonnet as fast model
+  // Fast mode — off by default (Opus 4.6 speed-optimized API config)
   fastMode: false,
-  fastModel: 'claude-sonnet-4-6',
 
   // Budget
   maxBudgetUsd: null,