cawdex 1.35.75 → 1.35.77

package/dist/permissions.js

@@ -1,6 +1,59 @@
 import chalk from 'chalk';
 import { saveConfig } from './config.js';
 import { evaluateCommand } from './execpolicy.js';
+export function explainPermission(tool, input, config) {
+    const lines = [
+        `tool: ${tool.name}`,
+        `permission mode: ${config.permissionMode}`,
+    ];
+    if (tool.name === 'bash') {
+        const command = String(input.command || '').trim();
+        const policy = evaluateCommand(command);
+        lines.push(`execpolicy: ${policy.decision}${policy.ruleId ? ` (${policy.ruleId})` : ''}`);
+        if (policy.reason)
+            lines.push(`execpolicy reason: ${policy.reason}`);
+        if (policy.decision === 'forbidden') {
+            lines.push('result: blocked before the normal permission prompt');
+            return {
+                decision: 'deny',
+                reason: policy.reason || 'blocked by execpolicy',
+                lines,
+            };
+        }
+        if (policy.decision === 'allow') {
+            lines.push('result: allowed by execpolicy before mode checks');
+            return {
+                decision: 'allow',
+                reason: `execpolicy${policy.ruleId ? ` ${policy.ruleId}` : ''}`,
+                lines,
+            };
+        }
+    }
+    if (config.permissionMode === 'yolo') {
+        lines.push('result: allowed because yolo permits all non-forbidden tools');
+        return { decision: 'allow', reason: 'permission mode yolo', lines };
+    }
+    if (tool.isReadOnly) {
+        lines.push('result: allowed because read-only tools do not prompt');
+        return { decision: 'allow', reason: 'read-only tool', lines };
+    }
+    if (config.alwaysAllowedTools?.includes(tool.name)) {
+        lines.push('result: allowed by the per-tool always-allow list');
+        return { decision: 'allow', reason: 'always-allow list', lines };
+    }
+    if (config.permissionMode === 'auto' && !tool.isDestructive) {
+        lines.push('result: allowed because auto permits non-destructive tools');
+        return { decision: 'allow', reason: 'auto mode non-destructive tool', lines };
+    }
+    lines.push('result: prompt required before execution');
+    return {
+        decision: 'prompt',
+        reason: config.permissionMode === 'auto'
+            ? 'destructive tool in auto mode'
+            : 'ask mode requires confirmation',
+        lines,
+    };
+}
 /**
  * Check if a tool call is allowed under the current permission mode.
  * Returns true if allowed, false if denied.

package/dist/permissions.js.map

@@ -1 +1 @@
-{"version":3,"file":"permissions.js","sourceRoot":"","sources":["../src/permissions.ts"],"names":[],"mappings":"AAYA,OAAO,KAAK,MAAM,OAAO,CAAC;AAG1B,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAElD;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,IAAU,EACV,KAA8B,EAC9B,MAAoB,EACpB,EAAsB;IAEtB,0DAA0D;IAC1D,qEAAqE;IACrE,kEAAkE;IAClE,kEAAkE;IAClE,oEAAoE;IACpE,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACzB,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;QACxC,MAAM,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;QACpC,IAAI,MAAM,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YACpC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,EAAE,KAAK,MAAM,CAAC,MAAM,IAAI,uBAAuB,EAAE,CAAC,CAAC,CAAC;YAChJ,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;YACrD,OAAO,KAAK,CAAC;QACf,CAAC;QACD,IAAI,MAAM,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,2DAA2D;YAC3D,OAAO,IAAI,CAAC;QACd,CAAC;QACD,gEAAgE;QAChE,wDAAwD;IAC1D,CAAC;IAED,qEAAqE;IACrE,wDAAwD;IACxD,IAAI,MAAM,CAAC,cAAc,KAAK,MAAM;QAAE,OAAO,IAAI,CAAC;IAElD,iCAAiC;IACjC,IAAI,IAAI,CAAC,UAAU;QAAE,OAAO,IAAI,CAAC;IAEjC,kEAAkE;IAClE,sEAAsE;IACtE,oEAAoE;IACpE,qEAAqE;IACrE,uEAAuE;IACvE,mCAAmC;IACnC,IAAI,MAAM,CAAC,kBAAkB,EAAE,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IAEhE,yDAAyD;IACzD,IAAI,MAAM,CAAC,cAAc,KAAK,MAAM,IAAI,CAAC,IAAI,CAAC,aAAa;QAAE,OAAO,IAAI,CAAC;IAEzE,qDAAqD;IACrD,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACzC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IACpD,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;IAE7B,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC,CAAC;IACvE,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAEtC,IAAI,CAAC,KAAK,QAAQ,EAAE,CAAC;QACnB,wEAAwE;QACxE,sEAAsE;QACtE,kEAAkE;QAClE,MAAM,CAAC,kBAAkB,GAAG,MAAM,CAAC,kBAAkB,IAAI,EAAE,CAAC;QAC5D,IAAI,CAAC,MAAM,CAAC,kBAAkB,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACnD,MAAM,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC5C,CAAC;QACD,UAAU,CAAC,MAAM,CAAC,CAAC;QACnB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,sBAAsB,IAAI,CAAC,IAAI,iEAAiE,CAAC,CAAC,CAAC;QACzH,OAAO,IAAI,CAAC;IACd,CAAC;IAED,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,KAAK,CAAC;AAC9C,CAAC;AAED,SAAS,cAAc,CAAC,IAAU,EAAE,KAA8B;IAChE,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;QAClB,KAAK,MAAM;YACT,OAAO,OAAO,KAAK,CAAC,OAAO,EAAE,CAAC;QAChC,KAAK,YAAY;YACf,OAAO,eAAe,KAAK,CAAC,SAAS,KAAK,CAAE,KAAK,CAAC,OAAkB,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,SAAS,CAAC;QAC1G,KAAK,WAAW;YACd,OAAO,WAAW,KAAK,CAAC,SAAS,EAAE,CAAC;QACtC;YACE,OAAO,KAAK,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;IACtD,CAAC;AACH,CAAC"}
+{"version":3,"file":"permissions.js","sourceRoot":"","sources":["../src/permissions.ts"],"names":[],"mappings":"AAYA,OAAO,KAAK,MAAM,OAAO,CAAC;AAG1B,OAAO,EAAE,UAAU,EAAE,MAAM,aAAa,CAAC;AACzC,OAAO,EAAE,eAAe,EAAE,MAAM,iBAAiB,CAAC;AAQlD,MAAM,UAAU,iBAAiB,CAC/B,IAAyD,EACzD,KAA8B,EAC9B,MAAoB;IAEpB,MAAM,KAAK,GAAa;QACtB,SAAS,IAAI,CAAC,IAAI,EAAE;QACpB,oBAAoB,MAAM,CAAC,cAAc,EAAE;KAC5C,CAAC;IAEF,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACzB,MAAM,OAAO,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC,IAAI,EAAE,CAAC;QACnD,MAAM,MAAM,GAAG,eAAe,CAAC,OAAO,CAAC,CAAC;QACxC,KAAK,CAAC,IAAI,CAAC,eAAe,MAAM,CAAC,QAAQ,GAAG,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,EAAE,EAAE,CAAC,CAAC;QAC1F,IAAI,MAAM,CAAC,MAAM;YAAE,KAAK,CAAC,IAAI,CAAC,sBAAsB,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC;QACrE,IAAI,MAAM,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YACpC,KAAK,CAAC,IAAI,CAAC,qDAAqD,CAAC,CAAC;YAClE,OAAO;gBACL,QAAQ,EAAE,MAAM;gBAChB,MAAM,EAAE,MAAM,CAAC,MAAM,IAAI,uBAAuB;gBAChD,KAAK;aACN,CAAC;QACJ,CAAC;QACD,IAAI,MAAM,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,KAAK,CAAC,IAAI,CAAC,kDAAkD,CAAC,CAAC;YAC/D,OAAO;gBACL,QAAQ,EAAE,OAAO;gBACjB,MAAM,EAAE,aAAa,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,IAAI,MAAM,CAAC,MAAM,EAAE,CAAC,CAAC,CAAC,EAAE,EAAE;gBAC/D,KAAK;aACN,CAAC;QACJ,CAAC;IACH,CAAC;IAED,IAAI,MAAM,CAAC,cAAc,KAAK,MAAM,EAAE,CAAC;QACrC,KAAK,CAAC,IAAI,CAAC,8DAA8D,CAAC,CAAC;QAC3E,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,sBAAsB,EAAE,KAAK,EAAE,CAAC;IACtE,CAAC;IAED,IAAI,IAAI,CAAC,UAAU,EAAE,CAAC;QACpB,KAAK,CAAC,IAAI,CAAC,uDAAuD,CAAC,CAAC;QACpE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,gBAAgB,EAAE,KAAK,EAAE,CAAC;IAChE,CAAC;IAED,IAAI,MAAM,CAAC,kBAAkB,EAAE,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;QACnD,KAAK,CAAC,IAAI,CAAC,mDAAmD,CAAC,CAAC;QAChE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,mBAAmB,EAAE,KAAK,EAAE,CAAC;IACnE,CAAC;IAED,IAAI,MAAM,CAAC,cAAc,KAAK,MAAM,IAAI,CAAC,IAAI,CAAC,aAAa,EAAE,CAAC;QAC5D,KAAK,CAAC,IAAI,CAAC,4DAA4D,CAAC,CAAC;QACzE,OAAO,EAAE,QAAQ,EAAE,OAAO,EAAE,MAAM,EAAE,gCAAgC,EAAE,KAAK,EAAE,CAAC;IAChF,CAAC;IAED,KAAK,CAAC,IAAI,CAAC,0CAA0C,CAAC,CAAC;IACvD,OAAO;QACL,QAAQ,EAAE,QAAQ;QAClB,MAAM,EAAE,MAAM,CAAC,cAAc,KAAK,MAAM;YACtC,CAAC,CAAC,+BAA+B;YACjC,CAAC,CAAC,gCAAgC;QACpC,KAAK;KACN,CAAC;AACJ,CAAC;AAED;;;;;;;;;GASG;AACH,MAAM,CAAC,KAAK,UAAU,eAAe,CACnC,IAAU,EACV,KAA8B,EAC9B,MAAoB,EACpB,EAAsB;IAEtB,0DAA0D;IAC1D,qEAAqE;IACrE,kEAAkE;IAClE,kEAAkE;IAClE,oEAAoE;IACpE,IAAI,IAAI,CAAC,IAAI,KAAK,MAAM,EAAE,CAAC;QACzB,MAAM,GAAG,GAAG,MAAM,CAAC,KAAK,CAAC,OAAO,IAAI,EAAE,CAAC,CAAC;QACxC,MAAM,MAAM,GAAG,eAAe,CAAC,GAAG,CAAC,CAAC;QACpC,IAAI,MAAM,CAAC,QAAQ,KAAK,WAAW,EAAE,CAAC;YACpC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,8BAA8B,MAAM,CAAC,MAAM,CAAC,CAAC,CAAC,KAAK,MAAM,CAAC,MAAM,GAAG,CAAC,CAAC,CAAC,EAAE,KAAK,MAAM,CAAC,MAAM,IAAI,uBAAuB,EAAE,CAAC,CAAC,CAAC;YAChJ,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,SAAS,GAAG,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC,CAAC,CAAC;YACrD,OAAO,KAAK,CAAC;QACf,CAAC;QACD,IAAI,MAAM,CAAC,QAAQ,KAAK,OAAO,EAAE,CAAC;YAChC,2DAA2D;YAC3D,OAAO,IAAI,CAAC;QACd,CAAC;QACD,gEAAgE;QAChE,wDAAwD;IAC1D,CAAC;IAED,qEAAqE;IACrE,wDAAwD;IACxD,IAAI,MAAM,CAAC,cAAc,KAAK,MAAM;QAAE,OAAO,IAAI,CAAC;IAElD,iCAAiC;IACjC,IAAI,IAAI,CAAC,UAAU;QAAE,OAAO,IAAI,CAAC;IAEjC,kEAAkE;IAClE,sEAAsE;IACtE,oEAAoE;IACpE,qEAAqE;IACrE,uEAAuE;IACvE,mCAAmC;IACnC,IAAI,MAAM,CAAC,kBAAkB,EAAE,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC;QAAE,OAAO,IAAI,CAAC;IAEhE,yDAAyD;IACzD,IAAI,MAAM,CAAC,cAAc,KAAK,MAAM,IAAI,CAAC,IAAI,CAAC,aAAa;QAAE,OAAO,IAAI,CAAC;IAEzE,qDAAqD;IACrD,MAAM,IAAI,GAAG,cAAc,CAAC,IAAI,EAAE,KAAK,CAAC,CAAC;IACzC,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,MAAM,CAAC,aAAa,IAAI,CAAC,IAAI,EAAE,CAAC,CAAC,CAAC;IACpD,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,IAAI,CAAC,CAAC,CAAC;IAE7B,MAAM,MAAM,GAAG,MAAM,EAAE,CAAC,QAAQ,CAAC,KAAK,CAAC,MAAM,CAAC,sBAAsB,CAAC,CAAC,CAAC;IACvE,MAAM,CAAC,GAAG,MAAM,CAAC,IAAI,EAAE,CAAC,WAAW,EAAE,CAAC;IAEtC,IAAI,CAAC,KAAK,QAAQ,EAAE,CAAC;QACnB,wEAAwE;QACxE,sEAAsE;QACtE,kEAAkE;QAClE,MAAM,CAAC,kBAAkB,GAAG,MAAM,CAAC,kBAAkB,IAAI,EAAE,CAAC;QAC5D,IAAI,CAAC,MAAM,CAAC,kBAAkB,CAAC,QAAQ,CAAC,IAAI,CAAC,IAAI,CAAC,EAAE,CAAC;YACnD,MAAM,CAAC,kBAAkB,CAAC,IAAI,CAAC,IAAI,CAAC,IAAI,CAAC,CAAC;QAC5C,CAAC;QACD,UAAU,CAAC,MAAM,CAAC,CAAC;QACnB,OAAO,CAAC,GAAG,CAAC,KAAK,CAAC,GAAG,CAAC,sBAAsB,IAAI,CAAC,IAAI,iEAAiE,CAAC,CAAC,CAAC;QACzH,OAAO,IAAI,CAAC;IACd,CAAC;IAED,OAAO,CAAC,KAAK,EAAE,IAAI,CAAC,KAAK,GAAG,IAAI,CAAC,KAAK,KAAK,CAAC;AAC9C,CAAC;AAED,SAAS,cAAc,CAAC,IAAU,EAAE,KAA8B;IAChE,QAAQ,IAAI,CAAC,IAAI,EAAE,CAAC;QAClB,KAAK,MAAM;YACT,OAAO,OAAO,KAAK,CAAC,OAAO,EAAE,CAAC;QAChC,KAAK,YAAY;YACf,OAAO,eAAe,KAAK,CAAC,SAAS,KAAK,CAAE,KAAK,CAAC,OAAkB,IAAI,EAAE,CAAC,CAAC,KAAK,CAAC,IAAI,CAAC,CAAC,MAAM,SAAS,CAAC;QAC1G,KAAK,WAAW;YACd,OAAO,WAAW,KAAK,CAAC,SAAS,EAAE,CAAC;QACtC;YACE,OAAO,KAAK,IAAI,CAAC,SAAS,CAAC,KAAK,CAAC,CAAC,KAAK,CAAC,CAAC,EAAE,GAAG,CAAC,EAAE,CAAC;IACtD,CAAC;AACH,CAAC"}

package/dist/pm2-manager.js

@@ -63,32 +63,32 @@ export function listPM2Services(cwd) {
  * @returns A prompt with guidance on creating ecosystem.config.js
  */
 export function buildEcosystemPrompt(cwd) {
-    return `To manage multiple services with PM2, create an ecosystem.config.js file in ${cwd}.
-
-Example structure:
-module.exports = {
-  apps: [
-    {
-      name: 'api-server',
-      script: './dist/server.js',
-      instances: 2,
-      exec_mode: 'cluster',
-      max_memory_restart: '500M',
-      env: {
-        NODE_ENV: 'production'
-      }
-    },
-    {
-      name: 'worker',
-      script: './dist/worker.js',
-      instances: 1,
-      max_memory_restart: '300M'
-    }
-  ]
-};
-
-Then start all services with: pm2 start ecosystem.config.js
-Monitor logs with: pm2 logs
+    return `To manage multiple services with PM2, create an ecosystem.config.js file in ${cwd}.
+
+Example structure:
+module.exports = {
+  apps: [
+    {
+      name: 'api-server',
+      script: './dist/server.js',
+      instances: 2,
+      exec_mode: 'cluster',
+      max_memory_restart: '500M',
+      env: {
+        NODE_ENV: 'production'
+      }
+    },
+    {
+      name: 'worker',
+      script: './dist/worker.js',
+      instances: 1,
+      max_memory_restart: '300M'
+    }
+  ]
+};
+
+Then start all services with: pm2 start ecosystem.config.js
+Monitor logs with: pm2 logs
 View status with: pm2 status`;
 }
 /**

package/dist/query.d.ts

@@ -4,7 +4,6 @@ import type { Mode } from './modes.js';
 export declare function resolveFirstTokenTimeoutMs(config: Pick<CawdexConfig, 'model' | 'provider'>): number;
 export declare function resolveStreamIdleTimeoutMs(fastDirect?: boolean): number;
 export declare function isKnownFlakyOpenRouterModel(config: Pick<CawdexConfig, 'model' | 'provider'>): boolean;
-export declare function fallbackModelForKnownFlakyTurn(config: CawdexConfig, usedFallbackModel?: boolean): string | null;
 export declare function isTurnCancelKeySequence(chunk: Buffer): boolean;
 export declare function shouldUseFastDirectReply(userQuery: string | undefined, mode: Mode, env?: NodeJS.ProcessEnv): boolean;
 export interface WorkingIndicator {

package/dist/query.js

@@ -9,7 +9,7 @@ import { buildSystemPrompt } from './system-prompt.js';
 import { runHooks } from './hooks.js';
 import { scanToolCall, printSecurityWarning } from './security.js';
 import { trackUsage } from './cost-tracker.js';
-import { shouldCompact, compactMessages, quickCompact, buildCompactionConfig, contextCapTokens, enforceContextCap, inferContextWindowTokens, } from './compaction.js';
+import { shouldCompact, compactMessages, quickCompact, estimateTokens, buildCompactionConfig, contextCapTokens, enforceContextCap, inferContextWindowTokens, } from './compaction.js';
 import { assistantTranscriptPrefix, theme, sym, printToolRun, printToolResult, printThinkingOpen, printThinkingText, printThinkingClose, printCost, printApiError, formatDuration, categorizeApiError } from './theme.js';
 import { isVoiceEnabled, getTtsConfig, getAccessibilityConfig, speakAssistantResponse, speak, speakUserEcho, } from './voice.js';
 import { isLikelyDestructive, describeDestructive, countWords, summarize } from './accessibility.js';
@@ -20,6 +20,7 @@ import * as liveQueue from './live-queue.js';
 import { isFooterActive, setFooterActivity, setFooterCost, writeScrollableLine } from './fixed-footer.js';
 import { applyQueuedInputChunk, drainQueuedInputBytes, queuedInputBytesToText } from './prompt-buffer.js';
 import { emit as dbgEmit } from './debug.js';
+import { applyAgentToolInstructions } from './agents-md.js';
 import { buildBenchmarkCompletionReminder, buildBenchmarkTrajectorySystemBlock, makeBenchmarkInvalidToolActionEvent, makeBenchmarkTraceEvent, writeBenchmarkTrace, } from './benchmark-trace.js';
 import { buildTodoStateBlock } from './tools/todo.js';
 import { buildRuntimeInfoBlock } from './runtime-info.js';
@@ -30,11 +31,11 @@ import { archiveLargeToolOutput } from './tool-output-archive.js';
 // not persisted — restart, see hint again. Keyed by sessionId so different
 // sessions get fresh hints.
 const _thinkingHintShownForSession = new Set();
-const INTERACTIVE_FIRST_TOKEN_TIMEOUT_MS = 12_000;
+const INTERACTIVE_FIRST_TOKEN_TIMEOUT_MS = 8_000;
 const INTERACTIVE_FLAKY_FIRST_TOKEN_TIMEOUT_MS = 6_000;
 const NON_INTERACTIVE_FIRST_TOKEN_TIMEOUT_MS = 60_000;
 const NON_INTERACTIVE_FLAKY_FIRST_TOKEN_TIMEOUT_MS = 20_000;
-const FAST_DIRECT_FIRST_TOKEN_TIMEOUT_MS = 8_000;
+const FAST_DIRECT_FIRST_TOKEN_TIMEOUT_MS = 5_000;
 const INTERACTIVE_STREAM_IDLE_TIMEOUT_MS = 45_000;
 const NON_INTERACTIVE_STREAM_IDLE_TIMEOUT_MS = 120_000;
 const FAST_DIRECT_STREAM_IDLE_TIMEOUT_MS = 20_000;
@@ -84,13 +85,6 @@ function fallbackModelForTurn(config, usedFallbackModel) {
         return null;
     return fallback;
 }
-export function fallbackModelForKnownFlakyTurn(config, usedFallbackModel = false) {
-    if (process.env.CAWDEX_ALLOW_FLAKY_MODELS === '1')
-        return null;
-    if (!isKnownFlakyOpenRouterModel(config))
-        return null;
-    return fallbackModelForTurn(config, usedFallbackModel);
-}
 export function isTurnCancelKeySequence(chunk) {
     const seq = chunk.toString('utf8');
     return (seq === '\x1b' ||
@@ -122,6 +116,7 @@ const FAST_DIRECT_POSITIVE = [
 ];
 const FAST_DIRECT_REPO_OR_TOOL = /\b(repo|repository|codebase|workspace|file|folder|directory|path|terminal|powershell|shell|command|npm|node|python|typescript|javascript|git|commit|diff|pr|pull request|branch|test|build|lint|install|package|debug|error|stack trace|log|fix|implement|refactor|review|audit|security|benchmark|run|execute|read|search|grep|edit|patch|write[- ]file|website|web\s*site|site|portfolio|landing\s+page|web\s+page|html|css|react|vue|svelte|vite|next\.?js|single[- ]file|dashboard|component|form|desktop|save|hireable|resume)\b/i;
 const FAST_DIRECT_CONTEXTUAL = /^(?:(?:please\s+)?(?:can|could|would|will)\s+you\s+(?:please\s+)?|please\s+)?(continue|carry on|resume|do it|same|again|that|this|those|these|it)\b/i;
+const FAST_DIRECT_REVISION_CONTEXT = /^(?:(?:please\s+)?(?:can|could|would|will)\s+you\s+(?:please\s+)?)?(?:make|change|update|revise|rewrite|redo|adjust|convert|turn)\s+(?:it|that|this|him|her|them|he|she)\b/i;
 export function shouldUseFastDirectReply(userQuery, mode, env = process.env) {
     if (env.CAWDEX_FAST_DIRECT === '0')
         return false;
@@ -136,6 +131,8 @@ export function shouldUseFastDirectReply(userQuery, mode, env = process.env) {
         return false;
     if (FAST_DIRECT_CONTEXTUAL.test(text))
         return false;
+    if (FAST_DIRECT_REVISION_CONTEXT.test(text))
+        return false;
     if (FAST_DIRECT_REPO_OR_TOOL.test(text))
         return false;
     return FAST_DIRECT_POSITIVE.some((pattern) => pattern.test(text));
@@ -154,9 +151,9 @@ function printInteractiveTurnAccepted(config) {
     }
 }
 export function formatWorkingIndicatorFrame(elapsedMs, frameIndex = 0, message = 'Working') {
-    const frames = ['\u25e6', '\u25c7', '\u25c6', '\u25c7'];
+    const frames = ['\u25dc', '\u25dd', '\u25de', '\u25df'];
     const frame = frames[Math.abs(frameIndex) % frames.length];
-    return `  ${frame} ${message} (${formatDuration(elapsedMs)} \u2022 esc to interrupt)`;
+    return `  ${frame} ${message} (${formatDuration(elapsedMs)} \u2022 Esc/F5 to interrupt)`;
 }
 function startWorkingIndicator(startedAtMs, screenReader, turn = 0) {
     if (screenReader)
@@ -166,11 +163,11 @@ function startWorkingIndicator(startedAtMs, screenReader, turn = 0) {
     if (!process.stdout.isTTY)
         return null;
     const messages = [
-        'Working hard on your request',
-        'Sumi ink settling',
-        'Edo lanterns lit',
-        'Kamon crest aligned',
-        'Neon alley quiet',
+        'Sumi ink moving',
+        'Edo lanterns cycling',
+        'Kamon crest pulsing',
+        'Neon shoji breathing',
+        'Signal blade drawn',
     ];
     let frame = 0;
     let stopped = false;
@@ -1129,18 +1126,21 @@ export async function runQuery(ctx) {
     // user's configured fallbackModel. After we use it, this latches so we
     // don't bounce back and forth between failing models in a single chain.
     let usedFallbackModel = false;
-    const immediateFallback = fallbackModelForKnownFlakyTurn(ctx.config, usedFallbackModel);
-    if (immediateFallback) {
-        usedFallbackModel = true;
-        const failedModel = ctx.config.model;
-        ctx.config.model = immediateFallback;
+    // Reliability path when fallback is intentionally disabled:
+    // retry the SAME configured model once on transient empty/timeout/unknown
+    // provider failures before surfacing a hard error.
+    let usedPrimaryRetry = false;
+    const retryPrimaryModelOnce = (reason) => {
+        if (usedPrimaryRetry)
+            return false;
+        usedPrimaryRetry = true;
+        const model = ctx.config.model;
         resetClient();
-        console.log(theme.warning(`  ${sym.warn} ${failedModel} is known to stall in interactive OpenRouter sessions; switching this turn to ${immediateFallback}.`));
-        console.log(theme.dim('    Override only if you really want it: CAWDEX_ALLOW_FLAKY_MODELS=1'));
-    }
-    if (!chainFastDirect) {
-        printInteractiveTurnAccepted(ctx.config);
-    }
+        console.log(theme.warning(`  ${sym.warn} ${reason} — retrying once on the same model ${model}.`));
+        console.log(theme.dim('    Fallback is disabled; this is a transient-recovery retry only.'));
+        turns--;
+        return true;
+    };
     // Tracks whether ANY reasoning tokens arrived across the entire chain.
     // Used at chain-end to print a one-time "/thinking is ON but this model
     // doesn't emit reasoning" hint. Hoisted to chain scope (not per-turn)
@@ -1210,8 +1210,11 @@ export async function runQuery(ctx) {
     // as new text, drowning the actual response).
     const isScreenReader = ctx.config.voice?.accessibility?.screenReader === true;
     const inputGuard = startInputSuppression(isScreenReader);
+    let earlyWorkingIndicator = null;
     try {
         if (!chainFastDirect) {
+            printInteractiveTurnAccepted(ctx.config);
+            earlyWorkingIndicator = startWorkingIndicator(chainStart, isScreenReader, 0);
             // Turn-boundary collapse runs BEFORE compaction. Every completed prior
             // turn becomes [user, "<final text>\n[Completed: used X, Y]"] — the
             // model no longer sees stale tool_calls that it might mistake for
@@ -1263,9 +1266,12 @@ export async function runQuery(ctx) {
             if (!fastDirect) {
                 replaceMessagesInPlace(ctx.messages, quickCompact(ctx.messages));
             }
+            const requestTools = fastDirect
+                ? []
+                : applyAgentToolInstructions(ALL_TOOLS, ctx.cwd, ctx.config.model);
             const systemPrompt = fastDirect
                 ? FAST_DIRECT_SYSTEM_PROMPT
-                : buildSystemPrompt(ctx.config, ctx.cwd, ctx.mode, userQuery);
+                : buildSystemPrompt(ctx.config, ctx.cwd, ctx.mode, userQuery, requestTools);
             let visibleMessages = fastDirect
                 ? (userQuery ? [{ role: 'user', content: userQuery }] : [])
                 : maskOldToolResults(ctx.messages);
@@ -1292,7 +1298,9 @@ export async function runQuery(ctx) {
             // override.
             const stateBlock = fastDirect ? null : buildStateBlock(visibleMessages);
             const runtimeInfoBlock = fastDirect ? null : buildRuntimeInfoBlock(ctx.cwd);
-            const repoMapBlock = fastDirect ? null : buildAutoRepoMapBlock(ctx.cwd, userQuery);
+            const repoMapBlock = fastDirect || ctx.mode === 'design'
+                ? null
+                : buildAutoRepoMapBlock(ctx.cwd, userQuery);
             const globalPlanBlock = fastDirect ? null : buildGlobalPlanBlock(visibleMessages);
             const todoStateBlock = fastDirect ? null : buildTodoStateBlock(ctx.cwd);
             const benchmarkTrajectoryBlock = !fastDirect && ctx.mode === 'benchmark'
@@ -1326,6 +1334,7 @@ export async function runQuery(ctx) {
             let lastCharWasNewline = false; // collapse 3+ consecutive newlines down to 2
             let consecutiveNewlines = 0;
             const turnStart = Date.now();
+            let usageRecorded = false;
             // Loop detection state: a stuck model can stream the SAME N-char
             // window of text 50+ times in a single API call (observed in the
             // wild with openrouter/owl-alpha emitting tool-call JSON as text).
@@ -1446,16 +1455,22 @@ export async function runQuery(ctx) {
             // line and then announce every subsequent token as "after the
             // waiting line", which is noisier than helpful).
             let firstTokenSeen = false;
+            let firstTokenLatencyMs = null;
             // Note: the outer `isScreenReader` declared at the top of runQuery
             // (line ~340) is in scope here via closure — no need for a second
             // declaration. Previously this re-declared inside the while loop
             // and TypeScript tolerated it as a different block scope, but it
             // was confusing and the audit flagged it as bug-bait.
             //
-            // Live waiting indicator on the response line. It keeps elapsed time
-            // and the interrupt hint visible while still clearing itself before
-            // the first model event writes real output.
-            let workingIndicator = startWorkingIndicator(turnStart, isScreenReader, turns);
+            // Live waiting indicator on the response line. It keeps motion and the
+            // interrupt hint visible while still clearing itself before the first
+            // model event writes real output.
+            let workingIndicator = turns === 1 ? earlyWorkingIndicator : null;
+            if (turns === 1)
+                earlyWorkingIndicator = null;
+            if (!workingIndicator) {
+                workingIndicator = startWorkingIndicator(turnStart, isScreenReader, turns);
+            }
             // Slow-model warning and first-token watchdog. The warning is a
             // UX hint; the watchdog is the hard recovery path for providers
             // that accept a request but then never produce a stream event.
@@ -1482,7 +1497,6 @@ export async function runQuery(ctx) {
                 const requestConfig = fastDirect
                     ? { ...ctx.config, maxTokens: Math.min(ctx.config.maxTokens ?? 700, 700) }
                     : ctx.config;
-                const requestTools = fastDirect ? [] : ALL_TOOLS;
                 const stream = streamChat(requestConfig, apiMessages, requestTools, streamAbort.signal);
                 const iterator = stream[Symbol.asyncIterator]();
                 while (true) {
@@ -1529,6 +1543,7 @@ export async function runQuery(ctx) {
                     // model warning timer; subsequent events are normal streaming.
                     if (!firstTokenSeen) {
                         firstTokenSeen = true;
+                        firstTokenLatencyMs = Date.now() - turnStart;
                         clearTimeout(slowTimer);
                         if (streamWaitTimer)
                             clearTimeout(streamWaitTimer);
@@ -1573,7 +1588,12 @@ export async function runQuery(ctx) {
                     else if (event.type === 'done') {
                         if (event.usage) {
                             const u = event.usage;
-                            const { cost, warning } = trackUsage(ctx.sessionId, ctx.config.model, u.prompt, u.completion);
+                            const turnDurationMs = Date.now() - turnStart;
+                            const { cost, warning } = trackUsage(ctx.sessionId, ctx.config.model, u.prompt, u.completion, {
+                                provider: ctx.config.provider,
+                                firstTokenMs: firstTokenLatencyMs,
+                                durationMs: turnDurationMs,
+                            });
                             chainStats.benchmarkUsageEvents.push({
                                 model: ctx.config.model,
                                 promptTokens: u.prompt,
@@ -1581,12 +1601,16 @@ export async function runQuery(ctx) {
                                 totalTokens: u.total || u.prompt + u.completion,
                                 estimatedCostUsd: cost,
                             });
-                            setFooterCost(cost, u.prompt, u.completion);
+                            usageRecorded = true;
+                            setFooterCost(cost, u.prompt, u.completion, {
+                                firstTokenMs: firstTokenLatencyMs,
+                                durationMs: turnDurationMs,
+                            });
                             // Single newline separator if we just streamed text, then the
                             // compact telemetry line.
                             if (hasOutput && !lastCharWasNewline)
                                 process.stdout.write('\n');
-                            printCost(u.prompt, u.completion, cost, warning, Date.now() - turnStart);
+                            printCost(u.prompt, u.completion, cost, warning, turnDurationMs);
                         }
                         try {
                             streamAbort.abort();
@@ -1596,6 +1620,28 @@ export async function runQuery(ctx) {
                         break;
                     }
                 }
+                if (!usageRecorded && (hasOutput || (toolCalls && toolCalls.length > 0))) {
+                    const promptEstimate = Math.max(1, estimateTokens(apiMessages));
+                    const completionEstimate = Math.max(1, Math.ceil(((fullText || '') + (toolCalls ? JSON.stringify(toolCalls) : '')).length / 3.5));
+                    const turnDurationMs = Date.now() - turnStart;
+                    const { cost } = trackUsage(ctx.sessionId, ctx.config.model, promptEstimate, completionEstimate, {
+                        provider: ctx.config.provider,
+                        firstTokenMs: firstTokenLatencyMs,
+                        durationMs: turnDurationMs,
+                    });
+                    chainStats.benchmarkUsageEvents.push({
+                        model: ctx.config.model,
+                        promptTokens: promptEstimate,
+                        completionTokens: completionEstimate,
+                        totalTokens: promptEstimate + completionEstimate,
+                        estimatedCostUsd: cost,
+                    });
+                    setFooterCost(cost, promptEstimate, completionEstimate, {
+                        firstTokenMs: firstTokenLatencyMs,
+                        durationMs: turnDurationMs,
+                    });
+                    usageRecorded = true;
+                }
                 clearTimeout(slowTimer);
                 if (streamWaitTimer)
                     clearTimeout(streamWaitTimer);
@@ -1629,9 +1675,12 @@ export async function runQuery(ctx) {
                         turns--;
                         continue;
                     }
+                    if (retryPrimaryModelOnce(`${failedModel} produced no stream events for ${formatDuration(firstTokenTimeoutMs)}`)) {
+                        continue;
+                    }
                     const timeoutMsg = `${failedModel} produced no stream events for ${formatDuration(firstTokenTimeoutMs)}`;
                     console.log(theme.error(`  ${sym.warn} ${timeoutMsg}.`));
-                    console.log(theme.dim('    No fallback is configured, so the request was cancelled and the prompt can recover.'));
+                    console.log(theme.dim('    Fallback is disabled and the same-model retry already ran; cancelling so the prompt can recover.'));
                     console.log(theme.dim('    Use /fallback <model-id> to enable automatic retry, or /model <known-good-model> to switch primary models.'));
                     ctx.messages.push({ role: 'assistant', content: `[Provider timeout: ${timeoutMsg}]` });
                     break;
@@ -1651,6 +1700,9 @@ export async function runQuery(ctx) {
                         turns--;
                         continue;
                     }
+                    if (!hasOutput && !toolCalls && retryPrimaryModelOnce(timeoutMsg)) {
+                        continue;
+                    }
                     if (fullText.trim()) {
                         const partial = `${fullText.trimEnd()}\n[Provider timeout: stream stalled before completion]`;
                         console.log(theme.warning(`  ${sym.warn} stream stalled before completion; returning the partial response.`));
@@ -1659,7 +1711,7 @@ export async function runQuery(ctx) {
                     }
                     else {
                         console.log(theme.error(`  ${sym.warn} ${timeoutMsg}.`));
-                        console.log(theme.dim('    The request was cancelled so the prompt can recover instead of hanging indefinitely.'));
+                        console.log(theme.dim('    Same-model retry already ran; cancelling so the prompt can recover instead of hanging indefinitely.'));
                         console.log(theme.dim('    Use /fallback <model-id> to enable automatic retry, or /model <known-good-model> to switch primary models.'));
                         ctx.messages.push({ role: 'assistant', content: `[Provider timeout: ${timeoutMsg}]` });
                     }
@@ -1732,6 +1784,9 @@ export async function runQuery(ctx) {
                     turns--; // this retry doesn't burn a turn slot from the max-turns budget
                     continue;
                 }
+                if (cat.category === 'unknown' && retryPrimaryModelOnce(`${ctx.config.model} returned a cryptic provider error`)) {
+                    continue;
+                }
                 printApiError(msg, {
                     baseURL: ctx.config.baseURL,
                     provider: ctx.config.provider,
@@ -1765,9 +1820,12 @@ export async function runQuery(ctx) {
                     turns--;
                     continue;
                 }
+                if (retryPrimaryModelOnce(`${failedModel} returned an empty response`)) {
+                    continue;
+                }
                 const emptyMsg = `${failedModel} returned an empty response`;
                 console.log(theme.error(`  ${sym.warn} ${emptyMsg}.`));
-                console.log(theme.dim('    No fallback is configured, so this empty response was not retried.'));
+                console.log(theme.dim('    Fallback is disabled and the same-model retry already ran.'));
                 console.log(theme.dim('    Use /fallback <model-id> to enable automatic retry, or /model <known-good-model> to switch primary models.'));
                 ctx.messages.push({ role: 'assistant', content: `[Provider empty response: ${emptyMsg}]` });
                 break;
@@ -1873,6 +1931,8 @@ export async function runQuery(ctx) {
         }
         // Chain ended; back to idle so F1 reports the correct state.
         setStatus({ state: 'idle' });
+        if (isFooterActive())
+            setFooterActivity('Ready', 0, null);
         // ── Voice: read the assistant's final response ────────────
         // Off the hot path — fire-and-forget so the next prompt appears
         // immediately. The playback runs in background; F2 pauses, F4 skips.
@@ -1972,6 +2032,8 @@ export async function runQuery(ctx) {
         }
     }
     finally {
+        earlyWorkingIndicator?.stop();
+        earlyWorkingIndicator = null;
         // Drain any queued user input typed during streaming. Stash on
         // globalThis for the REPL loop in index.ts to restore into the
         // next editable prompt. Enter typed mid-stream is preserved as
@@ -1985,6 +2047,8 @@ export async function runQuery(ctx) {
         // can't be aborted between turns by Shift+F5 (soft-cancel).
         globalThis.__turnAbortCtl = null;
         globalThis.__turnCancelCurrent = null;
+        if (isFooterActive())
+            setFooterActivity('Ready', 0, null);
     }
 }
 const TOOL_CALL_LOOP_THRESHOLD = 3;