moderado 0.4.26 → 0.4.33

package/dist/commands/chat.js

@@ -10,7 +10,7 @@ import { MemoryCredentialStore } from '../credentials.js';
 import { WindowsCredentialStore } from '../windows_credentials.js';
 import { TerminalApprovalHandler } from '../ui/terminal_approval.js';
 import { selectCompatibleModelOverlay, selectModelOverlay, showModelConnectionRequired } from '../ui/model_selector.js';
-import { connectProviderInteractive, isAuthenticationFailure, isModeradoCloudOAuthConnection, loginModeradoCloudInteractive, replaceProviderKeyInteractive } from '../ui/provider_connect.js';
+import { connectProviderInteractive, isAuthenticationFailure, isModeradoOAuthConnection, loginModeradoInteractive, replaceProviderKeyInteractive } from '../ui/provider_connect.js';
 import { initWorkspace } from './init.js';
 import { expandMentions, createWorkspaceFileSource } from '../ui/file_mentions.js';
 import { listFiles } from '../vendor/tools/index.js';
@@ -23,9 +23,15 @@ import { inspectGitWorkspace, readGitDiff } from '../vendor/tools/index.js';
 import { buildWorkspaceMap } from '../repo_map.js';
 import { evaluateTokenBudget, parseBudgetCommand } from '../budget.js';
 import { discoverSkills, formatSkillContext, formatSkillsList, getEnabledSkillNames, selectEnabledSkills, setSkillEnabled } from '../skills.js';
+import { getHelpCommandLines } from '../ui/help_content.js';
 export function createConnectionProvider(connection) {
     return new NvidiaAdapter({ apiKey: connection.apiKey, baseUrl: connection.baseUrl, providerId: connection.id, providerName: connection.displayName });
 }
+export function getProviderStatusName(connection) {
+    return connection?.id === 'moderado' || connection?.id === 'moderado-cloud'
+        ? 'Moderado'
+        : connection?.displayName;
+}
 export function formatSessionExitSummary(session, modelId) {
     const usage = session.usage;
     const tokens = usage.available
@@ -172,8 +178,18 @@ export function formatTurnFailureAnswer(status, errorEvent, modelId, otherConnec
     const cleanDetail = rawDetail
         .replace(/^(?:NVIDIA NIM|OpenAI|OpenRouter|[a-zA-Z0-9_-]+)\s*(?:request failed|rate limit exceeded|service or model unavailable)?\s*(?:\([^)]*\))?\s*(?:during\s*streamChat)?:\s*/i, '')
         .trim();
+    if (errorEvent?.code === 'ERR_GATEWAY_UNAUTHORIZED') {
+        return `⚠️ **Gateway Authorization (${errorEvent.code}):**\n${cleanDetail}`;
+    }
+    if (errorEvent?.code?.toUpperCase().includes('RATE_LIMIT')) {
+        return `⚠️ **Model Error (${errorEvent.code}):**\n${cleanDetail || 'Provider rate limit reached.'}\n\n👉 **Suggestion:** The provider temporarily limited requests for \`${modelId}\`. Please retry shortly or choose another model with \`/model\`.`;
+    }
     return `⚠️ **Model Error (${errorEvent?.code || 'ERR_INFERENCE_FAILED'}):**\n${cleanDetail || 'Model returned no response.'}\n\n👉 **Suggestion:** The model \`${modelId}\` encountered an error. Switch models with \`/model\` or choose a provider with \`/connect\`${altSuggestion}.`;
 }
+/** A turn whose provider rejected its credential may recover by re-entering a key once. */
+export function turnFailedAuthentication(status, errorEvent) {
+    return status === 'failed' && isAuthenticationFailure(errorEvent);
+}
 export function resolveTurnAssistantAnswer(result, previousHistoryLength, lastErrorEvent, streamedAnswer = '', currentModel = 'unknown', otherConnections = []) {
     if (result.status === 'cancelled') {
         const partial = cleanConversationalFiller(streamedAnswer);
@@ -192,10 +208,48 @@ export function resolveTurnAssistantAnswer(result, previousHistoryLength, lastEr
         isError: false,
     };
 }
-export function saveFailedTurnPrompt(history, prompt) {
+export function saveFailedTurnPrompt(history, prompt, partialResponse = '') {
     if (!prompt.trim())
         return history;
-    return [...history, { role: 'user', content: prompt }];
+    const interruptionNote = '[This response was interrupted before completion. Continue the task if asked.]';
+    const interruptedResponse = partialResponse.trim()
+        ? `${partialResponse.trim()}\n\n${interruptionNote}`
+        : '[This task was interrupted before a response was produced. Continue the task if asked.]';
+    return [...history, { role: 'user', content: prompt }, { role: 'assistant', content: interruptedResponse }];
+}
+const CONTINUE_INTERRUPTED_TASK_PROMPT = 'Continue the interrupted task from the saved conversation. Start from the unfinished work and avoid repeating completed steps.';
+/** Braille spinner frames for the live "still running" status row. */
+const ACTIVITY_SPINNER_FRAMES = ['⠋', '⠙', '⠹', '⠸', '⠼', '⠴', '⠦', '⠧', '⠇', '⠏'];
+function hasInterruptedTurn(messages) {
+    for (let index = messages.length - 1; index >= 0; index -= 1) {
+        const message = messages[index];
+        if (message.role !== 'assistant' || !message.content?.trim())
+            continue;
+        return message.content.includes('[This response was interrupted before completion.')
+            || message.content.includes('[This task was interrupted before a response was produced.');
+    }
+    return false;
+}
+function restoreLastSessionTurn(messages) {
+    let assistantIndex = -1;
+    for (let index = messages.length - 1; index >= 0; index -= 1) {
+        if (messages[index].role === 'assistant' && messages[index].content?.trim()) {
+            assistantIndex = index;
+            break;
+        }
+    }
+    if (assistantIndex < 0) {
+        const question = [...messages].reverse().find((message) => message.role === 'user' && message.content?.trim())?.content ?? '';
+        return { question, answer: question ? 'This task was interrupted. Type “continue” to resume with the current model.' : '' };
+    }
+    let question = '';
+    for (let index = assistantIndex - 1; index >= 0; index -= 1) {
+        if (messages[index].role === 'user' && messages[index].content?.trim()) {
+            question = messages[index].content ?? '';
+            break;
+        }
+    }
+    return { question, answer: messages[assistantIndex].content ?? '' };
 }
 export function isBareExitCommand(input) {
     return input.trim().toLowerCase() === 'exit';
@@ -429,7 +483,8 @@ export function formatDirectWebSearchAnswer(result, modelConnected = false) {
         .filter((line) => line && line !== '...' && line !== '---' && !SEARCH_BLOCK_METADATA.test(line) && !/^https?:\/\//i.test(line));
     if (!lines.length)
         return 'Web search returned no readable content.';
-    return `${lines.slice(0, 10).join('\n')}\n\n(${modelConnected ? 'model summary unavailable — ' : 'no model connected — '}raw search excerpt)`;
+    const excerpt = lines.slice(0, 10).join('\n');
+    return `${excerpt}\n\n${modelConnected ? 'Live web search excerpt.' : 'Live web search excerpt. Connect a model for a summary.'}`;
 }
 /** Environment configuration overrides the persistent search provider preference. */
 export function resolveWebSearchProvider(config) {
@@ -739,6 +794,7 @@ export async function handleChatSession(args, version, signal) {
             modelId: currentModel,
             mode: activeMode,
         });
+    currentModel = args.model ?? (activeSession.providerId === activeConnection?.id ? activeSession.modelId : undefined) ?? currentModel;
     if (activeSession.usage.available &&
         !activeSession.usage.costKnown &&
         activeSession.providerId === activeConnection?.id &&
@@ -779,9 +835,10 @@ export async function handleChatSession(args, version, signal) {
     const router = new Router();
     const loop = new AgentLoop();
     let conversationHistory = [...activeSession.messages];
-    let lastQuestion = '';
-    let lastAnswer = '';
-    let lastTurnDone = false;
+    const restoredTurn = restoreLastSessionTurn(conversationHistory);
+    let lastQuestion = restoredTurn.question;
+    let lastAnswer = restoredTurn.answer;
+    let lastTurnDone = conversationHistory.length > 0 && !hasInterruptedTurn(conversationHistory);
     let lastThoughtTime = 0;
     let lastOutputTokenRate;
     let lastTokenUsage;
@@ -797,10 +854,18 @@ export async function handleChatSession(args, version, signal) {
             isFirst = false;
         }
         else {
+            let autoContinueRequested = false;
             const turn = await promptInteractiveTurn({
+                getProviderName: () => getProviderStatusName(activeConnection),
                 model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(), workspace: canonicalWorkspace, version,
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, initialMode: activeMode, initialAutoApprove: activeAutoApprove, isFirstTurn: isFirst, signal, chatQuestion: lastQuestion || undefined, chatAnswer: lastAnswer || undefined, chatThoughtTime: lastThoughtTime, chatTaskDone: lastTurnDone, outputTokenRate: lastOutputTokenRate, tokenUsage: lastTokenUsage,
                 queuedCommands: commandQueue.items,
+                takeAutoContinuePrompt: () => {
+                    if (!autoContinueRequested)
+                        return undefined;
+                    autoContinueRequested = false;
+                    return CONTINUE_INTERRUPTED_TASK_PROMPT;
+                },
                 onExit: exitMessage,
                 questionHistory: conversationHistory.flatMap((message) => message.role === 'user' && message.content?.trim() ? [message.content] : []),
                 onModelSelect: async (drawFrame) => {
@@ -810,7 +875,7 @@ export async function handleChatSession(args, version, signal) {
                     }
                     if (activeConnection.kind === 'openai-compatible') {
                         const modelId = await selectCompatibleModelOverlay({
-                            apiKey: activeConnection.id === 'moderado-cloud' ? undefined : activeConnection.apiKey,
+                            apiKey: activeConnection.id === 'moderado' ? undefined : activeConnection.apiKey,
                             baseUrl: activeConnection.baseUrl,
                             providerId: activeConnection.id,
                             providerName: activeConnection.displayName,
@@ -823,15 +888,25 @@ export async function handleChatSession(args, version, signal) {
                             activeConnection = { ...activeConnection, defaultModel: modelId };
                             saveConnection(activeConnection);
                             config = loadConfig();
+                            activeSession.providerId = activeConnection.id;
+                            activeSession.providerName = activeConnection.displayName;
+                            activeSession.modelId = modelId;
+                            sessionStore.save(activeSession);
+                            autoContinueRequested = hasInterruptedTurn(activeSession.messages);
                         }
                         return modelId;
                     }
-                    const selection = await selectModelOverlay({ apiKey: activeConnection.id === 'moderado-cloud' ? undefined : activeConnection.apiKey, currentModel, signal, saveSelectionByDefault: true, drawFrame });
-                    if (selection.modelId) {
+                    const selection = await selectModelOverlay({ apiKey: activeConnection.id === 'moderado' ? undefined : activeConnection.apiKey, currentModel, signal, saveSelectionByDefault: true, drawFrame });
+                    if (selection.modelId && selection.modelId !== currentModel) {
                         currentModel = selection.modelId;
                         activeConnection = { ...activeConnection, defaultModel: selection.modelId };
                         saveConnection(activeConnection);
                         config = loadConfig();
+                        activeSession.providerId = activeConnection.id;
+                        activeSession.providerName = activeConnection.displayName;
+                        activeSession.modelId = selection.modelId;
+                        sessionStore.save(activeSession);
+                        autoContinueRequested = hasInterruptedTurn(activeSession.messages);
                     }
                     return selection.modelId;
                 },
@@ -839,7 +914,7 @@ export async function handleChatSession(args, version, signal) {
                     const connection = await connectProviderInteractive({
                         signal,
                         drawFrame,
-                        gatewayUrl: config.connections?.['moderado-cloud']?.baseUrl,
+                        gatewayUrl: config.connections?.moderado?.baseUrl,
                         activeConnection,
                         savedConnections: config.connections,
                         connectProviders: config.connectProviders,
@@ -961,15 +1036,17 @@ export async function handleChatSession(args, version, signal) {
                 onSession: async (command, drawFrame) => {
                     let action = command.slice('/session'.length).trim();
                     if (!action) {
-                        action = (await selectListPopup('Session', [
+                        const sessionActions = [
                             { label: 'New session', value: 'new', description: 'Start with an empty conversation.' },
                             { label: 'List or resume', value: 'resume', description: 'Load a saved session for this workspace.' },
+                            ...(hasInterruptedTurn(activeSession.messages) ? [{ label: 'Continue interrupted task', value: 'continue', description: 'Continue the last stopped task with the current model.' }] : []),
                             { label: 'Export transcript', value: 'export', description: 'Write a redacted Markdown transcript in this workspace.' },
                             { label: 'Undo latest agent change', value: 'undo', description: 'Restore the last checkpoint when safe.' },
                             { label: 'Redo latest agent change', value: 'redo', description: 'Reapply the last safely undone checkpoint.' },
                             { label: 'Share transcript', value: 'share', description: 'Choose an explicit workspace output path.' },
                             { label: 'Compact history', value: 'compact', description: 'Reduce older messages without calling a model.' },
-                        ], { drawFrame, signal })) ?? '';
+                        ];
+                        action = (await selectListPopup('Session', sessionActions, { drawFrame, signal })) ?? '';
                     }
                     if (action === 'new') {
                         activeSession = createSession(canonicalWorkspace, { providerId: activeConnection?.id, providerName: activeConnection?.displayName, modelId: currentModel, mode: activeMode });
@@ -983,6 +1060,14 @@ export async function handleChatSession(args, version, signal) {
                         sessionStore.save(activeSession);
                         return;
                     }
+                    if (action === 'continue') {
+                        if (!hasInterruptedTurn(activeSession.messages)) {
+                            drawFrame(renderBoxLines('Continue task', ['There is no interrupted task in this session.', '', 'Press Esc or Enter to return.'], 64));
+                            return;
+                        }
+                        autoContinueRequested = true;
+                        return;
+                    }
                     if (action === 'list' || action === 'resume') {
                         const sessions = sessionStore.listSessions(canonicalWorkspace);
                         const id = await selectListPopup('Saved sessions', sessions.map((entry) => ({
@@ -995,8 +1080,12 @@ export async function handleChatSession(args, version, signal) {
                             activeSession = loaded;
                             conversationHistory = [...loaded.messages];
                             sessionTokens = loaded.usage.totalTokens;
-                            currentModel = loaded.modelId ?? currentModel;
+                            currentModel = loaded.providerId === activeConnection?.id ? loaded.modelId ?? currentModel : currentModel;
                             activeMode = loaded.mode;
+                            const restored = restoreLastSessionTurn(conversationHistory);
+                            lastQuestion = restored.question;
+                            lastAnswer = restored.answer;
+                            lastTurnDone = conversationHistory.length > 0 && !hasInterruptedTurn(conversationHistory);
                         }
                         return;
                     }
@@ -1090,7 +1179,7 @@ export async function handleChatSession(args, version, signal) {
             lastTokenUsage = undefined;
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime, queuedCommands: commandQueue.items,
@@ -1106,7 +1195,7 @@ export async function handleChatSession(args, version, signal) {
             lastTokenUsage = undefined;
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime, queuedCommands: commandQueue.items,
@@ -1141,7 +1230,7 @@ export async function handleChatSession(args, version, signal) {
             }
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime,
@@ -1180,7 +1269,7 @@ export async function handleChatSession(args, version, signal) {
             lastTokenUsage = undefined;
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime,
@@ -1219,7 +1308,7 @@ export async function handleChatSession(args, version, signal) {
             lastOutputTokenRate = undefined;
             lastTokenUsage = undefined;
             process.stdout.write('\x1b[H\x1b[J');
-            process.stdout.write(renderFullWelcomeScreen({ model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(), usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode, autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer, chatThoughtTime: lastThoughtTime, queuedCommands: commandQueue.items }, process.stdout.rows));
+            process.stdout.write(renderFullWelcomeScreen({ provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(), usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode, autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer, chatThoughtTime: lastThoughtTime, queuedCommands: commandQueue.items }, process.stdout.rows));
             process.stdout.write(renderChatComposerCursor({ width: process.stdout.columns }, 0));
             continue;
         }
@@ -1228,27 +1317,13 @@ export async function handleChatSession(args, version, signal) {
         }
         if (trimmed === '/help') {
             lastQuestion = trimmed;
-            lastAnswer = 'Available slash commands:\n' +
-                '  /connect  - Choose a Gateway, hosted, or local provider\n' +
-                '  /model     - Switch active AI model\n' +
-                '  /init      - Scaffold AGENTS.md from workspace scan\n' +
-                '  /mcp       - Manage local MCP servers\n' +
-                '  /session   - Create, resume, undo, redo, share, export, or compact sessions\n' +
-                '  /queue     - Add, inspect, or clear queued follow-up commands\n' +
-                '  /workflow  - Inspect Git, build plans, or undo agent changes\n' +
-                '  /review    - Review current workspace changes for bugs\n' +
-                '  /map       - Show a compact workspace file map\n' +
-                '  /budget    - Set a per-task token limit (/budget 5000, /budget off)\n' +
-                '  /skills    - List skills; use /skills on|off NAME|PATTERN (* for all)\n' +
-                '  /clear     - Reset conversation memory\n' +
-                '  /help      - Display commands, shortcuts & version\n' +
-                '  /exit      - Exit Moderado';
+            lastAnswer = 'Available slash commands:\n' + getHelpCommandLines().map((line) => line.includes(' - ') ? line : `\n${line}`).join('\n');
             lastThoughtTime = 0.001;
             lastOutputTokenRate = undefined;
             lastTokenUsage = undefined;
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime,
@@ -1280,7 +1355,7 @@ export async function handleChatSession(args, version, signal) {
             sessionStore.save(activeSession);
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime,
@@ -1300,7 +1375,7 @@ export async function handleChatSession(args, version, signal) {
             sessionStore.save(activeSession);
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                 chatThoughtTime: lastThoughtTime,
@@ -1325,7 +1400,7 @@ export async function handleChatSession(args, version, signal) {
             lastTokenUsage = undefined;
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'Web search', tokens: Math.round(sessionTokens), cost: costLabel(),
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'Web search', tokens: Math.round(sessionTokens), cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                 autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
             }, process.stdout.rows));
@@ -1346,7 +1421,7 @@ export async function handleChatSession(args, version, signal) {
                 sessionStore.save(activeSession);
                 process.stdout.write('\x1b[H\x1b[J');
                 process.stdout.write(renderFullWelcomeScreen({
-                    model: currentModel ?? 'Web search', tokens: Math.round(sessionTokens), cost: costLabel(),
+                    provider: getProviderStatusName(activeConnection), model: currentModel ?? 'Web search', tokens: Math.round(sessionTokens), cost: costLabel(),
                     usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated, workspace: canonicalWorkspace, mode: activeMode,
                     autoApprove: activeAutoApprove, chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                     chatThoughtTime: lastThoughtTime,
@@ -1362,13 +1437,13 @@ export async function handleChatSession(args, version, signal) {
                 drawFrame: (popupLines) => {
                     process.stdout.write('\x1b[H\x1b[J');
                     process.stdout.write(renderWelcomePopupLayer({
-                        model: currentModel ?? 'No model connected — use /connect',
+                        provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect',
                         tokens: Math.round(sessionTokens), cost: costLabel(), workspace: canonicalWorkspace,
                         mode: activeMode, autoApprove: activeAutoApprove,
                         chatQuestion: lastQuestion, chatAnswer: lastAnswer,
                     }, popupLines, process.stdout.columns, process.stdout.rows));
                 },
-                gatewayUrl: config.connections?.['moderado-cloud']?.baseUrl,
+                gatewayUrl: config.connections?.moderado?.baseUrl,
                 activeConnection,
                 savedConnections: config.connections,
                 connectProviders: config.connectProviders,
@@ -1399,14 +1474,26 @@ export async function handleChatSession(args, version, signal) {
             sessionTokens = activeSession.usage.totalTokens;
             usageRecorded = true;
         };
+        let streamedAnswer = '';
         try {
             const requestAbort = new AbortController();
             const requestSignal = signal ? AbortSignal.any([signal, requestAbort.signal]) : requestAbort.signal;
-            let streamedAnswer = '';
+            let lastSessionCheckpointAt = 0;
+            const checkpointInterruptedTurn = () => {
+                activeSession.messages = saveFailedTurnPrompt(conversationHistory, effectivePrompt, streamedAnswer);
+                activeSession.providerId = activeConnection?.id;
+                activeSession.providerName = activeConnection?.displayName;
+                activeSession.modelId = currentModel;
+                activeSession.mode = activeMode;
+                sessionStore.save(activeSession);
+                lastSessionCheckpointAt = Date.now();
+            };
             lastAnswer = '';
             lastTurnDone = false;
             let taskStep;
-            let firstAssistantDeltaAt;
+            let activityLabel;
+            let activityFrame = 0;
+            let turnActive = true;
             let outputTokenRate;
             let budgetWarningIssued = false;
             let budgetStopped = false;
@@ -1414,12 +1501,32 @@ export async function handleChatSession(args, version, signal) {
             const usedModels = new Set();
             let answerPosition = { row: 15, column: getChatAnswerArea(process.stdout.columns).startColumn };
             let lastErrorEvent;
+            // Keep a live spinner on the status row for the whole turn: streaming
+            // text and silent tool execution both leave the screen looking settled.
+            const stopActivity = () => {
+                if (thinkingTimer) {
+                    clearInterval(thinkingTimer);
+                    thinkingTimer = undefined;
+                }
+            };
+            const paintActivity = (glyph, label) => {
+                process.stdout.write(renderChatThoughtTimeUpdate((Date.now() - startedAt) / 1000, taskStep, process.stdout.columns, { glyph, label }));
+            };
+            const startActivity = () => {
+                stopActivity();
+                if (!turnActive)
+                    return;
+                activityLabel = undefined;
+                thinkingTimer = setInterval(() => {
+                    paintActivity(ACTIVITY_SPINNER_FRAMES[activityFrame++ % ACTIVITY_SPINNER_FRAMES.length], activityLabel ?? 'Working');
+                }, 200);
+            };
             const redrawChatFrame = () => {
                 const thoughtTime = Math.max(0.001, (Date.now() - startedAt) / 1000);
                 const displayAnswer = streamedAnswer || lastAnswer;
                 process.stdout.write('\x1b[H\x1b[J');
                 process.stdout.write(renderFullWelcomeScreen({
-                    model: currentModel ?? 'No model connected — use /connect',
+                    provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect',
                     tokens: Math.round(sessionTokens + (usageRecorded ? 0 : turnUsage?.usage.totalTokens ?? 0)),
                     cost: costLabel(),
                     usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated,
@@ -1441,6 +1548,9 @@ export async function handleChatSession(args, version, signal) {
                     const area = getChatAnswerArea(process.stdout.columns);
                     answerPosition = renderChatAnswerDelta(displayAnswer, { row: 15, column: area.startColumn }, area.wrapWidth, area.startColumn);
                 }
+                // The frame paints a static status; keep the live row authoritative while the turn runs.
+                if (thinkingTimer)
+                    paintActivity(ACTIVITY_SPINNER_FRAMES[activityFrame++ % ACTIVITY_SPINNER_FRAMES.length], activityLabel ?? 'Working');
             };
             const onGenerationKeypress = (str, key) => {
                 handleGenerationKeypress(str, key, commandQueue, {
@@ -1456,6 +1566,7 @@ export async function handleChatSession(args, version, signal) {
                     },
                     onBtw: async (command) => {
                         const resume = suspendGenerationInput?.();
+                        stopActivity();
                         const ac = new AbortController();
                         const onBtwKey = (s, k) => {
                             if ((k && k.name === 'escape') || s === '\x1b') {
@@ -1467,7 +1578,7 @@ export async function handleChatSession(args, version, signal) {
                             await executeBtwQuery(command, async (popupLines, frameOpts) => {
                                 process.stdout.write('\x1b[H\x1b[J');
                                 process.stdout.write(renderWelcomePopupLayer({
-                                    model: currentModel ?? 'No model connected',
+                                    provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected',
                                     tokens: Math.round(sessionTokens),
                                     cost: costLabel(),
                                     usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated,
@@ -1496,6 +1607,7 @@ export async function handleChatSession(args, version, signal) {
                         }
                         finally {
                             process.stdin.removeListener('keypress', onBtwKey);
+                            startActivity();
                             redrawChatFrame();
                             resume?.();
                         }
@@ -1503,6 +1615,8 @@ export async function handleChatSession(args, version, signal) {
                 });
             };
             if (process.stdin.isTTY) {
+                const onGenerationResize = () => redrawChatFrame();
+                process.stdout.on('resize', onGenerationResize);
                 const attachGenerationListener = () => {
                     readline.emitKeypressEvents(process.stdin);
                     process.stdin.resume();
@@ -1522,16 +1636,13 @@ export async function handleChatSession(args, version, signal) {
                 };
                 removeGenerationListener = () => {
                     process.stdin.removeListener('keypress', onGenerationKeypress);
+                    process.stdout.removeListener('resize', onGenerationResize);
                     suspendGenerationInput = undefined;
                 };
             }
             // Move to the chat layout before the provider can emit its first event.
             redrawChatFrame();
-            thinkingTimer = setInterval(() => {
-                if (!firstAssistantDeltaAt && !turnUsage) {
-                    process.stdout.write(renderChatThoughtTimeUpdate((Date.now() - startedAt) / 1000, taskStep, process.stdout.columns));
-                }
-            }, 400);
+            startActivity();
             const evidenceTask = liveSearch ? buildSearchAnswerTask(effectivePrompt, liveSearch) : undefined;
             const runAgent = () => loop.run(isReview ? createReviewTask(reviewFiles) : evidenceTask ?? createAgentTask(effectivePrompt, activeMode), {
                 workspaceRoot: canonicalWorkspace, provider: provider, tools: isReview ? createReviewToolRegistry() : tools, approvalHandler, router, policy,
@@ -1543,9 +1654,21 @@ export async function handleChatSession(args, version, signal) {
                         usedModels.add(event.newModelId);
                     if (event.type === 'progress') {
                         taskStep = { step: event.step, maxSteps: event.maxSteps ?? 0 };
-                        if (!firstAssistantDeltaAt)
-                            process.stdout.write(renderChatThoughtTimeUpdate((Date.now() - startedAt) / 1000, taskStep, process.stdout.columns));
+                        activityLabel = event.status;
+                        paintActivity(ACTIVITY_SPINNER_FRAMES[activityFrame++ % ACTIVITY_SPINNER_FRAMES.length], activityLabel ?? 'Working');
+                    }
+                    if (event.type === 'tool_call_initiated') {
+                        activityLabel = `Running ${event.toolName}`;
+                        paintActivity(ACTIVITY_SPINNER_FRAMES[activityFrame++ % ACTIVITY_SPINNER_FRAMES.length], activityLabel);
                     }
+                    if (event.type === 'tool_result') {
+                        activityLabel = undefined;
+                        if (!thinkingTimer)
+                            startActivity();
+                    }
+                    // The approval prompt owns the screen until the tool result arrives.
+                    if (event.type === 'approval_request')
+                        stopActivity();
                     if (event.type === 'gateway_status')
                         process.stdout.write(renderGatewayFallbackUpdate(event, process.stdout.columns || 80));
                     if (event.type === 'usage') {
@@ -1566,12 +1689,9 @@ export async function handleChatSession(args, version, signal) {
                         }
                     }
                     if (event.type === 'assistant_delta') {
-                        if (!firstAssistantDeltaAt && event.delta.trim().length > 0) {
-                            firstAssistantDeltaAt = Date.now();
-                            if (thinkingTimer)
-                                clearInterval(thinkingTimer);
-                        }
                         streamedAnswer += event.delta;
+                        if (Date.now() - lastSessionCheckpointAt >= 1_000)
+                            checkpointInterruptedTurn();
                         const area = getChatAnswerArea(process.stdout.columns);
                         const renderedDelta = renderChatAnswerDelta(event.delta, answerPosition, area.wrapWidth, area.startColumn);
                         answerPosition = renderedDelta;
@@ -1593,27 +1713,24 @@ export async function handleChatSession(args, version, signal) {
                         checkpoints.recordPostWrite(canonicalWorkspace, paths);
                 },
             });
-            let result;
-            try {
-                result = await runAgent();
-            }
-            catch (error) {
-                if (!activeConnection || !isAuthenticationFailure(error))
-                    throw error;
-                if (thinkingTimer)
-                    clearInterval(thinkingTimer);
-                const cloudOAuth = isModeradoCloudOAuthConnection(activeConnection);
-                const cloudWithoutKey = activeConnection.id === 'moderado-cloud' && !activeConnection.apiKey;
-                process.stdout.write(cloudWithoutKey
+            /** Prompt for a replacement credential once, persist it, and activate it. Falsy when the user cancels. */
+            const recoverAuthenticationFailure = async () => {
+                if (!activeConnection)
+                    return undefined;
+                stopActivity();
+                const connection = activeConnection;
+                const moderadoOAuth = isModeradoOAuthConnection(connection);
+                const moderadoWithoutKey = connection.id === 'moderado' && !connection.apiKey;
+                process.stdout.write(moderadoWithoutKey
                     ? '\nThis free Gateway route requires a Moderado website API key. Sign in or enter a key to retry once.\n'
-                    : cloudOAuth
-                        ? '\nModerado Cloud rejected this browser login. Sign in again to retry once.\n'
-                        : `\nThe saved ${activeConnection.displayName} API key was rejected. Enter a replacement key to retry once.\n`);
-                const replacement = cloudOAuth || cloudWithoutKey
-                    ? await loginModeradoCloudInteractive({ signal, gatewayUrl: activeConnection.baseUrl })
-                    : await replaceProviderKeyInteractive(activeConnection, { signal });
+                    : moderadoOAuth
+                        ? '\nModerado rejected this browser login. Sign in again to retry once.\n'
+                        : `\nThe saved ${connection.displayName} API key was rejected. Enter a replacement key to retry once.\n`);
+                const replacement = moderadoOAuth || moderadoWithoutKey
+                    ? await loginModeradoInteractive({ signal, gatewayUrl: connection.baseUrl })
+                    : await replaceProviderKeyInteractive(connection, { signal });
                 if (!replacement)
-                    throw error;
+                    return undefined;
                 recordTurnUsage();
                 turnUsage = undefined;
                 usageRecorded = false;
@@ -1621,8 +1738,37 @@ export async function handleChatSession(args, version, signal) {
                 saveConnection(secured);
                 config = loadConfig();
                 activateConnection(secured);
+                startActivity();
+                return secured;
+            };
+            let result;
+            try {
+                checkpointInterruptedTurn();
                 result = await runAgent();
+                // The agent loop reports provider failures as error events and resolves
+                // with a failed result instead of throwing, so key re-entry must be
+                // offered here too. Retry once; a second failure falls through to the
+                // formatted error answer.
+                if (turnFailedAuthentication(result.status, lastErrorEvent) && activeConnection && await recoverAuthenticationFailure()) {
+                    lastErrorEvent = undefined;
+                    result = await runAgent();
+                }
             }
+            catch (error) {
+                if (!activeConnection || !isAuthenticationFailure(error))
+                    throw error;
+                if (!(await recoverAuthenticationFailure()))
+                    throw error;
+                result = await runAgent();
+            }
+            // The turn is over: replace the spinner with the settled state while
+            // usage accounting and catalog refresh finish.
+            turnActive = false;
+            stopActivity();
+            process.stdout.write(renderChatThoughtTimeUpdate((Date.now() - startedAt) / 1000, undefined, process.stdout.columns, {
+                glyph: result.status === 'completed' ? '✔' : '✖',
+                label: result.status === 'completed' ? 'Done' : result.status === 'cancelled' ? 'Stopped' : 'Failed',
+            }));
             if (turnUsage) {
                 let pricing;
                 const modelIds = new Set([...usedModels, result.selectedModel.id]);
@@ -1670,7 +1816,7 @@ export async function handleChatSession(args, version, signal) {
                 streamedAnswer = lastAnswer;
             }
             else if (resolution.isError) {
-                conversationHistory = saveFailedTurnPrompt(conversationHistory, effectivePrompt);
+                conversationHistory = saveFailedTurnPrompt(conversationHistory, effectivePrompt, streamedAnswer);
                 activeSession.messages = conversationHistory;
                 activeSession.providerId = activeConnection?.id;
                 activeSession.providerName = activeConnection?.displayName;
@@ -1694,14 +1840,16 @@ export async function handleChatSession(args, version, signal) {
             lastQuestion = trimmed;
             lastThoughtTime = Math.max(0.001, (Date.now() - startedAt) / 1000);
             lastOutputTokenRate = outputTokenRate;
-            if (thinkingTimer)
-                clearInterval(thinkingTimer);
+            turnActive = false;
+            stopActivity();
             removeGenerationListener?.();
             redrawChatFrame();
         }
         catch (err) {
-            if (thinkingTimer)
+            if (thinkingTimer) {
                 clearInterval(thinkingTimer);
+                thinkingTimer = undefined;
+            }
             removeGenerationListener?.();
             recordTurnUsage();
             if (turnUsage)
@@ -1711,7 +1859,7 @@ export async function handleChatSession(args, version, signal) {
                 ? ` (other configured providers: ${otherConnections.join(', ')})`
                 : '';
             lastQuestion = trimmed;
-            conversationHistory = saveFailedTurnPrompt(conversationHistory, effectivePrompt);
+            conversationHistory = saveFailedTurnPrompt(conversationHistory, effectivePrompt, streamedAnswer);
             activeSession.messages = conversationHistory;
             activeSession.providerId = activeConnection?.id;
             activeSession.providerName = activeConnection?.displayName;
@@ -1727,7 +1875,7 @@ export async function handleChatSession(args, version, signal) {
             lastOutputTokenRate = undefined;
             process.stdout.write('\x1b[H\x1b[J');
             process.stdout.write(renderFullWelcomeScreen({
-                model: currentModel ?? 'No model connected — use /connect',
+                provider: getProviderStatusName(activeConnection), model: currentModel ?? 'No model connected — use /connect',
                 tokens: Math.round(sessionTokens),
                 cost: costLabel(),
                 usageAvailable: activeSession.usage.available, usageEstimated: activeSession.usage.estimated,