crewly 1.2.1 → 1.2.3

package/dist/backend/backend/src/services/agent/agent-registration.service.js

@@ -14,6 +14,8 @@ import { ContextWindowMonitorService } from './context-window-monitor.service.js
 import { SubAgentMessageQueue } from '../messaging/sub-agent-message-queue.service.js';
 import { AgentSuspendService } from './agent-suspend.service.js';
 import { stripAnsiCodes } from '../../utils/terminal-output.utils.js';
+import { isPromptLine, containsSpinnerOrWorkingIndicator, containsProcessingIndicator, containsBusyStatusBar, containsRewindMode, containsGeminiProcessingKeywords, extractChatPrefix, stripTuiLineBorders, matchTuiPromptLine, } from '../../utils/terminal-string-ops.js';
+import { PtyActivityTrackerService } from './pty-activity-tracker.service.js';
 /**
  * Service responsible for the complex, multi-step process of agent initialization and registration.
  * Isolates the complex state management of agent startup with progressive escalation.
@@ -47,16 +49,10 @@ export class AgentRegistrationService {
     // which causes multiple Ctrl+C presses that can crash the runtime.
     sessionDeliveryMutex = new Map();
     // Terminal patterns are now centralized in TERMINAL_PATTERNS constant
-    // Keeping these as static getters for backwards compatibility within the class
+    // Keeping prompt chars as static getter for backwards compatibility within the class
     static get CLAUDE_PROMPT_INDICATORS() {
         return TERMINAL_PATTERNS.PROMPT_CHARS;
     }
-    static get CLAUDE_PROMPT_STREAM_PATTERN() {
-        return TERMINAL_PATTERNS.PROMPT_STREAM;
-    }
-    static get CLAUDE_PROCESSING_INDICATORS() {
-        return TERMINAL_PATTERNS.PROCESSING_INDICATORS;
-    }
     constructor(_legacyTmuxService, // Legacy parameter for backwards compatibility
     projectRoot, storageService) {
         this.logger = LoggerService.getInstance().createComponentLogger('AgentRegistrationService');
@@ -1508,6 +1504,14 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
             await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_SESSION_NAME, sessionName);
             await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_ROLE, role);
             await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.CREWLY_API_URL, `http://localhost:${WEB_CONSTANTS.PORTS.BACKEND}`);
+            // Pass Gemini API key to gemini-cli agents so they authenticate
+            // with the paid API key instead of the free-tier Google login.
+            if (runtimeType === RUNTIME_TYPES.GEMINI_CLI) {
+                const geminiApiKey = process.env[ENV_CONSTANTS.GEMINI_API_KEY];
+                if (geminiApiKey) {
+                    await sessionHelper.setEnvironmentVariable(sessionName, ENV_CONSTANTS.GEMINI_API_KEY, geminiApiKey);
+                }
+            }
             this.logger.info('Agent session created and environment variables set, initializing with registration', {
                 sessionName,
                 role,
@@ -1701,9 +1705,9 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
             if (!delivered) {
                 // Check if the agent is actively processing (busy) — the queue
                 // processor can re-queue instead of permanently failing the message.
-                const busyOutput = sessionHelper.capturePane(sessionName).slice(-2000);
-                const isBusy = TERMINAL_PATTERNS.BUSY_STATUS_BAR.test(busyOutput) ||
-                    TERMINAL_PATTERNS.PROCESSING_WITH_TEXT.test(busyOutput);
+                // Use PTY idle time instead of regex patterns for robust cross-runtime detection.
+                const idleMs = PtyActivityTrackerService.getInstance().getIdleTimeMs(sessionName);
+                const isBusy = idleMs < SESSION_COMMAND_DELAYS.AGENT_BUSY_IDLE_THRESHOLD_MS;
                 return {
                     success: false,
                     error: isBusy
@@ -1768,9 +1772,6 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
         if (!session) {
             return false;
         }
-        // Use runtime-specific pattern for stream detection to avoid false positives
-        // (e.g. Gemini's `> ` pattern matching markdown blockquotes in Claude Code output)
-        const streamPattern = this.getPromptPatternForRuntime(runtimeType);
         return new Promise((resolve) => {
             let resolved = false;
             const pollInterval = EVENT_DELIVERY_CONSTANTS.AGENT_READY_POLL_INTERVAL;
@@ -1817,9 +1818,11 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                 if (resolved)
                     return;
                 // Strip ANSI escape sequences before testing — raw PTY data contains
-                // cursor positioning, color codes, etc. that break regex matching (#106)
+                // cursor positioning, color codes, etc. that break pattern matching (#106)
                 const cleanData = stripAnsiCodes(data);
-                if (streamPattern.test(cleanData)) {
+                // Check each line for prompt pattern (string-based, no regex)
+                const hasPromptInStream = cleanData.split('\n').some(line => line.trim().length > 0 && isPromptLine(line, runtimeType));
+                if (hasPromptInStream) {
                     // Double-check with capturePane to avoid false positives from partial data
                     const output = sessionHelper.capturePane(sessionName);
                     if (this.isClaudeAtPrompt(output, runtimeType)) {
@@ -2052,7 +2055,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                 }
                 // Phase 1: Wait for Claude to be at prompt before sending
                 if (!messageSent) {
-                    const isAtPrompt = AgentRegistrationService.CLAUDE_PROMPT_STREAM_PATTERN.test(buffer);
+                    const isAtPrompt = buffer.split('\n').some(line => line.trim().length > 0 && isPromptLine(line));
                     if (isAtPrompt) {
                         sendMessageNow();
                     }
@@ -2063,9 +2066,9 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                     return; // Wait for Enter to be sent
                 }
                 // Look for processing indicators confirming delivery
-                const hasProcessingIndicator = AgentRegistrationService.CLAUDE_PROCESSING_INDICATORS.some((pattern) => pattern.test(buffer));
+                const hasProcessingIndicator = containsProcessingIndicator(buffer);
                 // Also check if prompt disappeared (Claude is working)
-                const promptStillVisible = AgentRegistrationService.CLAUDE_PROMPT_STREAM_PATTERN.test(buffer);
+                const promptStillVisible = buffer.split('\n').some(line => line.trim().length > 0 && isPromptLine(line));
                 // Use constant for minimum buffer check (P3.2 fix)
                 if (hasProcessingIndicator || (!promptStillVisible && buffer.length > EVENT_DELIVERY_CONSTANTS.MIN_BUFFER_FOR_PROCESSING_DETECTION)) {
                     this.logger.debug('Message delivery confirmed (event-driven)', {
@@ -2114,12 +2117,10 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                 if (!this.isClaudeAtPrompt(output, runtimeType)) {
                     if (attempt === maxAttempts) {
                         // On the final attempt, check if the agent is DEFINITELY busy
-                        // before force-delivering. If we see "esc to interrupt" or
-                        // processing indicators, the agent is actively working and
-                        // force-delivery risks corrupting its current task.
-                        const tailForBusyCheck = output.slice(-2000);
-                        const isBusy = TERMINAL_PATTERNS.BUSY_STATUS_BAR.test(tailForBusyCheck) ||
-                            TERMINAL_PATTERNS.PROCESSING_WITH_TEXT.test(tailForBusyCheck);
+                        // before force-delivering. Use PTY idle time for robust
+                        // cross-runtime detection instead of fragile regex patterns.
+                        const idleMs = PtyActivityTrackerService.getInstance().getIdleTimeMs(sessionName);
+                        const isBusy = idleMs < SESSION_COMMAND_DELAYS.AGENT_BUSY_IDLE_THRESHOLD_MS;
                         if (isBusy) {
                             this.logger.warn('Agent is busy (processing indicators detected), skipping force delivery', {
                                 sessionName,
@@ -2182,8 +2183,30 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                 //    for focus cycling and overlay dismissal.
                 if (isClaudeCode) {
                     if (attempt > 1) {
+                        // On retry: Ctrl+C to cancel stale input, then PTY resize to force
+                        // TUI re-render (SIGWINCH), then Tab to cycle Ink focus.
                         await sessionHelper.sendCtrlC(sessionName);
                         await delay(300);
+                        try {
+                            const session = sessionHelper.getSession(sessionName);
+                            if (session) {
+                                session.resize(81, 25);
+                                await delay(200);
+                                session.resize(80, 24);
+                                await delay(300);
+                            }
+                        }
+                        catch { /* non-fatal */ }
+                        await sessionHelper.sendKey(sessionName, 'Tab');
+                        await delay(300);
+                    }
+                    else {
+                        // First attempt: lightweight focus nudge — Tab key cycles Ink's
+                        // focusNext() to ensure the InputPrompt is active. This prevents
+                        // the "write succeeds but TUI ignores it" failure mode without
+                        // the overhead of Ctrl+C or PTY resize.
+                        await sessionHelper.sendKey(sessionName, 'Tab');
+                        await delay(200);
                     }
                 }
                 else {
@@ -2318,7 +2341,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                         //    definitive proof the agent accepted and is working.
                         //    Only check spinner/⏺ chars — NOT text words like
                         //    "thinking" which appear in historical response text.
-                        if (TERMINAL_PATTERNS.PROCESSING.test(currentOutput)) {
+                        if (containsSpinnerOrWorkingIndicator(currentOutput)) {
                             this.logger.debug('Processing indicators detected — message accepted', {
                                 sessionName,
                                 attempt,
@@ -2353,7 +2376,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                                 await delay(SESSION_COMMAND_DELAYS.MESSAGE_PROCESSING_DELAY);
                                 // Verify recovery
                                 const postEnterOutput = sessionHelper.capturePane(sessionName);
-                                if (TERMINAL_PATTERNS.PROCESSING.test(postEnterOutput) ||
+                                if (containsSpinnerOrWorkingIndicator(postEnterOutput) ||
                                     !this.isClaudeAtPrompt(postEnterOutput, runtimeType)) {
                                     this.logger.info('Enter recovery from prompt successful', {
                                         sessionName,
@@ -2400,7 +2423,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                             await delay(SESSION_COMMAND_DELAYS.MESSAGE_PROCESSING_DELAY);
                             // Verify recovery: check if processing started
                             const recoveryOutput = sessionHelper.capturePane(sessionName);
-                            if (TERMINAL_PATTERNS.PROCESSING.test(recoveryOutput)) {
+                            if (containsSpinnerOrWorkingIndicator(recoveryOutput)) {
                                 this.logger.info('Enter recovery successful — processing started', {
                                     sessionName,
                                     attempt,
@@ -2487,9 +2510,9 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                                 .filter((line) => !beforeLines.has(line))
                                 .join('\n');
                         }
-                        const hasProcessingIndicators = TERMINAL_PATTERNS.PROCESSING_WITH_TEXT.test(newContent || afterOutput.slice(-500));
+                        const hasProcessingIndicators = containsProcessingIndicator(newContent || afterOutput.slice(-500));
                         const hasGeminiIndicators = newContent.length > 0
-                            && /reading|thinking|processing|analyzing|generating|searching/i.test(newContent);
+                            && containsGeminiProcessingKeywords(newContent);
                         const significantLengthChange = Math.abs(lengthDiff) > 10;
                         // For Gemini CLI, contentChanged alone is sufficient evidence of
                         // delivery. The TUI redraws minimally (lengthDiff can be as low as
@@ -2568,19 +2591,11 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                 }
             }
         }
-        // Verification failed, but the message was physically written to the PTY.
-        // If the session is still alive, the agent will likely process it — return
-        // true to avoid false "Failed to deliver" errors shown to users (#99).
-        const backend = getSessionBackendSync();
-        const childAlive = backend?.isChildProcessAlive?.(sessionName);
-        if (childAlive !== false) {
-            this.logger.warn('Message delivery verification inconclusive but session alive — assuming success', {
-                sessionName,
-                maxAttempts,
-                messageLength: message.length,
-            });
-            return true;
-        }
+        this.logger.warn('Message delivery verification failed — all attempts exhausted', {
+            sessionName,
+            maxAttempts,
+            messageLength: message.length,
+        });
         this.logger.error('Message delivery failed after all retry attempts', {
             sessionName,
             maxAttempts,
@@ -2608,13 +2623,13 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
             }
             // Extract a search token from the message:
             // Strip [CHAT:uuid] prefix if present, then take the first 40 chars
-            const chatPrefixMatch = message.match(/^\[CHAT:[^\]]+\]\s*/);
-            const contentAfterPrefix = chatPrefixMatch
-                ? message.slice(chatPrefixMatch[0].length)
+            const { prefixLength } = extractChatPrefix(message);
+            const contentAfterPrefix = prefixLength > 0
+                ? message.slice(prefixLength)
                 : message;
             const searchToken = contentAfterPrefix.slice(0, 40).trim();
             // Also use [CHAT: as a secondary token if message has a CHAT prefix
-            const chatToken = chatPrefixMatch ? '[CHAT:' : null;
+            const chatToken = prefixLength > 0 ? '[CHAT:' : null;
             // Check last 20 non-empty lines for either token.
             // Gemini CLI TUI has status bars at the bottom (branch, sandbox, model info)
             // that push input content further up. 5 lines was insufficient.
@@ -2622,7 +2637,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
             const linesToCheck = lines.slice(-20);
             const isStuck = linesToCheck.some((line) => {
                 // Strip TUI box-drawing borders before checking (Gemini CLI wraps content in │...│)
-                const stripped = line.replace(/^[│┃║|\s]+/, '').replace(/[│┃║|\s]+$/, '');
+                const stripped = stripTuiLineBorders(line);
                 if (searchToken && (line.includes(searchToken) || stripped.includes(searchToken)))
                     return true;
                 if (chatToken && (line.includes(chatToken) || stripped.includes(chatToken)))
@@ -2672,9 +2687,9 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                 return false;
             // Extract search token: first 30 chars of the message content
             // (after stripping any [CHAT:uuid] prefix)
-            const chatPrefixMatch = message.match(/^\[CHAT:[^\]]+\]\s*/);
-            const contentAfterPrefix = chatPrefixMatch
-                ? message.slice(chatPrefixMatch[0].length)
+            const { prefixLength } = extractChatPrefix(message);
+            const contentAfterPrefix = prefixLength > 0
+                ? message.slice(prefixLength)
                 : message;
             const searchToken = contentAfterPrefix.slice(0, 30).trim();
             if (!searchToken)
@@ -2684,14 +2699,13 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
             // Or without borders: > text
             // The prompt line is the line with `> ` followed by actual content.
             const lines = output.split('\n');
-            const promptLineRegex = /^[│┃║|\s]*>\s+(.+)/;
             for (let i = lines.length - 1; i >= Math.max(0, lines.length - 25); i--) {
                 const line = lines[i];
-                const match = line.match(promptLineRegex);
-                if (match) {
-                    const promptContent = match[1].replace(/[│┃║|\s]+$/, '').trim();
+                const promptContent = matchTuiPromptLine(line);
+                if (promptContent !== null) {
+                    const trimmedContent = stripTuiLineBorders(promptContent).trim();
                     // Check if the prompt line content contains our message text
-                    if (promptContent.length > 5 && promptContent.includes(searchToken)) {
+                    if (trimmedContent.length > 5 && trimmedContent.includes(searchToken)) {
                         this.logger.warn('Text stuck at TUI prompt — Enter was not pressed', {
                             sessionName,
                             searchToken: searchToken.slice(0, 20),
@@ -2797,7 +2811,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
         for (const sessionName of sessionHelper.listSessions()) {
             try {
                 const output = sessionHelper.capturePane(sessionName);
-                if (TERMINAL_PATTERNS.REWIND_MODE.test(output)) {
+                if (containsRewindMode(output)) {
                     this.logger.warn('Rewind mode detected, sending q to exit', { sessionName });
                     sessionHelper.writeRaw(sessionName, 'q');
                     await delay(500);
@@ -2820,19 +2834,19 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                     continue;
                 // Look for any text sitting on the prompt line
                 const lines = output.split('\n');
-                const promptLineRegex = /^[│┃║|\s]*>\s+(.+)/;
                 for (let i = lines.length - 1; i >= Math.max(0, lines.length - 25); i--) {
-                    const match = lines[i].match(promptLineRegex);
-                    if (match) {
-                        const promptContent = match[1].replace(/[│┃║|\s]+$/, '').trim();
+                    const promptContent = matchTuiPromptLine(lines[i]);
+                    if (promptContent !== null) {
+                        const trimmedContent = stripTuiLineBorders(promptContent).trim();
                         // Only act on substantial text (> 10 chars) to avoid false positives
                         // from TUI rendering artifacts or short status text
-                        if (promptContent.length > 10) {
+                        if (trimmedContent.length > 10) {
                             // Skip known Gemini CLI idle placeholder text that sits at
                             // the `> ` prompt when no user input is present. These are
                             // NOT stuck messages — they are TUI decoration.
-                            const isPlaceholder = /^Type your message/i.test(promptContent) ||
-                                /^@[\w/.]+/.test(promptContent); // e.g., "@path/to/file"
+                            const lowerContent = trimmedContent.toLowerCase();
+                            const isPlaceholder = lowerContent.startsWith('type your message') ||
+                                trimmedContent.startsWith('@'); // e.g., "@path/to/file"
                             if (isPlaceholder) {
                                 break;
                             }
@@ -2927,8 +2941,8 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
      */
     trackSentMessage(sessionName, message) {
         // Extract a search snippet: skip [CHAT:uuid] prefix, take first 80 chars
-        const prefixMatch = message.match(/^\[CHAT:[^\]]+\]\s*/);
-        const contentStart = prefixMatch ? prefixMatch[0].length : 0;
+        const { prefixLength } = extractChatPrefix(message);
+        const contentStart = prefixLength;
         // Normalize whitespace: messages may contain \n from enhanced templates.
         // Terminal bottom text is join(' '), so \n in snippet would never match.
         const snippet = message.slice(contentStart, contentStart + 80).replace(/\s+/g, ' ').trim();
@@ -2942,27 +2956,10 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
         // Ensure the background scanner is running
         this.startStuckMessageDetector();
     }
-    /**
-     * Get the runtime-specific prompt regex pattern.
-     * Avoids false positives by using narrow patterns when runtime is known.
-     *
-     * @param runtimeType - The runtime type (claude-code, gemini-cli, etc.)
-     * @returns The appropriate prompt detection regex
-     */
-    getPromptPatternForRuntime(runtimeType) {
-        if (runtimeType === RUNTIME_TYPES.CLAUDE_CODE)
-            return TERMINAL_PATTERNS.CLAUDE_CODE_PROMPT;
-        if (runtimeType === RUNTIME_TYPES.GEMINI_CLI)
-            return TERMINAL_PATTERNS.GEMINI_CLI_PROMPT;
-        if (runtimeType === RUNTIME_TYPES.CODEX_CLI)
-            return TERMINAL_PATTERNS.CODEX_CLI_PROMPT;
-        return TERMINAL_PATTERNS.PROMPT_STREAM;
-    }
     /**
      * Check if the agent appears to be at an input prompt.
-     * Looks for prompt indicators (❯, ⏵, $, ❯❯, ⏵⏵) in terminal output.
-     * Also checks for busy indicators (esc to interrupt, spinners, ⏺)
-     * to avoid false negatives when the agent is processing.
+     * Delegates to the regex-free isAgentAtPrompt() from terminal-string-ops,
+     * with additional logging and per-line prompt detection using isPromptLine().
      *
      * @param terminalOutput - The terminal output to check
      * @param runtimeType - The runtime type for pattern selection
@@ -2979,57 +2976,12 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
         // Use 5000 chars to accommodate large tool outputs that push the prompt
         // further back in the buffer (#106).
         const tailSection = terminalOutput.slice(-5000);
-        const isGemini = runtimeType === RUNTIME_TYPES.GEMINI_CLI;
-        const isClaudeCode = runtimeType === RUNTIME_TYPES.CLAUDE_CODE;
-        const isCodex = runtimeType === RUNTIME_TYPES.CODEX_CLI;
-        const streamPattern = this.getPromptPatternForRuntime(runtimeType);
         // Check for prompt FIRST. Processing indicators like "thinking" or "analyzing"
         // can appear in the agent's previous response text and persist in the terminal
         // scroll buffer, causing false negatives if checked before the prompt.
-        if (streamPattern.test(tailSection)) {
-            return true;
-        }
-        // Fallback: check last several lines for prompt indicators.
-        // The prompt may not be on the very last line due to status bars,
-        // notifications, or terminal wrapping below the prompt.
         const lines = tailSection.split('\n').filter((line) => line.trim().length > 0);
         const linesToCheck = lines.slice(-10);
-        const hasPrompt = linesToCheck.some((line) => {
-            const trimmed = line.trim();
-            // Strip TUI box-drawing borders that Gemini CLI and other TUI frameworks
-            // wrap around prompts. Covers full Unicode box-drawing range (#106).
-            const stripped = trimmed
-                .replace(/^[\u2500-\u257F|+\-═║╭╮╰╯]+\s*/, '')
-                .replace(/\s*[\u2500-\u257F|+\-═║╭╮╰╯]+$/, '');
-            // Claude Code prompts: ❯, ⏵, $ alone on a line
-            if (!isGemini && !isCodex) {
-                if (['❯', '⏵', '$'].some(ch => trimmed === ch || stripped === ch)) {
-                    return true;
-                }
-                // ❯❯ = bypass permissions prompt (idle).
-                // Matches "❯❯", "❯❯ ", and "❯❯ bypass permissions on (shift+tab to cycle)".
-                // Note: ⏵⏵ appears in the status bar but is visible both when idle AND
-                // busy, so it cannot be used as a reliable prompt indicator.
-                if (trimmed.startsWith('❯❯')) {
-                    return true;
-                }
-            }
-            // Gemini CLI prompts: > or ! followed by space
-            if (!isClaudeCode) {
-                if (isCodex) {
-                    // Codex prompt uses `›`; avoid plain `> ` to prevent false-positives
-                    // from markdown blockquotes in agent output.
-                    if (trimmed.startsWith('› ') || stripped.startsWith('› ')) {
-                        return true;
-                    }
-                }
-                else if (trimmed.startsWith('> ') || trimmed.startsWith('! ') ||
-                    stripped.startsWith('> ') || stripped.startsWith('! ')) {
-                    return true;
-                }
-            }
-            return false;
-        });
+        const hasPrompt = linesToCheck.some(line => isPromptLine(line, runtimeType));
         if (hasPrompt) {
             return true;
         }
@@ -3037,14 +2989,14 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
         // Check last 10 lines (not just 5) because tool output can push processing
         // indicators further up while the status bar stays at the bottom.
         const recentLines = linesToCheck.join('\n');
-        if (TERMINAL_PATTERNS.PROCESSING_WITH_TEXT.test(recentLines)) {
+        if (containsProcessingIndicator(recentLines)) {
             this.logger.debug('Processing indicators present near bottom of output');
             return false;
         }
         // Check for "esc to interrupt" in the status bar — this is a definitive
         // busy signal. Claude Code only shows this text while actively processing.
         // It disappears when the agent returns to idle at the prompt.
-        if (TERMINAL_PATTERNS.BUSY_STATUS_BAR.test(recentLines)) {
+        if (containsBusyStatusBar(recentLines)) {
             this.logger.debug('Busy status bar detected (esc to interrupt)');
             return false;
         }
@@ -3302,7 +3254,7 @@ After checking in, just say "Ready for tasks" and wait for me to send you work.`
                             return false;
                         const currentOutput = sessionHelper.capturePane(sessionName);
                         // Processing indicators (spinners) = definitive success
-                        if (TERMINAL_PATTERNS.PROCESSING.test(currentOutput)) {
+                        if (containsSpinnerOrWorkingIndicator(currentOutput)) {
                             this.logger.debug('Kickoff delivered — processing indicators detected', {
                                 sessionName, checkIndex: i,
                             });