commandmate 0.3.5 → 0.4.0

package/README.md

@@ -10,45 +10,70 @@
 [English](./README.md) | [日本語](./docs/ja/README.md)
 
 <p align="center">
-  <img src="./docs/images/demo-mobile.gif" alt="CommandMate mobile demo" width="300">
+  <img src="./docs/images/demo-desktop.gif" width="600" alt="CommandMate Desktop Demo" />
 </p>
 
-> **Claude Code keeps coding while you're away. Check in from your phone.**
+> **Move issues forward, not terminal tabs.**
 
-Not a "remote control" — a **mobile dev cockpit**.
+CommandMate is an IDE for issue-driven AI development.
 
 ```bash
 npx commandmate
 ```
 
-**From install to mobile monitoring in 60 seconds.** macOS / Linux · Node.js v20+ · npm · git · tmux · openssl
+**From install to your first session in 60 seconds.** macOS / Linux · Node.js v20+ · npm · git · tmux
 
 ---
 
-AI writes code for you — but you're stuck watching a terminal, afraid to walk away.
-Close the lid and every session dies.
-**CommandMate keeps it alive, and puts the controls on your phone.**
+Instead of jumping straight into implementation, you define an issue, refine it with AI, review the direction, generate a plan, and then let your coding agent execute. CommandMate helps you run multiple issues in parallel with Git worktrees, choose the right agent for each issue, and keep work moving even when you leave your desk.
 
-Of course, it works great on desktop too — the two-column layout gives you a full overview of all sessions and worktrees at a glance.
+If your workflow is shifting from "writing code yourself" to "defining issues, reviewing direction, and accepting outcomes," CommandMate can become the center of your development workflow.
 
 <p align="center">
-  <img src="./docs/images/demo-desktop.gif" alt="CommandMate desktop demo" width="600">
+  <img src="./docs/images/demo-mobile.gif" width="300" alt="CommandMate Mobile Demo" />
 </p>
 
+Works on desktop and mobile — monitor and steer sessions from any browser, including your phone.
+
+---
+
+## Issue-Driven Development
+
+CommandMate recommends the following development method. By adopting this process, humans can focus on defining issues and verifying final outputs.
+
+```
+Define Issue → Refine with AI → Review Direction → Generate Plan → Agent Executes
+```
+
+| Step | Command | What happens |
+|------|---------|-------------|
+| Refine the issue | `/issue-enhance` | AI asks clarifying questions and fills in missing details |
+| Review the issue | `/multi-stage-issue-review` | Multi-stage review (consistency, impact scope) with automated fixes |
+| Review the design | `/multi-stage-design-review` | 4-stage review (general → consistency → impact → security) |
+| Plan the work | `/work-plan` | Generates a task breakdown with dependencies |
+| Implement via TDD | `/tdd-impl` | Red-Green-Refactor cycle, automated |
+| Verify acceptance | `/acceptance-test` | Validates all acceptance criteria from the issue |
+| Create the PR | `/create-pr` | Auto-generates title, description, and labels |
+| Dev (full) | `/pm-auto-dev` | TDD implementation → acceptance test → refactoring → progress report |
+| Issue → Dev (full) | `/pm-auto-issue2dev` | Issue review → design review → work plan → TDD → acceptance test → refactoring → progress report |
+| Design → Dev (full) | `/pm-auto-design2dev` | Design review → work plan → TDD → acceptance test → refactoring → progress report |
+
+For details, see the [issues](https://github.com/Kewton/CommandMate/issues), [dev reports](./dev-reports/issue/), and [workflow examples](./docs/en/user-guide/workflow-examples.md) in the CommandMate repository.
+
 ---
 
 ## Key Features
 
 | Feature | What it does | Why it matters |
 |---------|-------------|----------------|
-| **Auto Yes Mode** | Agent runs without stopping for confirmations | No babysitting — Claude Code keeps working while you're away |
-| **Git Worktree Sessions** | One session per worktree, parallel execution | Multiple tasks progress simultaneously |
-| **Mobile Web UI** | Full session control from any browser | Monitor and steer from your phone |
-| **File Viewer** | Browse worktree files from the browser | Review code changes without touching your PC |
-| **Markdown Editor** | Edit Markdown files in the browser | Update AI instructions on the go |
+| **Git Worktree Sessions** | One session per worktree, parallel execution | Multiple issues progress simultaneously without interference |
+| **Multi-Agent Support** | Choose Claude Code, Codex, Gemini, or local models per issue | Pick the right agent for each task |
+| **Auto Yes Mode** | Agent runs without stopping for confirmations | No babysitting — the agent keeps working while you're away |
+| **Web UI (Desktop & Mobile)** | Full session control from any browser | Monitor and steer from your desk or your phone |
+| **File Viewer & Markdown Editor** | Browse and edit worktree files in the browser | Review changes and update AI instructions without opening an IDE |
 | **Screenshot Instructions** | Attach images to your prompts | Snap a bug → "Fix this" — the agent sees the screenshot |
+| **Scheduled Execution** | Cron-based auto-run via CMATE.md | Daily reviews, nightly tests — agents work on a schedule |
 | **Token Authentication** | SHA-256 hashed token + HTTPS + rate limiting | Secure remote access — no credentials leaked, brute-force protected |
-| **Scheduled Execution** | Cron-based auto-run via CMATE.md | Daily reviews, nightly tests — Claude Code works on a schedule |
 
 ---
 
@@ -56,48 +81,11 @@ Of course, it works great on desktop too — the two-column layout gives you a f
 
 | Scenario | How CommandMate helps |
 |----------|----------------------|
-| **Couch coding** | Start a task on your PC, then monitor and steer from the sofa |
-| **Commute review** | Review AI-generated code changes on the train |
-| **Overnight runs** | Let Claude Code work all night — check progress from bed |
+| **Parallel issue development** | Run multiple issues in separate worktrees, each with its own agent session |
+| **Issue refinement** | Define an issue, let AI fill gaps, review before any code is written |
+| **Overnight execution** | Queue issues with scheduled execution — check progress in the morning |
+| **Mobile review** | Review AI-generated changes and steer direction from your phone |
 | **Visual bug fix** | Snap a UI bug on your phone, send it with "Fix this" |
-| **Parallel tasks** | Run multiple worktree sessions, manage them all from one dashboard |
-
----
-
-## Comparison
-
-| Feature | CommandMate | Remote Control (Official) | Happy Coder | claude-squad | Omnara |
-|---------|:-----------:|:------------------------:|:-----------:|:------------:|:------:|
-| Auto Yes Mode | Yes | No | No | Yes (TUI only) | No |
-| Git Worktree Management | Yes | No | No | Yes (TUI only) | No |
-| Parallel Sessions | Yes | **No (1 only)** | Yes | Yes | No |
-| Mobile Web UI | Yes | Yes (claude.ai) | Yes | **No** | Yes |
-| File Viewer | Yes | No | No | No | No |
-| Markdown Editor | Yes | No | No | No | No |
-| Screenshot Instructions | Yes | No | No | Not possible | No |
-| Scheduled Execution | Yes | No | No | No | No |
-| Survives Laptop Close | Yes (daemon) | **No (terminal must stay open)** | Yes | Yes | Yes |
-| Token Authentication | Yes | N/A (Anthropic account) | N/A (app) | No | N/A (cloud) |
-| Free / OSS | Yes | Requires Pro/Max | Free + Paid | Yes | $20/mo |
-| Runs 100% Locally | Yes | Via Anthropic API | Server-routed | Yes | Cloud fallback |
-
----
-
-## Screenshots
-
-### Desktop
-
-![Desktop view](./docs/images/screenshot-desktop.png)
-
-### Mobile
-
-| Top Page | Worktree (History) | Worktree (Terminal) |
-|----------|-------------------|-------------------|
-| ![Mobile view](./docs/images/screenshot-mobile.png) | ![Mobile - History](./docs/images/screenshot-worktree-mobile.png) | ![Mobile - Terminal](./docs/images/screenshot-worktree-mobile-terminal.png) |
-
-### Worktree Detail (Desktop)
-
-![Desktop - Worktree detail](./docs/images/screenshot-worktree-desktop.png)
 
 ---
 
@@ -302,6 +290,28 @@ npm start
 
 ---
 
+<details>
+<summary><strong>Comparison</strong></summary>
+
+| Feature | CommandMate | Remote Control (Official) | Happy Coder | claude-squad | Omnara |
+|---------|:-----------:|:------------------------:|:-----------:|:------------:|:------:|
+| Auto Yes Mode | Yes | No | No | Yes (TUI only) | No |
+| Git Worktree Management | Yes | No | No | Yes (TUI only) | No |
+| Parallel Sessions | Yes | **No (1 only)** | Yes | Yes | No |
+| Mobile Web UI | Yes | Yes (claude.ai) | Yes | **No** | Yes |
+| File Viewer | Yes | No | No | No | No |
+| Markdown Editor | Yes | No | No | No | No |
+| Screenshot Instructions | Yes | No | No | Not possible | No |
+| Scheduled Execution | Yes | No | No | No | No |
+| Survives Laptop Close | Yes (daemon) | **No (terminal must stay open)** | Yes | Yes | Yes |
+| Token Authentication | Yes | N/A (Anthropic account) | N/A (app) | No | N/A (cloud) |
+| Free / OSS | Yes | Requires Pro/Max | Free + Paid | Yes | $20/mo |
+| Runs 100% Locally | Yes | Via Anthropic API | Server-routed | Yes | Cloud fallback |
+
+</details>
+
+---
+
 ## Documentation
 
 | Document | Description |

package/dist/server/server.js

@@ -40,6 +40,7 @@ const db_instance_1 = require("./src/lib/db-instance");
 const response_poller_1 = require("./src/lib/response-poller");
 const auto_yes_manager_1 = require("./src/lib/auto-yes-manager");
 const schedule_manager_1 = require("./src/lib/schedule-manager");
+const resource_cleanup_1 = require("./src/lib/resource-cleanup");
 const db_migrations_1 = require("./src/lib/db-migrations");
 const env_1 = require("./src/lib/env");
 const db_repository_1 = require("./src/lib/db-repository");
@@ -230,6 +231,8 @@ app.prepare().then(() => {
         await initializeWorktrees();
         // [S3-010] Initialize schedule manager AFTER worktrees are ready
         (0, schedule_manager_1.initScheduleManager)();
+        // Issue #404: Initialize resource cleanup AFTER schedule manager
+        (0, resource_cleanup_1.initResourceCleanup)();
     });
     // Graceful shutdown with timeout
     let isShuttingDown = false;
@@ -246,6 +249,8 @@ app.prepare().then(() => {
         (0, auto_yes_manager_1.stopAllAutoYesPolling)();
         // Issue #294: Stop all scheduled executions (SIGKILL fire-and-forget)
         (0, schedule_manager_1.stopAllSchedules)();
+        // Issue #404: Stop resource cleanup timer
+        (0, resource_cleanup_1.stopResourceCleanup)();
         // Close WebSocket connections immediately (don't wait)
         (0, ws_server_1.closeWebSocket)();
         // Force exit after 3 seconds if graceful shutdown fails

package/dist/server/src/lib/auto-yes-manager.js

@@ -29,6 +29,9 @@ exports.detectAndRespondToPrompt = detectAndRespondToPrompt;
 exports.startAutoYesPolling = startAutoYesPolling;
 exports.stopAutoYesPolling = stopAutoYesPolling;
 exports.stopAllAutoYesPolling = stopAllAutoYesPolling;
+exports.deleteAutoYesState = deleteAutoYesState;
+exports.getAutoYesStateWorktreeIds = getAutoYesStateWorktreeIds;
+exports.getAutoYesPollerWorktreeIds = getAutoYesPollerWorktreeIds;
 const cli_session_1 = require("./cli-session");
 const prompt_detector_1 = require("./prompt-detector");
 const auto_yes_resolver_1 = require("./auto-yes-resolver");
@@ -37,6 +40,8 @@ const manager_1 = require("./cli-tools/manager");
 const cli_patterns_1 = require("./cli-patterns");
 const auto_yes_config_1 = require("../config/auto-yes-config");
 const prompt_key_1 = require("./prompt-key");
+const errors_1 = require("./errors");
+const tmux_capture_cache_1 = require("./tmux-capture-cache");
 // =============================================================================
 // Constants (Issue #138)
 // =============================================================================
@@ -62,16 +67,6 @@ exports.MAX_CONCURRENT_POLLERS = 50;
 exports.THINKING_CHECK_LINE_COUNT = 50;
 /** Worktree ID validation pattern (security: prevent command injection) */
 const WORKTREE_ID_PATTERN = /^[a-zA-Z0-9_-]+$/;
-/**
- * Extract error message from unknown error type.
- * Provides consistent error message extraction across the module (DRY).
- *
- * @param error - Unknown error object
- * @returns Error message string, or 'Unknown error' for non-Error values
- */
-function getErrorMessage(error) {
-    return error instanceof Error ? error.message : 'Unknown error';
-}
 /** In-memory storage for auto-yes states (globalThis for hot reload persistence) */
 const autoYesStates = globalThis.__autoYesStates ??
     (globalThis.__autoYesStates = new Map());
@@ -485,12 +480,18 @@ async function detectAndRespondToPrompt(worktreeId, pollerState, cliToolId, clea
         const manager = manager_1.CLIToolManager.getInstance();
         const cliTool = manager.getTool(cliToolId);
         const sessionName = cliTool.getSessionName(worktreeId);
-        await (0, prompt_answer_sender_1.sendPromptAnswer)({
-            sessionName,
-            answer,
-            cliToolId,
-            promptData: promptDetection.promptData,
-        });
+        try {
+            await (0, prompt_answer_sender_1.sendPromptAnswer)({
+                sessionName,
+                answer,
+                cliToolId,
+                promptData: promptDetection.promptData,
+            });
+        }
+        finally {
+            // Issue #405: Ensure cache invalidation even if sendPromptAnswer throws
+            (0, tmux_capture_cache_1.invalidateCache)(sessionName);
+        }
         // 5. Update timestamp and reset error count
         updateLastServerResponseTimestamp(worktreeId, Date.now());
         resetErrorCount(worktreeId);
@@ -505,7 +506,7 @@ async function detectAndRespondToPrompt(worktreeId, pollerState, cliToolId, clea
         // incrementErrorCount is called here, and this function never throws,
         // preventing double incrementErrorCount in the outer pollAutoYes() catch.
         incrementErrorCount(worktreeId);
-        console.warn(`[Auto-Yes Poller] Error in detectAndRespondToPrompt for worktree ${worktreeId}: ${getErrorMessage(error)}`);
+        console.warn(`[Auto-Yes Poller] Error in detectAndRespondToPrompt for worktree ${worktreeId}: ${(0, errors_1.getErrorMessage)(error)}`);
         return 'error';
     }
 }
@@ -557,7 +558,7 @@ async function pollAutoYes(worktreeId, cliToolId) {
         // errors only. detectAndRespondToPrompt() catches its own errors and returns
         // 'error' instead of throwing (preventing double incrementErrorCount).
         incrementErrorCount(worktreeId);
-        console.warn(`[Auto-Yes Poller] Error for worktree ${worktreeId}: ${getErrorMessage(error)}`);
+        console.warn(`[Auto-Yes Poller] Error for worktree ${worktreeId}: ${(0, errors_1.getErrorMessage)(error)}`);
     }
     // Schedule next poll (catch block fallthrough or other paths)
     scheduleNextPoll(worktreeId, cliToolId);
@@ -656,3 +657,42 @@ function stopAllAutoYesPolling() {
     }
     autoYesPollerStates.clear();
 }
+// =============================================================================
+// Cleanup Functions (Issue #404: Resource leak prevention)
+// =============================================================================
+/**
+ * Delete the auto-yes state for a worktree.
+ * Used during worktree deletion to prevent memory leaks in the autoYesStates Map.
+ *
+ * [SEC-404-001] Validates worktreeId before deletion.
+ *
+ * @param worktreeId - Worktree identifier (must pass isValidWorktreeId)
+ * @returns true if worktreeId was valid (deletion attempted), false if invalid
+ */
+function deleteAutoYesState(worktreeId) {
+    if (!isValidWorktreeId(worktreeId)) {
+        return false;
+    }
+    autoYesStates.delete(worktreeId);
+    return true;
+}
+/**
+ * Get all worktree IDs that have auto-yes state entries.
+ * Used by periodic resource cleanup to detect orphaned entries.
+ *
+ * @internal Exported for resource-cleanup and testing purposes.
+ * @returns Array of worktree IDs present in the autoYesStates Map
+ */
+function getAutoYesStateWorktreeIds() {
+    return Array.from(autoYesStates.keys());
+}
+/**
+ * Get all worktree IDs that have active auto-yes poller entries.
+ * Used by periodic resource cleanup to detect orphaned entries.
+ *
+ * @internal Exported for resource-cleanup and testing purposes.
+ * @returns Array of worktree IDs present in the autoYesPollerStates Map
+ */
+function getAutoYesPollerWorktreeIds() {
+    return Array.from(autoYesPollerStates.keys());
+}

package/dist/server/src/lib/claude-session.js

@@ -20,6 +20,7 @@ exports.restartClaudeSession = restartClaudeSession;
 const tmux_1 = require("./tmux");
 const cli_patterns_1 = require("./cli-patterns");
 const pasted_text_helper_1 = require("./pasted-text-helper");
+const tmux_capture_cache_1 = require("./tmux-capture-cache");
 const child_process_1 = require("child_process");
 const util_1 = require("util");
 const promises_1 = require("fs/promises");
@@ -256,8 +257,9 @@ async function getCleanPaneOutput(sessionName, lines = 50) {
  * Verify that Claude CLI is actually running inside a tmux session
  * Detects broken sessions where tmux exists but Claude failed to start
  *
- * @internal Exported for testing purposes only.
- * Follows clearCachedClaudePath() precedent (L148-156).
+ * Issue #405: Promoted from @internal to production export.
+ * Used by worktrees/route.ts and worktrees/[id]/route.ts for
+ * health-aware session status with listSessions() batch optimization.
  *
  * @param sessionName - tmux session name
  * @returns HealthCheckResult with healthy status and optional reason
@@ -627,6 +629,8 @@ async function sendMessageToClaude(worktreeId, message) {
     if (message.includes('\n')) {
         await (0, pasted_text_helper_1.detectAndResendIfPastedText)(sessionName);
     }
+    // Issue #405: Invalidate cache after sending message
+    (0, tmux_capture_cache_1.invalidateCache)(sessionName);
     console.log(`Sent message to Claude session: ${sessionName}`);
 }
 /**
@@ -675,7 +679,9 @@ async function stopClaudeSession(worktreeId) {
         if (exists) {
             await (0, tmux_1.sendKeys)(sessionName, '', false);
             // Send Ctrl+D (ASCII 4)
-            await execAsync(`tmux send-keys -t "${sessionName}" C-d`);
+            await (0, tmux_1.sendSpecialKey)(sessionName, 'C-d');
+            // Issue #405: Invalidate cache after sending stop signal
+            (0, tmux_capture_cache_1.invalidateCache)(sessionName);
             // Wait a moment for Claude to exit
             await new Promise((resolve) => setTimeout(resolve, 500));
         }

package/dist/server/src/lib/cli-session.js

@@ -2,14 +2,20 @@
 /**
  * Generic CLI session management
  * Manages CLI tool sessions (Claude, Codex, Gemini) within tmux
+ *
+ * Issue #405: Cache integration via tmux-capture-cache.ts
+ * - captureSessionOutput() uses getOrFetchCapture() for cache-backed capture
+ * - captureSessionOutputFresh() bypasses cache for prompt-response verification
  */
 Object.defineProperty(exports, "__esModule", { value: true });
 exports.isSessionRunning = isSessionRunning;
 exports.captureSessionOutput = captureSessionOutput;
+exports.captureSessionOutputFresh = captureSessionOutputFresh;
 exports.getSessionName = getSessionName;
 const tmux_1 = require("./tmux");
 const manager_1 = require("./cli-tools/manager");
 const logger_1 = require("./logger");
+const tmux_capture_cache_1 = require("./tmux-capture-cache");
 const logger = (0, logger_1.createLogger)('cli-session');
 /**
  * Check if CLI tool session is running
@@ -25,7 +31,10 @@ async function isSessionRunning(worktreeId, cliToolId) {
     return await (0, tmux_1.hasSession)(sessionName);
 }
 /**
- * Capture CLI session output
+ * Capture CLI session output (cache-backed)
+ *
+ * Issue #405: Uses getOrFetchCapture() for TTL-based caching with singleflight
+ * deduplication. Interface is unchanged for backward compatibility.
  *
  * @param worktreeId - Worktree ID
  * @param cliToolId - CLI tool ID (claude, codex, gemini)
@@ -38,27 +47,68 @@ async function captureSessionOutput(worktreeId, cliToolId, lines = 1000) {
     const manager = manager_1.CLIToolManager.getInstance();
     const cliTool = manager.getTool(cliToolId);
     const sessionName = cliTool.getSessionName(worktreeId);
-    // Check if session exists
-    const exists = await (0, tmux_1.hasSession)(sessionName);
-    if (!exists) {
-        log.debug('captureSessionOutput:sessionNotFound', { sessionName });
-        throw new Error(`${cliTool.name} session ${sessionName} does not exist`);
-    }
     try {
-        const output = await (0, tmux_1.capturePane)(sessionName, { startLine: -lines });
-        const actualLines = output.split('\n').length;
+        const output = await (0, tmux_capture_cache_1.getOrFetchCapture)(sessionName, lines, async () => {
+            // fetchFn: check session existence then capture
+            const exists = await (0, tmux_1.hasSession)(sessionName);
+            if (!exists) {
+                throw new Error(`${cliTool.name} session ${sessionName} does not exist`);
+            }
+            return await (0, tmux_1.capturePane)(sessionName, { startLine: -tmux_capture_cache_1.CACHE_MAX_CAPTURE_LINES });
+        });
         log.debug('captureSessionOutput:success', {
-            actualLines,
+            actualLines: output.split('\n').length,
             lastFewLines: output.split('\n').slice(-3).join(' | '),
         });
         return output;
     }
     catch (error) {
         const errorMessage = error instanceof Error ? error.message : String(error);
+        // Preserve original error messages (session not found vs capture failure)
+        if (errorMessage.includes('does not exist')) {
+            log.debug('captureSessionOutput:sessionNotFound', { sessionName });
+            throw error;
+        }
         log.error('captureSessionOutput:failed', { error: errorMessage });
         throw new Error(`Failed to capture ${cliTool.name} output: ${errorMessage}`);
     }
 }
+/**
+ * Capture CLI session output bypassing cache (fresh capture).
+ *
+ * Issue #405: Used by prompt-response endpoint to ensure fresh output
+ * for prompt re-verification. Writes back to cache on success.
+ *
+ * @param worktreeId - Worktree ID
+ * @param cliToolId - CLI tool ID
+ * @param lines - Number of lines to capture (default: 5000)
+ * @returns Captured output
+ */
+async function captureSessionOutputFresh(worktreeId, cliToolId, lines = 5000) {
+    const log = logger.withContext({ worktreeId, cliToolId });
+    log.debug('captureSessionOutputFresh:start', { requestedLines: lines });
+    const manager = manager_1.CLIToolManager.getInstance();
+    const cliTool = manager.getTool(cliToolId);
+    const sessionName = cliTool.getSessionName(worktreeId);
+    try {
+        const output = await (0, tmux_1.capturePane)(sessionName, { startLine: -lines });
+        // Write back to cache if non-empty [SEC4-007]
+        if (output.length > 0) {
+            (0, tmux_capture_cache_1.setCachedCapture)(sessionName, output, lines);
+        }
+        else {
+            (0, tmux_capture_cache_1.invalidateCache)(sessionName);
+        }
+        return (0, tmux_capture_cache_1.sliceOutput)(output, lines);
+    }
+    catch (error) {
+        // [DA3-005] Invalidate cache on error (TOCTOU safety)
+        (0, tmux_capture_cache_1.invalidateCache)(sessionName);
+        const errorMessage = error instanceof Error ? error.message : String(error);
+        log.error('captureSessionOutputFresh:failed', { error: errorMessage });
+        throw new Error(`Failed to capture ${cliTool.name} output: ${errorMessage}`);
+    }
+}
 /**
  * Get session name for a CLI tool and worktree
  *

package/dist/server/src/lib/cli-tools/codex.js

@@ -7,10 +7,8 @@ Object.defineProperty(exports, "__esModule", { value: true });
 exports.CodexTool = void 0;
 const base_1 = require("./base");
 const tmux_1 = require("../tmux");
-const child_process_1 = require("child_process");
-const util_1 = require("util");
 const pasted_text_helper_1 = require("../pasted-text-helper");
-const execAsync = (0, util_1.promisify)(child_process_1.exec);
+const tmux_capture_cache_1 = require("../tmux-capture-cache");
 /**
  * Extract error message from unknown error type (DRY)
  * Same pattern as claude-session.ts getErrorMessage()
@@ -80,9 +78,9 @@ class CodexTool extends base_1.BaseCLITool {
             await new Promise((resolve) => setTimeout(resolve, CODEX_MODEL_SELECT_WAIT_MS));
             // T2.6: Skip model selection dialog by sending Down arrow + Enter
             // This selects the default model and proceeds to the prompt
-            await execAsync(`tmux send-keys -t "${sessionName}" Down`);
+            await (0, tmux_1.sendSpecialKeys)(sessionName, ['Down']);
             await new Promise((resolve) => setTimeout(resolve, CODEX_MODEL_SELECT_WAIT_MS));
-            await execAsync(`tmux send-keys -t "${sessionName}" Enter`);
+            await (0, tmux_1.sendSpecialKeys)(sessionName, ['Enter']);
             await new Promise((resolve) => setTimeout(resolve, CODEX_MODEL_SELECT_WAIT_MS));
             console.log(`✓ Started Codex session: ${sessionName}`);
         }
@@ -110,7 +108,7 @@ class CodexTool extends base_1.BaseCLITool {
             // Wait a moment for the text to be typed
             await new Promise((resolve) => setTimeout(resolve, 100));
             // Send Enter key separately
-            await execAsync(`tmux send-keys -t "${sessionName}" C-m`);
+            await (0, tmux_1.sendSpecialKey)(sessionName, 'C-m');
             // Wait a moment for the message to be processed
             await new Promise((resolve) => setTimeout(resolve, 200));
             // Issue #212: Detect [Pasted text] and resend Enter for multi-line messages
@@ -118,6 +116,8 @@ class CodexTool extends base_1.BaseCLITool {
             if (message.includes('\n')) {
                 await (0, pasted_text_helper_1.detectAndResendIfPastedText)(sessionName);
             }
+            // Issue #405: Invalidate cache after sending message
+            (0, tmux_capture_cache_1.invalidateCache)(sessionName);
             console.log(`✓ Sent message to Codex session: ${sessionName}`);
         }
         catch (error) {
@@ -137,7 +137,7 @@ class CodexTool extends base_1.BaseCLITool {
             const exists = await (0, tmux_1.hasSession)(sessionName);
             if (exists) {
                 // Send Ctrl+D (ASCII 4)
-                await execAsync(`tmux send-keys -t "${sessionName}" C-d`);
+                await (0, tmux_1.sendSpecialKey)(sessionName, 'C-d');
                 // Wait a moment for Codex to exit
                 await new Promise((resolve) => setTimeout(resolve, 500));
             }

package/dist/server/src/lib/cli-tools/gemini.js

@@ -13,6 +13,7 @@ exports.GeminiTool = void 0;
 const base_1 = require("./base");
 const tmux_1 = require("../tmux");
 const pasted_text_helper_1 = require("../pasted-text-helper");
+const tmux_capture_cache_1 = require("../tmux-capture-cache");
 /**
  * Extract error message from unknown error type (DRY)
  */
@@ -141,6 +142,8 @@ class GeminiTool extends base_1.BaseCLITool {
             if (message.includes('\n')) {
                 await (0, pasted_text_helper_1.detectAndResendIfPastedText)(sessionName);
             }
+            // Issue #405: Invalidate cache after sending message
+            (0, tmux_capture_cache_1.invalidateCache)(sessionName);
             console.log(`✓ Sent message to Gemini session: ${sessionName}`);
         }
         catch (error) {