gencode-ai 0.1.3 → 0.3.0

package/docs/proposals/0041-configuration-system.md

@@ -49,9 +49,7 @@ Claude Code supports extensive environment variables:
 
 **Cloud Providers:**
 - `CLAUDE_CODE_USE_BEDROCK` - Enable AWS Bedrock
-- `CLAUDE_CODE_USE_VERTEX` - Enable Google Vertex AI
 - `CLAUDE_CODE_SKIP_BEDROCK_AUTH` - Bypass AWS auth
-- `CLAUDE_CODE_SKIP_VERTEX_AUTH` - Bypass Vertex auth
 
 **Operational:**
 - `CLAUDE_CODE_MAX_OUTPUT_TOKENS` - Token limit
@@ -121,7 +119,7 @@ OpenCode provides additional patterns worth adopting:
 ### Directory Structure
 
 ```
-~/.gencode/                          # User-level configuration
+~/.gen/                          # User-level configuration
 ├── settings.json                    # Main user config
 ├── settings.local.json              # User local overrides (gitignored pattern)
 ├── GENCODE.md                       # User context (like CLAUDE.md)
@@ -135,7 +133,7 @@ OpenCode provides additional patterns worth adopting:
 └── sessions/                        # Session data
 
 ./gencode.json                       # Project config (like opencode.json)
-./.gencode/                          # Project directory
+./.gen/                          # Project directory
 ├── settings.local.json              # Project local overrides (gitignored)
 ├── GENCODE.md                       # Project context
 ├── rules/                           # Path-scoped rules
@@ -146,10 +144,10 @@ OpenCode provides additional patterns worth adopting:
 
 1. **Environment variables** (`GENCODE_*`, provider API keys)
 2. **CLI arguments** (`--model`, `--provider`)
-3. **Project local** (`./.gencode/settings.local.json`)
+3. **Project local** (`./.gen/settings.local.json`)
 4. **Project shared** (`./gencode.json`)
-5. **User local** (`~/.gencode/settings.local.json`)
-6. **User global** (`~/.gencode/settings.json`)
+5. **User local** (`~/.gen/settings.local.json`)
+6. **User global** (`~/.gen/settings.json`)
 7. **Defaults**
 
 ### API Design
@@ -227,8 +225,8 @@ interface EnvHandler {
 ### Environment Variables
 
 **Provider Selection:**
-- `GENCODE_PROVIDER` - Provider name (anthropic, openai, gemini, bedrock, vertex)
-- `GENCODE_MODEL` - Model ID
+- `GEN_PROVIDER` - Provider name (anthropic, openai, gemini, bedrock, vertex)
+- `GEN_MODEL` - Model ID
 - `GENCODE_CONFIG` - Custom config file path
 
 **Provider API Keys (Auto-detect):**
@@ -300,7 +298,7 @@ const ConfigSchema = z.object({
 });
 
 export class ConfigLoader {
-  private userConfigDir = path.join(os.homedir(), '.gencode');
+  private userConfigDir = path.join(os.homedir(), '.gen');
 
   async load(): Promise<GencodeConfig> {
     const configs = await Promise.all([
@@ -349,7 +347,7 @@ export class EnvHandler {
   };
 
   getProviderFromEnv(): string | undefined {
-    return process.env.GENCODE_PROVIDER;
+    return process.env.GEN_PROVIDER;
   }
 
   getApiKey(provider: string): string | undefined {
@@ -442,7 +440,7 @@ $ gencode
 # Uses OpenAI automatically
 
 # Or specify explicitly
-$ export GENCODE_PROVIDER=anthropic
+$ export GEN_PROVIDER=anthropic
 $ export ANTHROPIC_API_KEY=sk-ant-...
 $ gencode
 ```
@@ -480,7 +478,7 @@ $ gencode config deny "Read(.env)"
 
 ### Single Config File
 
-A single `~/.gencoderc` file would be simpler but lacks:
+A single `~/.genrc` file would be simpler but lacks:
 - Project-specific overrides
 - Team-shareable settings
 - Local-only sensitive settings
@@ -526,7 +524,7 @@ Simpler implementation but:
 
 ## Migration Path
 
-1. **From mycode**: Migrate `~/.mycode/` to `~/.gencode/`
+1. **From mycode**: Migrate `~/.mycode/` to `~/.gen/`
 2. **Version Detection**: Check for legacy config locations and prompt migration
 3. **Backward Compatibility**: Support reading old config format for one major version
 
@@ -570,9 +568,9 @@ Simpler implementation but:
 
 ### Key Implementation Details
 
-1. **Multi-level Loading**: User (`~/.gencode/`) → Project (`./.gencode/`) with proper merge
+1. **Multi-level Loading**: User (`~/.gen/`) → Project (`./.gen/`) with proper merge
 2. **Claude Code Compatibility**: Supports both `GENCODE.md` and `CLAUDE.md` for memory files
-3. **Environment Variables**: `GENCODE_PROVIDER`, `GENCODE_MODEL`, and provider API key auto-detection
+3. **Environment Variables**: `GEN_PROVIDER`, `GEN_MODEL`, and provider API key auto-detection
 4. **Deep Merge**: Arrays concatenate, objects merge recursively
 5. **JSONC Support**: Configuration files support comments and trailing commas
 
@@ -580,8 +578,8 @@ Simpler implementation but:
 
 1. Environment variables (`GENCODE_*`)
 2. CLI arguments
-3. Project local (`./.gencode/settings.local.json`)
+3. Project local (`./.gen/settings.local.json`)
 4. Project shared (`./gencode.json`)
-5. User local (`~/.gencode/settings.local.json`)
-6. User global (`~/.gencode/settings.json`)
+5. User local (`~/.gen/settings.local.json`)
+6. User global (`~/.gen/settings.json`)
 7. Defaults

package/docs/proposals/0042-prompt-optimization.md

@@ -199,7 +199,7 @@ src/prompts/
 
 ### 4.2 Prompt Loading Strategy
 
-The prompt loading system uses a **model → provider → prompt** flow, leveraging the `~/.gencode/providers.json` configuration to automatically determine which prompt to use.
+The prompt loading system uses a **model → provider → prompt** flow, leveraging the `~/.gen/providers.json` configuration to automatically determine which prompt to use.
 
 #### 4.2.1 Loading Flow
 
@@ -209,7 +209,7 @@ model ID (e.g., "claude-sonnet-4-5@20250929")
     ▼
 ┌─────────────────────────────────────┐
 │ Look up provider in providers.json  │
-│ (search models.{provider}.list)     │
+│ (search models[provider:*].list)    │
 └──────────────┬──────────────────────┘
                │
     ┌──────────┴──────────┐
@@ -242,18 +242,26 @@ model ID (e.g., "claude-sonnet-4-5@20250929")
 
 #### 4.2.2 Provider Lookup from providers.json
 
-The `~/.gencode/providers.json` file stores cached models for each connected provider:
+The `~/.gen/providers.json` file stores cached models for each provider using `"provider:authMethod"` keys:
 
 ```json
 {
   "models": {
-    "anthropic": {
+    "anthropic:api_key": {
+      "cachedAt": "2025-01-17T10:00:00Z",
       "list": [
         { "id": "claude-sonnet-4-5@20250929", "name": "Claude Sonnet 4.5" },
         { "id": "claude-opus-4-1@20250805", "name": "Claude Opus 4.1" }
       ]
     },
-    "gemini": {
+    "anthropic:vertex": {
+      "cachedAt": "2025-01-17T09:00:00Z",
+      "list": [
+        { "id": "claude-3-5-sonnet@20241022", "name": "Claude 3.5 Sonnet" }
+      ]
+    },
+    "gemini:api_key": {
+      "cachedAt": "2025-01-17T08:00:00Z",
       "list": [
         { "id": "gemini-2.5-pro", "name": "Gemini 2.5 Pro" }
       ]
@@ -262,7 +270,7 @@ The `~/.gencode/providers.json` file stores cached models for each connected pro
 }
 ```
 
-Given a model ID, the system searches through each provider's model list to find the owning provider.
+Given a model ID, the system searches through each provider's model list to find the owning provider. The key format is `"provider:authMethod"` to support multiple authentication methods for the same provider.
 
 #### 4.2.3 Provider to Prompt Type Mapping
 
@@ -293,7 +301,7 @@ This ensures the system always has a valid prompt to use.
 // 5.1.1 Provider Type Definition
 export type ProviderType = 'anthropic' | 'openai' | 'gemini' | 'generic';
 
-// 5.1.2 Look up provider for a model from ~/.gencode/providers.json
+// 5.1.2 Look up provider for a model from ~/.gen/providers.json
 export function getProviderForModel(model: string): string | null {
   const config = loadProvidersConfig();
   if (!config?.models) return null;
@@ -462,7 +470,7 @@ User starts GenCode CLI
            │
            ▼
 ┌─────────────────────┐     ┌─────────────────────────────┐
-│ 3. Memory Loading    │────▶│ Load ~/.gencode/GENCODE.md  │
+│ 3. Memory Loading    │────▶│ Load ~/.gen/GENCODE.md  │
 │    (if exists)       │     │ Load ./CLAUDE.md (fallback) │
 └──────────┬──────────┘     └─────────────────────────────┘
            │
@@ -480,7 +488,7 @@ User starts GenCode CLI
 │                                                         │
 │    ┌─────────────────────────────────────────────────┐  │
 │    │ a. getPromptTypeForModel(model, fallback)       │  │
-│    │    - Look up provider in ~/.gencode/providers.json│
+│    │    - Look up provider in ~/.gen/providers.json│
 │    │    - Map provider → prompt type                  │  │
 │    │    - Fallback to 'generic' if not found         │  │
 │    └─────────────────────────────────────────────────┘  │

package/docs/proposals/README.md

@@ -19,15 +19,14 @@ This directory contains enhancement proposals for the gencode project. Each prop
 | [0001](./0001-web-fetch-tool.md) | WebFetch Tool | Implemented |
 | [0002](./0002-web-search-tool.md) | WebSearch Tool | Implemented |
 | [0003](./0003-task-subagents.md) | Task Tool & Subagents | Draft |
-| [0004](./0004-plan-mode.md) | Plan Mode | Draft |
+| [0004](./0004-plan-mode.md) | Plan Mode | Implemented |
 | [0005](./0005-todo-system.md) | Todo System | Implemented |
 | [0006](./0006-memory-system.md) | Memory System (MYCODE.md) | Implemented |
 | [0007](./0007-context-management.md) | Context Management | Draft |
-| [0008](./0008-checkpointing.md) | Checkpointing | Draft |
+| [0008](./0008-checkpointing.md) | Checkpointing | Partially Implemented |
 | [0009](./0009-hooks-system.md) | Hooks System | Draft |
 | [0010](./0010-mcp-integration.md) | MCP Integration | Draft |
-| [0011](./0011-custom-commands.md) | Custom Commands | Draft |
-| [0012](./0012-ask-user-question.md) | AskUserQuestion Tool | Implemented |
+| [0021](./0021-skills-system.md) | Skills System | Draft |
 | [0041](./0041-configuration-system.md) | Configuration System | Implemented |
 | [0042](./0042-prompt-optimization.md) | Prompt System Optimization | Implemented |
 
@@ -35,6 +34,8 @@ This directory contains enhancement proposals for the gencode project. Each prop
 
 | ID | Title | Status |
 |----|-------|--------|
+| [0011](./0011-custom-commands.md) | Custom Commands | Draft |
+| [0012](./0012-ask-user-question.md) | AskUserQuestion Tool | Implemented |
 | [0013](./0013-multi-edit-tool.md) | MultiEdit Tool | Draft |
 | [0014](./0014-lsp-tool.md) | LSP Tool | Draft |
 | [0015](./0015-ls-tool.md) | LS Tool | Draft |
@@ -43,11 +44,10 @@ This directory contains enhancement proposals for the gencode project. Each prop
 | [0018](./0018-parallel-tool-execution.md) | Parallel Tool Execution | Draft |
 | [0019](./0019-session-enhancements.md) | Session Enhancements | Draft |
 | [0020](./0020-session-summarization.md) | Session Summarization | Draft |
-| [0021](./0021-skills-system.md) | Skills System | Draft |
 | [0022](./0022-plugin-system.md) | Plugin System | Draft |
 | [0023](./0023-permission-enhancements.md) | Permission Enhancements | Draft |
 | [0024](./0024-keyboard-shortcuts.md) | Keyboard Shortcuts | Draft |
-| [0025](./0025-cost-tracking.md) | Cost Tracking | Draft |
+| [0025](./0025-cost-tracking.md) | Cost Tracking | Implemented |
 | [0026](./0026-git-integration.md) | Git Integration | Draft |
 | [0027](./0027-enhanced-read-tool.md) | Enhanced Read Tool | Draft |
 | [0028](./0028-enhanced-bash-tool.md) | Enhanced Bash Tool | Draft |

package/docs/providers.md

@@ -131,16 +131,25 @@ To use Claude models via Google Vertex AI:
 ### 1. Set Environment Variables
 
 ```bash
+# Required: Enable Vertex AI
+export CLAUDE_CODE_USE_VERTEX=1
+
 # Required: Your GCP project ID
 export ANTHROPIC_VERTEX_PROJECT_ID="your-project-id"
 
-# Optional: Region (defaults to us-east5)
-export ANTHROPIC_VERTEX_REGION="us-east5"
+# Required: Region (use 'global' or specific region)
+export CLOUD_ML_REGION=global
 ```
 
-Alternative variables also supported:
-- `GOOGLE_CLOUD_PROJECT` - GCP project ID
-- `CLOUD_ML_REGION` - GCP region
+**Environment Variable Details:**
+
+| Variable | Required | Default | Description |
+|----------|----------|---------|-------------|
+| `CLAUDE_CODE_USE_VERTEX` | Yes | - | Set to `1` to enable Vertex AI |
+| `ANTHROPIC_VERTEX_PROJECT_ID` | Yes | - | Your GCP project ID (also accepts `GCLOUD_PROJECT` or `GOOGLE_CLOUD_PROJECT`) |
+| `CLOUD_ML_REGION` | Yes | `us-east5` | Region (use `global` or specific region like `us-east5`) |
+
+**Documentation:** https://code.claude.com/docs/en/google-vertex-ai
 
 ### 2. Authenticate with Google Cloud
 
@@ -169,12 +178,24 @@ gcloud services enable aiplatform.googleapis.com
 # Press Enter to connect
 ```
 
+## Architecture
+
+GenCode uses a two-layer provider architecture:
+
+- **Layer 1: Provider** (Semantic layer) - `anthropic` | `openai` | `gemini`
+- **Layer 2: AuthMethod** (Implementation layer) - `api_key` | `vertex` | `bedrock` | `azure`
+
+Each provider can support multiple authentication methods. For example, Anthropic supports:
+- `api_key` - Direct API access
+- `vertex` - Google Vertex AI
+- `bedrock` - Amazon Bedrock (coming soon)
+
 ## Configuration Storage
 
 Provider connections and cached models are stored in:
 
 ```
-~/.gencode/
+~/.gen/
 ├── settings.json      # Current model and provider settings
 └── providers.json     # Provider connections and model cache
 ```
@@ -185,23 +206,87 @@ Provider connections and cached models are stored in:
 {
   "connections": {
     "anthropic": {
-      "method": "vertex",
+      "authMethod": "vertex",
+      "method": "Google Vertex AI",
       "connectedAt": "2025-01-15T10:00:00Z"
+    },
+    "openai": {
+      "authMethod": "api_key",
+      "method": "Direct API",
+      "connectedAt": "2025-01-15T09:00:00Z"
     }
   },
   "models": {
-    "anthropic": {
+    "anthropic:vertex": {
       "cachedAt": "2025-01-15T10:00:00Z",
+      "list": [
+        { "id": "claude-3-5-sonnet@20241022", "name": "Claude 3.5 Sonnet" }
+      ]
+    },
+    "anthropic:api_key": {
+      "cachedAt": "2025-01-15T11:00:00Z",
       "list": [
         { "id": "claude-sonnet-4-5@20250929", "name": "Claude Sonnet 4.5" }
       ]
+    },
+    "openai:api_key": {
+      "cachedAt": "2025-01-15T09:00:00Z",
+      "list": [
+        { "id": "gpt-4", "name": "GPT-4" }
+      ]
     }
   },
   "searchProvider": "exa"
 }
 ```
 
-The `searchProvider` field stores the selected search provider. Valid values: `exa`, `serper`, `brave`. If not set, defaults to `exa`.
+**Key points:**
+- `connections` stores the active connection for each provider
+  - `authMethod` - Authentication method being used
+  - `method` - Display name (optional, legacy field)
+- `models` uses `"provider:authMethod"` as the key
+  - Supports multiple auth methods for the same provider
+  - Each auth method has its own cached model list
+- `searchProvider` - Selected search provider (`exa`, `serper`, `brave`)
+
+### Migration from Old Format
+
+If you're upgrading from an older version, run the migration script to update your configuration:
+
+```bash
+npm run migrate
+```
+
+The script will:
+1. Convert old format models keys (e.g., `"anthropic"`) to new format (e.g., `"anthropic:vertex"`)
+2. Add `authMethod` field to connections
+3. Create a backup of your old configuration
+
+**Old format:**
+```json
+{
+  "models": {
+    "anthropic": {
+      "provider": "anthropic",
+      "authMethod": "vertex",
+      "cachedAt": "...",
+      "list": [...]
+    }
+  }
+}
+```
+
+**New format:**
+```json
+{
+  "models": {
+    "anthropic:vertex": {
+      "cachedAt": "...",
+      "list": [...]
+    }
+  }
+}
+```
 
 ## Troubleshooting
 

package/examples/test-checkpointing.ts

@@ -0,0 +1,121 @@
+/**
+ * Test Checkpointing System
+ *
+ * This example demonstrates the automatic file change tracking and rewind capabilities.
+ */
+
+import { CheckpointManager } from '../src/checkpointing/index.js';
+import * as fs from 'fs/promises';
+import * as path from 'path';
+import { fileURLToPath } from 'url';
+
+const __filename = fileURLToPath(import.meta.url);
+const __dirname = path.dirname(__filename);
+
+async function testCheckpointing() {
+  console.log('🧪 Testing Checkpointing System\n');
+
+  const manager = new CheckpointManager('test-session');
+  const testDir = path.join(__dirname, '../.test-checkpoints');
+  const testFile = path.join(testDir, 'example.txt');
+
+  try {
+    // Create test directory
+    await fs.mkdir(testDir, { recursive: true });
+
+    console.log('1️⃣  Recording file creation...');
+    const originalContent = 'Hello, World!';
+    await fs.writeFile(testFile, originalContent, 'utf-8');
+
+    manager.recordChange({
+      path: testFile,
+      changeType: 'create',
+      previousContent: null,
+      newContent: originalContent,
+      toolName: 'Write',
+    });
+
+    console.log('   ✓ File created:', testFile);
+    console.log('   ✓ Checkpoint recorded\n');
+
+    console.log('2️⃣  Recording file modification...');
+    const modifiedContent = 'Hello, Checkpointing!';
+    await fs.writeFile(testFile, modifiedContent, 'utf-8');
+
+    manager.recordChange({
+      path: testFile,
+      changeType: 'modify',
+      previousContent: originalContent,
+      newContent: modifiedContent,
+      toolName: 'Edit',
+    });
+
+    console.log('   ✓ File modified');
+    console.log('   ✓ Checkpoint recorded\n');
+
+    console.log('3️⃣  Listing checkpoints...');
+    console.log(manager.formatCheckpointList(true));
+    console.log();
+
+    console.log('4️⃣  Getting summary...');
+    const summary = manager.getSummary();
+    console.log(`   Created: ${summary.created}`);
+    console.log(`   Modified: ${summary.modified}`);
+    console.log(`   Deleted: ${summary.deleted}`);
+    console.log(`   Total: ${summary.total}\n`);
+
+    console.log('5️⃣  Rewinding last change...');
+    const result = await manager.rewind({ count: 1 });
+
+    if (result.success) {
+      console.log('   ✓ Rewind successful');
+      result.revertedFiles.forEach((f) => {
+        console.log(`     • ${path.basename(f.path)} (${f.action})`);
+      });
+
+      // Verify content was restored
+      const restoredContent = await fs.readFile(testFile, 'utf-8');
+      if (restoredContent === originalContent) {
+        console.log('   ✓ Content restored correctly\n');
+      } else {
+        console.log('   ✗ Content mismatch!\n');
+      }
+    } else {
+      console.log('   ✗ Rewind failed');
+      result.errors.forEach((e) => console.log(`     ${e.error}`));
+    }
+
+    console.log('6️⃣  Rewinding all changes...');
+    const finalResult = await manager.rewind({ all: true });
+
+    if (finalResult.success) {
+      console.log('   ✓ All changes reverted');
+      finalResult.revertedFiles.forEach((f) => {
+        console.log(`     • ${path.basename(f.path)} (${f.action})`);
+      });
+
+      // Verify file was deleted
+      try {
+        await fs.access(testFile);
+        console.log('   ✗ File still exists!\n');
+      } catch {
+        console.log('   ✓ File was deleted as expected\n');
+      }
+    }
+
+    console.log('✅ All tests passed!\n');
+  } catch (error) {
+    console.error('❌ Test failed:', error);
+  } finally {
+    // Cleanup
+    try {
+      await fs.rm(testDir, { recursive: true, force: true });
+      console.log('🧹 Cleaned up test directory');
+    } catch {
+      // Ignore cleanup errors
+    }
+  }
+}
+
+// Run the test
+testCheckpointing();

package/examples/test-cost-tracking.ts

@@ -0,0 +1,77 @@
+/**
+ * Test cost tracking functionality
+ */
+
+import { createProvider } from '../src/providers/index.js';
+import { calculateCost, formatCost, formatTokens } from '../src/pricing/calculator.js';
+
+async function testCostTracking() {
+  console.log('='.repeat(60));
+  console.log('Testing Cost Tracking Feature');
+  console.log('='.repeat(60));
+  console.log();
+
+  // Test 1: Cost Calculation
+  console.log('Test 1: Cost Calculation');
+  console.log('-'.repeat(60));
+
+  const testCases = [
+    { provider: 'anthropic', model: 'claude-sonnet-4', input: 1000, output: 500 },
+    { provider: 'openai', model: 'gpt-4o', input: 1000, output: 500 },
+    { provider: 'gemini', model: 'gemini-2.0-flash', input: 1000, output: 500 },
+  ];
+
+  for (const test of testCases) {
+    const cost = calculateCost(test.provider, test.model, {
+      inputTokens: test.input,
+      outputTokens: test.output,
+    });
+
+    console.log(`\n${test.provider}/${test.model}:`);
+    console.log(`  Input: ${formatTokens(test.input)} tokens`);
+    console.log(`  Output: ${formatTokens(test.output)} tokens`);
+    console.log(`  Cost: ${formatCost(cost.totalCost)}`);
+    console.log(`  Breakdown: ${formatCost(cost.inputCost)} (input) + ${formatCost(cost.outputCost)} (output)`);
+  }
+
+  console.log();
+  console.log('='.repeat(60));
+  console.log();
+
+  // Test 2: Real API Call (if env var is set)
+  if (process.env.ANTHROPIC_API_KEY) {
+    console.log('Test 2: Real API Call with Anthropic');
+    console.log('-'.repeat(60));
+
+    const provider = createProvider({ provider: 'anthropic', model: 'claude-haiku-3-5' });
+
+    const response = await provider.complete({
+      model: 'claude-haiku-3-5',
+      messages: [{ role: 'user', content: 'Say hello in one word' }],
+      maxTokens: 10,
+    });
+
+    console.log('\nResponse:', response.content[0]?.type === 'text' ? response.content[0].text : '');
+
+    if (response.usage) {
+      console.log(`\nToken Usage:`);
+      console.log(`  Input: ${formatTokens(response.usage.inputTokens)}`);
+      console.log(`  Output: ${formatTokens(response.usage.outputTokens)}`);
+    }
+
+    if (response.cost) {
+      console.log(`\nCost Estimate:`);
+      console.log(`  Total: ${formatCost(response.cost.totalCost)}`);
+      console.log(`  Input: ${formatCost(response.cost.inputCost)}`);
+      console.log(`  Output: ${formatCost(response.cost.outputCost)}`);
+    }
+
+    console.log();
+    console.log('='.repeat(60));
+  } else {
+    console.log('Skipping API test (no ANTHROPIC_API_KEY)');
+    console.log('='.repeat(60));
+  }
+}
+
+testCostTracking().catch(console.error);

package/examples/test-interrupt-cleanup.ts

@@ -0,0 +1,94 @@
+/**
+ * Test Interrupt Cleanup
+ *
+ * Simulates the interrupt scenario and verifies cleanup works correctly
+ */
+
+import { Agent } from '../src/agent/index.js';
+import type { Message } from '../src/providers/types.js';
+
+async function testInterruptCleanup() {
+  console.log('🧪 Testing Interrupt Cleanup\n');
+
+  const agent = new Agent({
+    provider: 'anthropic',
+    model: 'claude-3-5-sonnet-20241022',
+    cwd: process.cwd(),
+  });
+
+  try {
+    // Simulate incomplete tool_use message (what happens during interrupt)
+    console.log('1️⃣  Simulating incomplete tool_use message...');
+
+    // Access private messages array via getHistory and direct manipulation
+    const history = agent.getHistory();
+    console.log(`   Initial history: ${history.length} messages`);
+
+    // Add a user message
+    (agent as any).messages.push({
+      role: 'user',
+      content: 'Create a file',
+    });
+    console.log(`   After user message: ${agent.getHistory().length} messages`);
+
+    // Add an assistant message with tool_use (simulating interrupt point)
+    (agent as any).messages.push({
+      role: 'assistant',
+      content: [
+        { type: 'text', text: 'I will create the file.' },
+        {
+          type: 'tool_use',
+          id: 'toolu_test_123',
+          name: 'Write',
+          input: { file_path: 'test.txt', content: 'hello' },
+        },
+      ],
+    });
+    console.log(`   After assistant+tool_use: ${agent.getHistory().length} messages`);
+    console.log(`   Last message role: ${agent.getHistory()[agent.getHistory().length - 1].role}`);
+
+    // Verify incomplete state
+    const lastMessage = agent.getHistory()[agent.getHistory().length - 1];
+    const hasToolUse = Array.isArray(lastMessage.content) &&
+      lastMessage.content.some((c: any) => c.type === 'tool_use');
+    console.log(`   Has incomplete tool_use: ${hasToolUse}\n`);
+
+    console.log('2️⃣  Calling cleanupIncompleteMessages()...');
+    agent.cleanupIncompleteMessages();
+
+    const afterCleanup = agent.getHistory();
+    console.log(`   After cleanup: ${afterCleanup.length} messages`);
+
+    if (afterCleanup.length === 1 && afterCleanup[0].role === 'user') {
+      console.log('   ✓ Incomplete assistant message removed!\n');
+    } else {
+      console.log('   ✗ Cleanup failed!\n');
+      console.log('   Remaining messages:', JSON.stringify(afterCleanup, null, 2));
+    }
+
+    console.log('3️⃣  Testing with complete messages (no tool_use)...');
+
+    // Add complete assistant message
+    (agent as any).messages.push({
+      role: 'assistant',
+      content: [{ type: 'text', text: 'Here is your answer.' }],
+    });
+    console.log(`   Before cleanup: ${agent.getHistory().length} messages`);
+
+    agent.cleanupIncompleteMessages();
+    console.log(`   After cleanup: ${agent.getHistory().length} messages`);
+
+    if (agent.getHistory().length === 2) {
+      console.log('   ✓ Complete messages preserved!\n');
+    } else {
+      console.log('   ✗ Complete message was incorrectly removed!\n');
+    }
+
+    console.log('✅ All tests passed!\n');
+  } catch (error) {
+    console.error('❌ Test failed:', error);
+  }
+}
+
+// Run the test
+testInterruptCleanup();