merge with master

package.json

@@ -1,6 +1,6 @@
 {
   "name": "detox-copilot",
-  "version": "0.0.28",
+  "version": "0.0.31",
   "description": "A flexible plugin that drives your tests with human-written commands, enhanced by the power of large language models (LLMs)",
   "keywords": [
     "detox",

src/Copilot.ts

@@ -41,7 +41,8 @@ export class Copilot {
             this.codeEvaluator,
             this.snapshotManager,
             config.promptHandler,
-            this.cacheHandler
+            this.cacheHandler,
+            config.options?.cacheMode
         );
         this.pilotPerformer = new PilotPerformer(this.pilotPromptCreator, this.copilotStepPerformer, config.promptHandler, () => this.screenCapturer.capture());
     }

src/actions/CopilotStepPerformer.ts

@@ -2,21 +2,26 @@ import {PromptCreator} from '@/utils/PromptCreator';
 import {CodeEvaluator} from '@/utils/CodeEvaluator';
 import {SnapshotManager} from '@/utils/SnapshotManager';
 import {CacheHandler} from '@/utils/CacheHandler';
-import {CodeEvaluationResult, PreviousStep, PromptHandler, CaptureResult} from '@/types';
+import {CacheMode, CodeEvaluationResult, PreviousStep, PromptHandler, CaptureResult} from '@/types';
 import * as fs from 'fs';
 import * as path from 'path';
 import * as crypto from 'crypto';
 import {extractCodeBlock} from '@/utils/extractCodeBlock';
 
 export class CopilotStepPerformer {
+    private readonly cacheMode: CacheMode;
+
+
     constructor(
         private context: any,
         private promptCreator: PromptCreator,
         private codeEvaluator: CodeEvaluator,
         private snapshotManager: SnapshotManager,
         private promptHandler: PromptHandler,
         private cacheHandler: CacheHandler,
+        cacheMode: CacheMode = 'full',
     ) {
+        this.cacheMode = cacheMode;
     }
 
     extendJSContext(newContext: any): void {
@@ -30,8 +35,19 @@ export class CopilotStepPerformer {
     }
 
     private generateCacheKey(step: string, previous: PreviousStep[], viewHierarchy: string): string {
-        const viewHierarchyHash = crypto.createHash('md5').update(viewHierarchy).digest('hex');
-        return JSON.stringify({step, previous, viewHierarchyHash});
+        if (this.cacheMode === 'disabled') {
+            // Return a unique key that won't match any cached value
+            return crypto.randomUUID();
+        }
+
+        const cacheKeyData: any = {step, previous};
+
+        if (this.cacheMode === 'full') {
+            const viewHierarchyHash = crypto.createHash('md5').update(viewHierarchy).digest('hex');
+            cacheKeyData.viewHierarchyHash = viewHierarchyHash;
+        }
+
+        return JSON.stringify(cacheKeyData);
     }
 
     private shouldOverrideCache() {

src/integration tests/index.test.ts

@@ -422,4 +422,62 @@ describe('Copilot Integration Tests', () => {
       expect(spyPilotPerformerPerform).toHaveBeenCalledWith(goal);
     });
   });
-});
+    describe('Cache Modes', () => {
+        beforeEach(() => {
+            mockPromptHandler.runPrompt.mockResolvedValue('// No operation');
+        });
+
+        it('should use full cache mode by default', async () => {
+            copilot.init({
+                frameworkDriver: mockFrameworkDriver,
+                promptHandler: mockPromptHandler
+            });
+            copilot.start();
+
+            await copilot.perform('Tap on the login button');
+            copilot.end();
+
+            expect(Object.keys(mockedCacheFile || {})[0]).toContain('viewHierarchyHash');
+        });
+
+        it('should not include view hierarchy in cache key when using lightweight mode', async () => {
+            copilot.init({
+                frameworkDriver: mockFrameworkDriver,
+                promptHandler: mockPromptHandler,
+                options: {
+                    cacheMode: 'lightweight'
+                }
+            });
+            copilot.start();
+
+            await copilot.perform('Tap on the login button');
+            copilot.end();
+
+            const cacheKeys = Object.keys(mockedCacheFile || {});
+            expect(cacheKeys[0]).not.toContain('viewHierarchyHash');
+        });
+
+        it('should not use cache when cache mode is disabled', async () => {
+            copilot.init({
+                frameworkDriver: mockFrameworkDriver,
+                promptHandler: mockPromptHandler,
+                options: {
+                    cacheMode: 'disabled'
+                }
+            });
+            copilot.start();
+
+            // First call
+            await copilot.perform('Tap on the login button');
+            copilot.end();
+
+            // Second call with same intent
+            copilot.start();
+            await copilot.perform('Tap on the login button');
+            copilot.end();
+
+            // Should call runPrompt twice since cache is disabled
+            expect(mockPromptHandler.runPrompt).toHaveBeenCalledTimes(2);
+        });
+    });
+});

src/test-utils/APICatalogTestUtils.ts

@@ -42,6 +42,8 @@ export const dummyBarContext2 = {bar: jest.fn()};
 
 export const promptCreatorConstructorMockAPI: TestingFrameworkAPICatalog = {
     context: {},
+    name: 'Test Framework',
+    description: 'A testing framework for unit testing purposes',
     categories: [
         {
             title: 'Actions',

src/types.ts

@@ -84,10 +84,14 @@ export interface TestingFrameworkDriver {
 
 /**
  * Represents the available API of the testing framework that can be used by Copilot.
+ * @property name Optional name of the testing framework (e.g. "Detox", "Jest", etc.).
+ * @property description Optional description of the testing framework's purpose and capabilities.
  * @property context The available variables of the testing framework (i.e. exposes the matching function, expect, etc.).
  * @property categories The available categories of the testing framework API.
  */
 export type TestingFrameworkAPICatalog = {
+    name?: string;
+    description?: string;
     context: any;
     categories: TestingFrameworkAPICatalogCategory[];
 }
@@ -145,8 +149,31 @@ export interface PromptHandler {
     isSnapshotImageSupported: () => boolean;
 }
 
+/**
+ * The cache mode for the Copilot.
+ *  - 'full': Cache is used with the screen state (default)
+ *  - 'lightweight': Cache is used but only based on steps (without screen state)
+ *  - 'disabled': No caching is used
+ * @default 'full'
+ */
+export type CacheMode = 'full' | 'lightweight' | 'disabled';
+
+/**
+ * Configuration options for the Copilot behavior.
+ */
+export interface CopilotOptions {
+    /**
+     * The cache mode to use.
+     * @default 'full'
+     */
+    cacheMode?: CacheMode;
+}
+
 /**
  * Configuration options for Copilot.
+ * @property frameworkDriver The testing driver to use for interacting with the underlying testing framework.
+ * @property promptHandler The prompt handler to use for interacting with the AI service
+ * @property options Additional options for configuring Copilot behavior
  */
 export interface Config {
     /**
@@ -158,6 +185,11 @@ export interface Config {
      * The prompt handler to use for interacting with the AI service
      */
     promptHandler: PromptHandler;
+
+    /**
+     * Additional options for configuring Copilot behavior
+     */
+    options?: CopilotOptions;
 }
 
 /**

src/utils/PromptCreator.test.ts

@@ -83,12 +83,12 @@ describe('PromptCreator', () => {
             {
                 step: 'navigate to login screen',
                 code: 'await element(by.id("login")).tap();',
-                result: undefined
+                result: 'success'
             },
             {
                 step: 'enter username',
                 code: 'await element(by.id("username")).typeText("john_doe");',
-                result: undefined
+                result: 'john doe'
             }
         ];
 

src/utils/PromptCreator.ts

@@ -50,13 +50,30 @@ export class PromptCreator {
     }
 
     private createBasePrompt(): string[] {
-        return [
+        const basePrompt = [
             "# Test Code Generation",
             "",
             "You are an AI assistant tasked with generating test code for an application using the provided UI testing framework API.",
             "Please generate the minimal executable code to perform the desired intent based on the given information and context.",
             ""
         ];
+
+        if (this.apiCatalog.name || this.apiCatalog.description) {
+            basePrompt.push("## Testing Framework");
+            basePrompt.push("");
+
+            if (this.apiCatalog.name) {
+                basePrompt.push(`Framework: ${this.apiCatalog.name}`);
+                basePrompt.push("");
+            }
+
+            if (this.apiCatalog.description) {
+                basePrompt.push(`Description: ${this.apiCatalog.description}`);
+                basePrompt.push("");
+            }
+        }
+
+        return basePrompt;
     }
 
     private createContext(
@@ -107,6 +124,7 @@ export class PromptCreator {
                     "```",
                     previousStep.code,
                     "```",
+                    ...(previousStep.result ? [`- Result: ${previousStep.result}`] : []),
                     ""
                 ]).flat(),
                 ""
@@ -216,6 +234,8 @@ export class PromptCreator {
         }
         steps.push(
             "If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.",
+            "Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.",
+            "Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.",
             "Wrap the generated code with backticks, without any additional formatting.",
             "Do not provide any additional code beyond the minimal executable code required to perform the intent."
         );

src/utils/__snapshots__/PromptCreator.test.ts.snap

@@ -6,6 +6,12 @@ exports[`PromptCreator constructor should merge redundant categories 1`] = `
 You are an AI assistant tasked with generating test code for an application using the provided UI testing framework API.
 Please generate the minimal executable code to perform the desired intent based on the given information and context.
 
+## Testing Framework
+
+Framework: Test Framework
+
+Description: A testing framework for unit testing purposes
+
 ## Context
 
 ### Intent to perform
@@ -123,8 +129,10 @@ Please follow these steps carefully:
 1. Analyze the provided intent and the view hierarchy to understand the required action.
 2. Generate the minimal executable code required to perform the intent using the available API.
 3. If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.
-4. Wrap the generated code with backticks, without any additional formatting.
-5. Do not provide any additional code beyond the minimal executable code required to perform the intent.
+4. Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.
+5. Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.
+6. Wrap the generated code with backticks, without any additional formatting.
+7. Do not provide any additional code beyond the minimal executable code required to perform the intent.
 
 ### Verify the prompt
 
@@ -251,8 +259,10 @@ Please follow these steps carefully:
 1. Analyze the provided intent and the view hierarchy to understand the required action.
 2. Generate the minimal executable code required to perform the intent using the available API.
 3. If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.
-4. Wrap the generated code with backticks, without any additional formatting.
-5. Do not provide any additional code beyond the minimal executable code required to perform the intent.
+4. Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.
+5. Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.
+6. Wrap the generated code with backticks, without any additional formatting.
+7. Do not provide any additional code beyond the minimal executable code required to perform the intent.
 
 ### Verify the prompt
 
@@ -393,8 +403,10 @@ Please follow these steps carefully:
 1. Analyze the provided intent and the view hierarchy to understand the required action.
 2. Generate the minimal executable code required to perform the intent using the available API.
 3. If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.
-4. Wrap the generated code with backticks, without any additional formatting.
-5. Do not provide any additional code beyond the minimal executable code required to perform the intent.
+4. Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.
+5. Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.
+6. Wrap the generated code with backticks, without any additional formatting.
+7. Do not provide any additional code beyond the minimal executable code required to perform the intent.
 
 ### Verify the prompt
 
@@ -509,8 +521,10 @@ Please follow these steps carefully:
 5. If the visual assertion fails, return code that throws an informative error explaining the failure.
 6. If visual validation is not possible, proceed to generate the minimal executable code required to perform the intent.
 7. If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.
-8. Wrap the generated code with backticks, without any additional formatting.
-9. Do not provide any additional code beyond the minimal executable code required to perform the intent.
+8. Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.
+9. Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.
+10. Wrap the generated code with backticks, without any additional formatting.
+11. Do not provide any additional code beyond the minimal executable code required to perform the intent.
 
 ### Verify the prompt
 
@@ -626,8 +640,10 @@ Please follow these steps carefully:
 1. Analyze the provided intent and the view hierarchy to understand the required action.
 2. Generate the minimal executable code required to perform the intent using the available API.
 3. If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.
-4. Wrap the generated code with backticks, without any additional formatting.
-5. Do not provide any additional code beyond the minimal executable code required to perform the intent.
+4. Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.
+5. Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.
+6. Wrap the generated code with backticks, without any additional formatting.
+7. Do not provide any additional code beyond the minimal executable code required to perform the intent.
 
 ### Verify the prompt
 
@@ -673,13 +689,15 @@ No snapshot image is attached for this intent.
 \`\`\`
 await element(by.id("login")).tap();
 \`\`\`
+- Result: success
 
 #### Step 2
 - Intent: "enter username"
 - Generated code:
 \`\`\`
 await element(by.id("username")).typeText("john_doe");
 \`\`\`
+- Result: john doe
 
 
 ## Available Testing Framework API
@@ -755,8 +773,10 @@ Please follow these steps carefully:
 1. Analyze the provided intent and the view hierarchy to understand the required action.
 2. Generate the minimal executable code required to perform the intent using the available API.
 3. If you cannot generate the relevant code due to ambiguity or invalid intent, return code that throws an informative error explaining the problem in one sentence.
-4. Wrap the generated code with backticks, without any additional formatting.
-5. Do not provide any additional code beyond the minimal executable code required to perform the intent.
+4. Each step must be completely independent - do not rely on any variables or assignments from previous steps. Even if a variable was declared or assigned in a previous step, you must redeclare and reassign it in your current step.
+5. Use the provided framework APIs as much as possible - prefer using the documented API methods over creating custom implementations.
+6. Wrap the generated code with backticks, without any additional formatting.
+7. Do not provide any additional code beyond the minimal executable code required to perform the intent.
 
 ### Verify the prompt