Implement prompt generator and schema parser for prompt engineering (#23)

Signed-off-by: Dae❤️ <74119677+daeisbae@users.noreply.github.com>

Author: daeisbae

src/agent/structured-output/prompt-generator.ts

@@ -0,0 +1,86 @@
+import { PromptTemplate } from '@langchain/prompts'
+
+/**
+ * Configuration interface for PromptTemplate
+ * @interface PromptTemplateConfig
+ */
+export interface PromptTemplateConfig {
+    template: string
+    inputVariables: string[]
+    partialVariables: Instruction
+}
+
+/**
+ * Interface for formatting instructions
+ * @interface Instruction
+ */
+export interface Instruction {
+    formatInstructions: string
+}
+
+/**
+ * Enum defining types of prompts that can be generated
+ * @enum {string}
+ */
+export enum PromptType {
+    Folder = 'folder',
+    File = 'file',
+}
+
+/**
+ * Class responsible for generating prompts based on different PromptTypes
+ * @class PromptGenerator
+ */
+export class PromptGenerator {
+    protected template: PromptTemplate
+    protected promptType: PromptType
+
+    /**
+     * Creates an instance of PromptGenerator
+     * @example
+     * const generator = new PromptGenerator({
+     *     template: "The following instruction is given:\n{formatInstructions}\nBelow is the code for your task: {code}",
+     *     inputVariables: ["code"],
+     *     partialVariables: {
+     *         formatInstructions: CodePrompt // Import CodePrompt from prompt.ts
+     *     }}, PromptType.File);
+     * @example
+     * const generator = new PromptGenerator({
+     *    template: "The following instruction is given:\n{formatInstructions}\nBelow are the AI summaries for the codebase:\n{ai_summaries}",
+     *   inputVariables: ["ai_summaries"],
+     *  partialVariables: {
+     *     formatInstructions: FolderPrompt // Import FolderPrompt from prompt.ts
+     * }}, PromptType.Folder);
+     * @param {PromptTemplateConfig} config - Configuration for the prompt template
+     * @param {PromptType} promptType - Type of prompt to generate
+     */
+    constructor(config: PromptTemplateConfig, promptType: PromptType) {
+        this.template = new PromptTemplate({
+            ...config,
+        })
+        this.promptType = promptType
+    }
+
+    /**
+     * Generates a formatted prompt based on the prompt type and inputs
+     * @param {string} [code] - Optional code string for file prompts
+     * @param {string[]} [ai_summaries] - Optional array of AI summaries for folder prompts
+     * @returns {Promise<string>} Formatted prompt string
+     * @throws {Error} If prompt type is invalid or required input is missing
+     */
+    async generate(code?: string, ai_summaries?: string[]): Promise<string> {
+        const promptMap: Record<PromptType, string> = {
+            [PromptType.Folder]: ai_summaries?.join('\n') ?? '',
+            [PromptType.File]: code ?? '',
+        }
+
+        const userPrompt: string = promptMap[this.promptType]
+        if (!userPrompt.length) {
+            throw new Error(
+                'Invalid prompt type or missing input for prompt generation'
+            )
+        }
+
+        return this.template.format( this.promptType === PromptType.File ? { code: userPrompt } : { ai_summaries: userPrompt }) 
+    }
+}

src/agent/structured-output/schema-parser.ts

@@ -0,0 +1,25 @@
+import { StructuredOutputParser } from '@/langchain/output_parsers'
+import { ZodSchema } from 'zod'
+
+
+/**
+ * Parser class that converts LLM outputs into structured objects
+ * @class SchemaParser
+ */
+export class SchemaParser {
+    private outputParser: StructuredOutputParser
+    private formatInstructions: string
+
+    constructor(schema: ZodSchema) {
+        this.outputParser = new StructuredOutputParser(schema)
+        this.formatInstructions = this.outputParser.getFormalInstructions()
+    }
+
+    get formalInstructions() {
+        return this.formatInstructions;
+    }
+
+    async parse(output: string): Promise<Object> {
+        return await this.outputParser.parse(output)
+    }
+}