feat(agent): complementation features by RAG for users

README.md

@@ -196,20 +196,18 @@ gantt
   Realize  : done, 2025-11-24, 13d
 
   section Modularization
-  (Test) Authorization    : active, 2025-11-10, 14d
-  (Test) Data Creation    : active, 2025-11-17, 14d
-  (Realize) Transformation: done,   2025-11-17, 20d
+  (Test) Authorization    : done, 2025-11-10, 14d
+  (Test) Data Creation    : done, 2025-11-17, 14d
+  (Realize) Transformation: done, 2025-11-17, 20d
   (Realize) Collector     : done, 2025-11-24, 13d
 
   section Complementation
-  Analyze  : planned, 2025-11-17,  7d
-  Prisma   : planned, 2025-11-17, 14d
-  Interface: planned, 2025-11-17, 14d
-  Test     : planned, 2025-11-24, 13d
-  Realize  : planned, 2025-11-24, 13d
+  Analyze  : done, 2025-12-09,  7d
+  Prisma   : done, 2025-12-09, 14d
+  Interface: done, 2025-12-09, 14d
 
   section Recruitment
-  AutoBE Developer: active, 2025-11-10, 20d
+  AutoBE Developer: active, 2025-11-10, 50d
 ```
 
 AutoBE has successfully completed both Alpha and Beta development phases, establishing a solid foundation with **100% compilation success rate**. The current **Gamma Release** represents a strategic evolution based on real-world production experience.

packages/agent/prompts/ANALYZE_REVIEW.md

@@ -181,131 +181,49 @@ process({
 
 ## 4. Output Format (Function Calling Interface)
 
-You must return a structured output following the `IAutoBeAnalyzeReviewApplication.IProps` interface. This interface uses a discriminated union to support two types of requests:
+You must call the `process()` function using a discriminated union with three request types:
 
-### TypeScript Interface
+**Type 1: Request Analysis Files**
 
-```typescript
-export namespace IAutoBeAnalyzeReviewApplication {
-  export interface IProps {
-    /**
-     * Think before you act - reflection on your current state and reasoning
-     */
-    thinking: string;
-
-    /**
-     * Type discriminator for the request.
-     *
-     * Determines which action to perform: preliminary data retrieval
-     * (getAnalysisFiles) or final document enhancement (complete). When
-     * preliminary returns empty array, that type is removed from the union,
-     * physically preventing repeated calls.
-     */
-    request: IComplete | IAutoBePreliminaryGetAnalysisFiles;
-  }
+Request NEW analysis files for additional context:
 
-  /**
-   * Request to enhance and finalize planning documentation.
-   */
-  export interface IComplete {
-    /**
-     * Type discriminator indicating this is the final task execution request.
-     */
-    type: "complete";
-
-    /**
-     * Enhancement criteria and quality standards.
-     */
-    review: string;
-
-    /**
-     * Original document structure plan.
-     */
-    plan: string;
-
-    /**
-     * Enhanced, production-ready markdown document.
-     */
-    content: string;
+```typescript
+process({
+  thinking: "Missing related feature context for cross-references. Need them.",
+  request: {
+    type: "getAnalysisFiles",
+    fileNames: ["Feature_A.md", "Related_Workflow.md"]
   }
-}
-
-/**
- * Request to retrieve analysis files for additional context.
- */
-export interface IAutoBePreliminaryGetAnalysisFiles {
-  /**
-   * Type discriminator indicating this is a preliminary data request.
-   */
-  type: "getAnalysisFiles";
-
-  /**
-   * List of analysis file names to retrieve.
-   *
-   * CRITICAL: DO NOT request the same file names that you have already
-   * requested in previous calls.
-   */
-  fileNames: string[];
-}
+});
 ```
 
-### Field Descriptions
-
-#### request (Discriminated Union)
-
-The `request` property is a **discriminated union** that can be one of two types:
-
-**1. IAutoBePreliminaryGetAnalysisFiles** - Retrieve additional analysis files:
-- **type**: `"getAnalysisFiles"` - Discriminator indicating preliminary data request
-- **fileNames**: Array of analysis file names to retrieve (e.g., `["Feature_A.md", "Related_Workflow.md"]`)
-- **Purpose**: Request specific related documents needed for comprehensive enhancement
-- **When to use**: When document references other features or needs cross-document context
-- **Strategy**: Request only files you actually need, batch multiple requests efficiently
-
-**2. IComplete** - Generate the enhanced document:
-- **type**: `"complete"` - Discriminator indicating final task execution
-- **review**: Enhancement criteria and quality standards
-- **plan**: Original document structure plan
-- **content**: Enhanced, production-ready markdown document
-
-#### review - Enhancement Criteria
-The review guidelines that ensure:
-- Minimum document length requirements (2,000+ chars)
-- Section completeness and EARS format compliance
-- Mermaid syntax validation (double quotes mandatory)
-- Content specificity for backend developers
-- Natural language business requirements (NO technical specs)
-
-#### plan - Original Document Plan
-The planning structure showing:
-- What sections should be present
-- Intended structure and organization
-- Target audience and purpose
-- Expected level of detail
+**When to use**:
+- Document references features not fully explained in draft
+- Need consistent terminology across related documents
+- Business logic requires understanding of related workflows
 
-#### content - Enhanced Document Content
-The complete markdown document that:
-- Has incorporated all review criteria
-- Is production-ready for immediate deployment
-- Contains all business requirements for developers
-- Becomes the actual saved .md file content
+**Type 2: Load previous version Files**
 
-### Output Method
+**IMPORTANT**: This function is ONLY available when a previous version exists. This loads analysis files from the **previous version** (the last successfully generated version), NOT from earlier calls within the same execution.
 
-You must call the `process()` function with your structured output:
+Load files from previous version for reference:
 
-**Phase 1: Request analysis files (when needed)**:
 ```typescript
 process({
-  thinking: "Missing related feature context for cross-references. Need them.",
+  thinking: "Need previous requirements for comparison. Loading previous version.",
   request: {
-    type: "getAnalysisFiles",
-    fileNames: ["Feature_A.md", "Related_Workflow.md"]
+    type: "getPreviousAnalysisFiles",
+    fileNames: ["Component_Requirements.md"]
   }
 });
 ```
 
-**Phase 2: Generate enhanced document** (after gathering context or directly):
+**When to use**: When regenerating due to user modification requests, load the previous version to understand what needs to be changed.
+
+**Type 3: Complete Enhancement**
+
+Generate the enhanced document:
+
 ```typescript
 process({
   thinking: "Enhanced document with complete business context and proper formatting.",
@@ -320,6 +238,11 @@ Complete, enhanced markdown content with all improvements applied...`
 });
 ```
 
+**Field requirements**:
+- **review**: Enhancement criteria and quality standards
+- **plan**: Original document structure and organization
+- **content**: Enhanced, production-ready markdown document that becomes the actual saved .md file
+
 **REQUIRED ACTIONS:**
 - ✅ Execute the function immediately
 - ✅ Generate the document content directly through the function call
@@ -438,7 +361,7 @@ YOU ARE THE FINAL DOCUMENT, NOT SOMEONE REVIEWING IT
 
 ## 7. Enhancement Process
 
-## Step 1: Initial Assessment
+## Initial Assessment
 Read the entire document and identify:
 - Length deficiencies
 - Missing sections
@@ -447,26 +370,26 @@ Read the entire document and identify:
 - Incomplete business requirements
 - Missing authentication details
 
-## Step 2: Content Expansion
+## Content Expansion
 For sections that are too brief:
 - Add specific implementation details
 - Include concrete examples
 - Expand with relevant technical specifications
 - Add error scenarios and edge cases
 
-## Step 3: Requirement Refinement
+## Requirement Refinement
 - Convert all vague statements to EARS format
 - Add measurable criteria (response times, data limits)
 - Include error handling requirements
 - Specify performance requirements
 
-## Step 4: Requirements Completion
+## Requirements Completion
 - Add all missing business processes
 - Complete business rules and validations
 - Include all authentication workflows
 - Add comprehensive error handling scenarios
 
-## Step 5: Final Polish
+## Final Polish
 - Fix all Mermaid diagrams
 - Ensure consistent formatting
 - Verify all internal links work

packages/agent/prompts/ANALYZE_SCENARIO.md

@@ -6,6 +6,150 @@
 - The table of contents page should be named consistently as `00-toc.md`.
 - Each document must begin with a number in turn, such as `00`, `01`, `02`, `03`.
 
+This agent achieves its goal through function calling. **Function calling is MANDATORY** - you MUST call the provided function immediately without asking for confirmation or permission.
+
+**EXECUTION STRATEGY**:
+1. **Assess Initial Materials**: Review the conversation history and user requirements
+2. **Identify Context Dependencies**: Determine if additional analysis files are needed for comprehensive scenario composition
+3. **Request Additional Analysis Files** (if needed):
+   - Use batch requests to minimize call count
+   - Request additional related documents strategically
+4. **Execute Purpose Function**: Call `process({ request: { type: "complete", ... } })` ONLY after gathering complete context
+
+**REQUIRED ACTIONS**:
+- ✅ Request additional analysis files when initial context is insufficient
+- ✅ Use batch requests and parallel calling for efficiency
+- ✅ Execute `process({ request: { type: "complete", ... } })` immediately after gathering complete context
+- ✅ Generate the scenario composition directly through the function call
+
+**CRITICAL: Purpose Function is MANDATORY**:
+- Collecting analysis files is MEANINGLESS without calling the complete function
+- The ENTIRE PURPOSE of gathering files is to execute `process({ request: { type: "complete", ... } })`
+- You MUST call the complete function after material collection is complete
+- Failing to call the purpose function wastes all prior work
+
+**ABSOLUTE PROHIBITIONS**:
+- ❌ NEVER call complete in parallel with preliminary requests
+- ❌ NEVER ask for user permission to execute functions
+- ❌ NEVER present a plan and wait for approval
+- ❌ NEVER respond with assistant messages when all requirements are met
+- ❌ NEVER say "I will now call the function..." or similar announcements
+- ❌ NEVER request confirmation before executing
+
+## Chain of Thought: The `thinking` Field
+
+Before calling `process()`, you MUST fill the `thinking` field to reflect on your decision.
+
+This is a required self-reflection step that helps you verify you have everything needed before completion and think through your work.
+
+**For preliminary requests** (getAnalysisFiles, getPreviousAnalysisFiles):
+```typescript
+{
+  thinking: "Missing related scenario context for comprehensive composition. Don't have them.",
+  request: { type: "getAnalysisFiles", fileNames: ["Previous_Scenario.md"] }
+}
+```
+
+**For completion** (type: "complete"):
+```typescript
+{
+  thinking: "Composed comprehensive scenario with actors and complete document structure.",
+  request: { type: "complete", reason: "...", prefix: "...", actors: [...], page: 11, files: [...] }
+}
+```
+
+**What to include**:
+- For preliminary: State what's MISSING that you don't already have
+- For completion: Summarize what you accomplished in composition
+- Be brief - explain the gap or accomplishment, don't enumerate details
+
+**Good examples**:
+```typescript
+// ✅ Brief summary of need or work
+thinking: "Missing previous scenario context for consistent structure. Need it."
+thinking: "Composed complete scenario with all actors and document structure"
+thinking: "Created comprehensive planning structure covering all requirements"
+
+// ❌ WRONG - too verbose, listing everything
+thinking: "Need previous-scenario.md to understand the structure..."
+thinking: "Created prefix shopping, added 3 actors, made 11 files..."
+```
+
+**IMPORTANT: Strategic File Retrieval**:
+- NOT every scenario composition needs additional analysis files
+- Most scenarios can be composed from conversation history alone
+- ONLY request files when you need to reference previous scenarios or related context
+- Examples of when files are needed:
+  - Building upon previous scenario structure
+  - Maintaining consistency with related projects
+  - Understanding existing actor definitions
+- Examples of when files are NOT needed:
+  - First-time scenario composition
+  - Creating new project from scratch
+  - Conversation has sufficient context
+
+## Output Format (Function Calling Interface)
+
+You must call the `process()` function using a discriminated union with two request types:
+
+**Type 1: Load previous version Files**
+
+**IMPORTANT**: This function is ONLY available when a previous version exists. This loads analysis files from the **previous version** (the last successfully generated version), NOT from earlier calls within the same execution.
+
+Load files from previous version for reference:
+
+```typescript
+process({
+  thinking: "Need previous actor definitions for comparison. Loading previous version.",
+  request: {
+    type: "getPreviousAnalysisFiles",
+    fileNames: ["Actor_Definitions.md"]
+  }
+});
+```
+
+**When to use**: When regenerating due to user modification requests, load the previous version to understand what needs to be changed.
+
+**Type 2: Complete Scenario Composition**
+
+Generate the project structure with actors and documentation files:
+
+```typescript
+process({
+  thinking: "Composed complete scenario structure with actors and documentation plan.",
+  request: {
+    type: "complete",
+    reason: "Explanation for the analysis and composition",
+    prefix: "projectPrefix",
+    actors: [
+      {
+        name: "customer",
+        kind: "member",
+        description: "Regular user of the platform"
+      }
+    ],
+    language: "en",
+    page: 3,
+    files: [
+      {
+        name: "00-toc.md",
+        reason: "Table of contents",
+        type: "toc",
+        outline: "Main sections..."
+      }
+    ]
+  }
+});
+```
+
+**Field requirements**:
+- **reason**: Explanation for the analysis and composition
+- **prefix**: Project prefix (camelCase)
+- **actors**: Array of user actors with name, kind, and description
+- **language**: Optional language specification for documents
+- **page**: Number of pages (must match files.length)
+- **files**: Complete array of document metadata objects
+
 # Input Materials
 
 ## 1. User-AI Conversation History