@grec0/memory-bank-mcp 0.1.41 → 0.2.1

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
package/README.md CHANGED
@@ -59,6 +59,13 @@ With Memory Bank, AIs:
59
59
  - **🆔 Identity Management**: Tracks who is doing what (GitHub Copilot, Cursor, etc.)
60
60
  - **🔒 Atomic Locks**: File-system based locking safe across different processes/IDEs
61
61
 
62
+ ### Task Orchestration (Smart Routing) 🧭 NEW
63
+ - **🎯 Intelligent Routing**: Analyzes tasks BEFORE implementation to determine ownership
64
+ - **📋 Enriched Project Registry**: Projects have responsibilities, ownership, and exports metadata
65
+ - **🤖 AI Reasoning**: Uses reasoning models to distribute work across projects
66
+ - **🔀 Auto-Delegation**: Automatically identifies what should be delegated to other projects
67
+ - **📦 Import Suggestions**: Recommends what to import from other projects instead of reimplementing
68
+
62
69
  ## 📋 Requirements
63
70
 
64
71
  - **Node.js** >= 18.0.0
@@ -294,13 +301,16 @@ Reads generated documentation.
294
301
  1. Index code
295
302
  memorybank_index_code({ projectId: "my-project" })
296
303
 
297
- 2. Generate documentation
304
+ 2. Generate documentation (also updates global registry)
298
305
  memorybank_generate_project_docs({ projectId: "my-project" })
299
306
 
300
307
  3. Query documentation at the start of each session
301
308
  memorybank_get_project_docs({ projectId: "my-project", document: "activeContext" })
302
309
 
303
- 4. Search specific code
310
+ 4. Route task BEFORE implementing (mandatory in auto-index mode)
311
+ memorybank_route_task({ projectId: "my-project", taskDescription: "..." })
312
+
313
+ 5. Search specific code
304
314
  memorybank_search({ projectId: "my-project", query: "..." })
305
315
  ```
306
316
 
@@ -308,6 +318,21 @@ Reads generated documentation.
308
318
 
309
319
  If you configure `MEMORYBANK_AUTO_UPDATE_DOCS=true`, documents will be automatically regenerated after each indexing. This is useful for keeping documentation always up to date but consumes more API tokens.
310
320
 
321
+ ### Upgrading Existing Projects 🆕
322
+
323
+ If you have projects already initialized with a previous version, simply regenerate the docs to enable Task Orchestration:
324
+
325
+ ```json
326
+ // For each existing project:
327
+ memorybank_generate_project_docs({ "projectId": "your-project", "force": true })
328
+ ```
329
+
330
+ This will:
331
+ 1. Regenerate all 6 markdown documents
332
+ 2. **NEW**: Extract responsibilities, ownership, and exports
333
+ 3. **NEW**: Update `global_registry.json` with enriched metadata
334
+ 4. Enable `memorybank_route_task` to work with this project
335
+
311
336
  ---
312
337
 
313
338
  ## 🤖 Multi-Agent Coordination
@@ -382,6 +407,71 @@ memorybank_delegate_task({
382
407
  })
383
408
  ```
384
409
 
410
+ ### Task Orchestration (Smart Routing) 🧭 NEW
411
+
412
+ The **Task Orchestrator** analyzes tasks BEFORE implementation to prevent agents from creating code that belongs to other projects.
413
+
414
+ #### Why is this needed?
415
+
416
+ Without orchestration, agents often:
417
+ - ❌ Create DTOs in the API project when `lib-dtos` exists
418
+ - ❌ Duplicate utilities that are already in `shared-utils`
419
+ - ❌ Implement features that belong to other microservices
420
+ - ❌ Violate architectural boundaries unknowingly
421
+
422
+ With the orchestrator:
423
+ - ✅ Know exactly what belongs to this project
424
+ - ✅ Automatically delegate work to the right project
425
+ - ✅ Get import suggestions instead of reimplementing
426
+ - ✅ Respect ecosystem boundaries
427
+
428
+ #### How It Works
429
+
430
+ 1. **Enriched Registry**: When you run `memorybank_generate_project_docs`, it automatically extracts:
431
+ - `responsibilities`: What this project is responsible for
432
+ - `owns`: Files/folders that belong to this project
433
+ - `exports`: What this project provides to others
434
+ - `projectType`: api, library, frontend, backend, etc.
435
+
436
+ 2. **Route Before Implementing**: Call `memorybank_route_task` BEFORE any code changes:
437
+ ```json
438
+ memorybank_route_task({
439
+ "projectId": "my-api",
440
+ "taskDescription": "Create DTOs for user management and expose REST endpoints"
441
+ })
442
+ ```
443
+
444
+ 3. **Orchestrator Response**:
445
+ ```json
446
+ {
447
+ "action": "partial_delegate",
448
+ "myResponsibilities": [
449
+ "Create REST endpoints in src/controllers/",
450
+ "Implement business logic in src/services/"
451
+ ],
452
+ "delegations": [
453
+ {
454
+ "targetProjectId": "lib-dtos",
455
+ "taskTitle": "Create UserDTO and UserResponseDTO",
456
+ "reason": "DTOs belong to lib-dtos per project responsibilities"
457
+ }
458
+ ],
459
+ "suggestedImports": [
460
+ "import { UserDTO } from 'lib-dtos'"
461
+ ],
462
+ "architectureNotes": "Use shared DTOs to maintain consistency across services"
463
+ }
464
+ ```
465
+
466
+ #### Possible Actions
467
+
468
+ | Action | Meaning |
469
+ |--------|--------|
470
+ | `implement_here` | Everything belongs to this project, proceed |
471
+ | `delegate_all` | Nothing belongs here, delegate everything |
472
+ | `partial_delegate` | Some parts belong here, delegate the rest |
473
+ | `needs_clarification` | Task is ambiguous, ask user for details |
474
+
385
475
  ---
386
476
 
387
477
  ## 🔀 Multi-Project: Cross-Project Queries
@@ -524,9 +614,36 @@ Analyzes project indexing coverage.
524
614
  }
525
615
  ```
526
616
 
617
+ ### `memorybank_route_task` 🆕
618
+
619
+ Analyzes a task and determines what belongs to this project vs what should be delegated. **MUST be called BEFORE any implementation.**
620
+
621
+ **Parameters:**
622
+ - `projectId` **(REQUIRED)**: Project requesting the routing
623
+ - `taskDescription` **(REQUIRED)**: Detailed description of what needs to be implemented
624
+
625
+ **Example:**
626
+ ```json
627
+ {
628
+ "projectId": "my-api",
629
+ "taskDescription": "Create user registration endpoint with validation and DTOs"
630
+ }
631
+ ```
632
+
633
+ **Response:**
634
+ ```json
635
+ {
636
+ "action": "partial_delegate",
637
+ "myResponsibilities": ["Create POST /users endpoint", "Add validation middleware"],
638
+ "delegations": [{ "targetProjectId": "lib-dtos", "taskTitle": "Create UserDTO" }],
639
+ "suggestedImports": ["import { UserDTO } from 'lib-dtos'"],
640
+ "architectureNotes": "Follow REST conventions, use shared DTOs"
641
+ }
642
+ ```
643
+
527
644
  ### `memorybank_generate_project_docs`
528
645
 
529
- Generates structured project documentation using AI with reasoning.
646
+ Generates structured project documentation using AI with reasoning. **Also automatically updates the global registry with enriched project metadata** (responsibilities, owns, exports, projectType).
530
647
 
531
648
  **Parameters:**
532
649
  - `projectId` **(REQUIRED)**: Project identifier
@@ -721,11 +838,13 @@ Memory Bank includes instruction templates in two formats to configure agent beh
721
838
  **For active development with automatic synchronization.**
722
839
 
723
840
  - ✅ Agent consults Memory Bank automatically
841
+ - ✅ **Routes tasks before implementing** (Rule 0.5)
724
842
  - ✅ Reindexes EVERY file after modifying it
725
843
  - ✅ Keeps Memory Bank always up to date
726
844
  - ✅ Can read/write files directly
845
+ - ✅ **Auto-delegates to other projects** when appropriate
727
846
 
728
- **Ideal for**: Active development, rapid iteration, teams.
847
+ **Ideal for**: Active development, rapid iteration, teams, **multi-project ecosystems**.
729
848
 
730
849
  ### 3. Sandboxed Mode
731
850
 
@@ -889,9 +1008,37 @@ Claude: [executes memorybank_generate_project_docs({ projectId: "my-app" })]
889
1008
  - techContext.md: Tech stack
890
1009
  ...
891
1010
 
1011
+ Also updated the global registry with project responsibilities.
892
1012
  You can query them anytime with memorybank_get_project_docs
893
1013
  ```
894
1014
 
1015
+ ### 5. Task Orchestration (Multi-Project) 🆕
1016
+
1017
+ ```
1018
+ User: I need to add a UserDTO for the new registration endpoint
1019
+ Claude: [executes memorybank_route_task({
1020
+ projectId: "my-api",
1021
+ taskDescription: "Create UserDTO for registration endpoint"
1022
+ })]
1023
+
1024
+ The orchestrator analyzed your ecosystem and determined:
1025
+
1026
+ ❌ DTOs should NOT be created in my-api
1027
+ ✅ DTOs belong to lib-dtos project
1028
+
1029
+ I'll delegate the DTO creation to lib-dtos and import it:
1030
+
1031
+ [executes memorybank_delegate_task({
1032
+ projectId: "my-api",
1033
+ targetProjectId: "lib-dtos",
1034
+ title: "Create UserDTO",
1035
+ description: "DTO for user registration with email, password fields"
1036
+ })]
1037
+
1038
+ Task delegated! Once lib-dtos creates the DTO, you can:
1039
+ import { UserDTO } from 'lib-dtos'
1040
+ ```
1041
+
895
1042
  ---
896
1043
 
897
1044
  ## 🔧 Configuration Files
@@ -1008,6 +1008,72 @@ ${chunk.content}
1008
1008
  const files = fs.readdirSync(docsPath);
1009
1009
  return files.filter(f => f.endsWith(".md"));
1010
1010
  }
1011
+ // ==========================================
1012
+ // Project Summary Generation (for Registry)
1013
+ // ==========================================
1014
+ /**
1015
+ * Generates a structured summary of the project for the global registry.
1016
+ * Extracts responsibilities, ownership patterns, and project type from projectBrief.
1017
+ * This is called automatically after generateAllDocuments.
1018
+ */
1019
+ async generateProjectSummary(projectId) {
1020
+ const projectBrief = this.getDocument(projectId, "projectBrief");
1021
+ const techContext = this.getDocument(projectId, "techContext");
1022
+ if (!projectBrief) {
1023
+ console.error(`Cannot generate project summary: projectBrief not found for ${projectId}`);
1024
+ return null;
1025
+ }
1026
+ console.error(`Generating project summary for registry: ${projectId}`);
1027
+ const prompt = `Analyze the following project documentation and extract a structured summary.
1028
+
1029
+ PROJECT BRIEF:
1030
+ ${projectBrief.content}
1031
+
1032
+ ${techContext ? `TECHNICAL CONTEXT:\n${techContext.content}` : ''}
1033
+
1034
+ Extract the following information in JSON format:
1035
+ {
1036
+ "description": "A concise 1-2 sentence description of what this project does",
1037
+ "responsibilities": ["List of 3-5 specific things this project is responsible for"],
1038
+ "owns": ["File patterns or types this project owns, e.g., '*DTO.ts', 'services/', 'controllers/'"],
1039
+ "projectType": "One of: api, library, frontend, backend, cli, service, monorepo, fullstack",
1040
+ "exports": "Package name if it's a library (e.g., '@company/lib-dtos'), or null if not applicable",
1041
+ "keywords": ["5-8 keywords describing this project"]
1042
+ }
1043
+
1044
+ IMPORTANT:
1045
+ - Be specific about responsibilities - they help the orchestrator decide where code belongs
1046
+ - For "owns", think about what file patterns ONLY this project should create
1047
+ - If it's a library, identify what it exports/provides to other projects
1048
+
1049
+ Respond ONLY with the JSON object, no markdown or explanation.`;
1050
+ try {
1051
+ const result = await this.callResponsesAPI(prompt);
1052
+ // Parse JSON response
1053
+ const jsonMatch = result.content.match(/\{[\s\S]*\}/);
1054
+ if (!jsonMatch) {
1055
+ console.error(`Failed to parse project summary JSON for ${projectId}`);
1056
+ return null;
1057
+ }
1058
+ const summary = JSON.parse(jsonMatch[0]);
1059
+ console.error(`Project summary generated for ${projectId}:`);
1060
+ console.error(` - Type: ${summary.projectType}`);
1061
+ console.error(` - Responsibilities: ${summary.responsibilities?.length || 0}`);
1062
+ console.error(` - Owns: ${summary.owns?.length || 0}`);
1063
+ return {
1064
+ description: summary.description || '',
1065
+ responsibilities: summary.responsibilities || [],
1066
+ owns: summary.owns || [],
1067
+ projectType: summary.projectType || 'unknown',
1068
+ exports: summary.exports || undefined,
1069
+ keywords: summary.keywords || [],
1070
+ };
1071
+ }
1072
+ catch (error) {
1073
+ console.error(`Error generating project summary: ${error}`);
1074
+ return null;
1075
+ }
1076
+ }
1011
1077
  }
1012
1078
  /**
1013
1079
  * Creates a Project Knowledge Service from environment variables
@@ -22,10 +22,10 @@ export class RegistryManager {
22
22
  async saveRegistry(registry) {
23
23
  await fs.writeFile(this.globalPath, JSON.stringify(registry, null, 2), 'utf-8');
24
24
  }
25
- async registerProject(projectId, workspacePath, description, keywords = [], embeddingService) {
25
+ async registerProject(projectId, workspacePath, description, keywords = [], embeddingService, enhancedInfo) {
26
26
  const registry = await this.ensureRegistry();
27
27
  const idx = registry.projects.findIndex(p => p.projectId === projectId);
28
- // Preserve existing description/keywords if not provided
28
+ // Preserve existing data if not provided
29
29
  const existing = idx >= 0 ? registry.projects[idx] : null;
30
30
  const card = {
31
31
  projectId,
@@ -33,7 +33,12 @@ export class RegistryManager {
33
33
  description: description || existing?.description || '',
34
34
  keywords: keywords.length > 0 ? keywords : (existing?.keywords || []),
35
35
  lastActive: new Date().toISOString(),
36
- status: 'ACTIVE'
36
+ status: 'ACTIVE',
37
+ // Enhanced fields - preserve existing if not provided
38
+ responsibilities: enhancedInfo?.responsibilities || existing?.responsibilities,
39
+ owns: enhancedInfo?.owns || existing?.owns,
40
+ exports: enhancedInfo?.exports || existing?.exports,
41
+ projectType: enhancedInfo?.projectType || existing?.projectType,
37
42
  };
38
43
  if (idx >= 0) {
39
44
  registry.projects[idx] = card;
@@ -53,12 +58,30 @@ export class RegistryManager {
53
58
  }
54
59
  }
55
60
  async updateProjectEmbedding(card, embeddingService) {
56
- const text = `Project: ${card.projectId}\nDescription: ${card.description || ''}\nKeywords: ${card.keywords.join(', ')}`;
61
+ // Build rich text for embedding including responsibilities
62
+ const textParts = [
63
+ `Project: ${card.projectId}`,
64
+ `Description: ${card.description || ''}`,
65
+ `Keywords: ${card.keywords.join(', ')}`,
66
+ ];
67
+ if (card.responsibilities && card.responsibilities.length > 0) {
68
+ textParts.push(`Responsibilities: ${card.responsibilities.join('. ')}`);
69
+ }
70
+ if (card.owns && card.owns.length > 0) {
71
+ textParts.push(`Owns: ${card.owns.join(', ')}`);
72
+ }
73
+ if (card.projectType) {
74
+ textParts.push(`Type: ${card.projectType}`);
75
+ }
76
+ if (card.exports) {
77
+ textParts.push(`Exports: ${card.exports}`);
78
+ }
79
+ const text = textParts.join('\n');
57
80
  const result = await embeddingService.generateEmbedding(card.projectId, text);
58
81
  await this.projectVectorStore.upsertProject({
59
82
  id: card.projectId,
60
83
  vector: result.vector,
61
- name: card.projectId, // Using ID as name for now if name not available
84
+ name: card.projectId,
62
85
  description: card.description || '',
63
86
  tags: card.keywords,
64
87
  path: card.path,
@@ -101,6 +124,14 @@ export class RegistryManager {
101
124
  const registry = await this.ensureRegistry();
102
125
  return registry.projects.find(p => p.projectId === projectId);
103
126
  }
127
+ /**
128
+ * Gets all projects from the registry.
129
+ * Useful for the orchestrator to analyze responsibilities across all projects.
130
+ */
131
+ async getAllProjects() {
132
+ const registry = await this.ensureRegistry();
133
+ return registry.projects;
134
+ }
104
135
  /**
105
136
  * Syncs all projects from the JSON registry to the vector store.
106
137
  * Useful for migrating existing projects to the new semantic discovery system.
@@ -1,2 +1,2 @@
1
1
  // Version of the MCP Kanban server
2
- export const VERSION = "0.1.39";
2
+ export const VERSION = "0.2.1";
package/dist/index.js CHANGED
@@ -83,6 +83,7 @@ import { manageAgentsTool, manageAgentsToolDefinition } from "./tools/manageAgen
83
83
  import { discoverProjectsTool, discoverProjectsToolDefinition } from "./tools/discoverProjects.js";
84
84
  import { delegateTaskTool, delegateTaskToolDefinition } from "./tools/delegateTask.js";
85
85
  import { syncProjectsTool, syncProjectsToolDefinition } from "./tools/syncProjects.js";
86
+ import { routeTaskTool, routeTaskToolDefinition } from "./tools/routeTask.js";
86
87
  import { RegistryManager } from "./common/registryManager.js";
87
88
  import { VERSION } from "./common/version.js";
88
89
  // Global services
@@ -864,6 +865,19 @@ server.tool(delegateTaskToolDefinition.name, delegateTaskToolDefinition.descript
864
865
  content: [{ type: "text", text: JSON.stringify(result, null, 2) }],
865
866
  };
866
867
  });
868
+ // Tool: Route Task (Orchestrator)
869
+ server.tool(routeTaskToolDefinition.name, routeTaskToolDefinition.description, {
870
+ projectId: z.string().describe("ID del proyecto que está solicitando el enrutamiento"),
871
+ taskDescription: z.string().describe("Descripción detallada de la tarea a implementar")
872
+ }, async (args) => {
873
+ const result = await routeTaskTool({
874
+ projectId: args.projectId,
875
+ taskDescription: args.taskDescription
876
+ });
877
+ return {
878
+ content: [{ type: "text", text: JSON.stringify(result, null, 2) }],
879
+ };
880
+ });
867
881
  /**
868
882
  * Starts the stdio server
869
883
  */
@@ -900,6 +914,7 @@ async function startStdioServer() {
900
914
  console.error(" - memorybank_complete_task: Marcar una tarea como completada");
901
915
  console.error(" - memorybank_discover_projects: Descubrir otros proyectos en el ecosistema");
902
916
  console.error(" - memorybank_delegate_task: Delegar tareas a otros proyectos");
917
+ console.error(" - memorybank_route_task: Orquestador - analiza y distribuye tareas");
903
918
  console.error("");
904
919
  console.error("Available resources:");
905
920
  console.error(" - memory://{projectId}/active: Contexto activo");
@@ -4,6 +4,8 @@
4
4
  */
5
5
  import { AgentBoard } from "../common/agentBoard.js";
6
6
  import { sessionLogger } from "../common/sessionLogger.js";
7
+ import { RegistryManager } from "../common/registryManager.js";
8
+ import { EmbeddingService } from "../common/embeddingService.js";
7
9
  import { sessionState } from "../common/sessionState.js";
8
10
  /**
9
11
  * Generates project documentation using AI reasoning
@@ -94,6 +96,41 @@ export async function generateProjectDocs(params, projectKnowledgeService, vecto
94
96
  console.error(` - Reasoning tokens: ${result.totalReasoningTokens}`);
95
97
  console.error(` - Output tokens: ${result.totalOutputTokens}`);
96
98
  console.error(` - Estimated cost: $${totalCost.toFixed(4)}`);
99
+ // === AUTO-UPDATE GLOBAL REGISTRY ===
100
+ // Generate project summary and update registry with enriched info
101
+ if (result.documentsGenerated.length > 0 || result.documentsUpdated.length > 0) {
102
+ try {
103
+ console.error(`\n=== Updating Global Registry ===`);
104
+ const summary = await projectKnowledgeService.generateProjectSummary(projectId);
105
+ if (summary) {
106
+ const registryManager = new RegistryManager();
107
+ let embeddingService;
108
+ if (process.env.OPENAI_API_KEY) {
109
+ try {
110
+ embeddingService = new EmbeddingService(process.env.OPENAI_API_KEY);
111
+ }
112
+ catch (e) {
113
+ console.error(`Warning: Could not init embedding service: ${e}`);
114
+ }
115
+ }
116
+ await registryManager.registerProject(projectId, workspaceRoot, summary.description, summary.keywords, embeddingService, {
117
+ responsibilities: summary.responsibilities,
118
+ owns: summary.owns,
119
+ exports: summary.exports,
120
+ projectType: summary.projectType,
121
+ });
122
+ console.error(`Registry updated for ${projectId}:`);
123
+ console.error(` - Description: ${summary.description.slice(0, 80)}...`);
124
+ console.error(` - Type: ${summary.projectType}`);
125
+ console.error(` - Responsibilities: ${summary.responsibilities.length}`);
126
+ message += ` Registry updated with project summary.`;
127
+ }
128
+ }
129
+ catch (regError) {
130
+ console.error(`Warning: Failed to update registry: ${regError}`);
131
+ // Don't fail the whole operation if registry update fails
132
+ }
133
+ }
97
134
  return {
98
135
  success: result.success,
99
136
  message,
@@ -15,3 +15,5 @@ export * from "./syncProjects.js";
15
15
  export * from "./generateProjectDocs.js";
16
16
  export * from "./getProjectDocs.js";
17
17
  export * from "./manageAgents.js";
18
+ // Export Task Orchestration tools
19
+ export * from "./routeTask.js";
@@ -0,0 +1,302 @@
1
+ /**
2
+ * @fileoverview Task Routing Orchestrator
3
+ * Uses AI reasoning to analyze tasks and distribute work across projects
4
+ * based on their responsibilities. MANDATORY before any implementation.
5
+ */
6
+ import OpenAI from "openai";
7
+ import { RegistryManager } from "../common/registryManager.js";
8
+ /**
9
+ * Builds a context string describing all projects and their responsibilities
10
+ */
11
+ function buildProjectsContext(projects, currentProjectId) {
12
+ if (projects.length === 0) {
13
+ return "No other projects registered in the workspace.";
14
+ }
15
+ let context = "## Registered Projects in Workspace\n\n";
16
+ for (const project of projects) {
17
+ const isCurrent = project.projectId === currentProjectId;
18
+ context += `### ${project.projectId}${isCurrent ? ' (CURRENT - requesting agent)' : ''}\n`;
19
+ context += `- **Path**: ${project.path}\n`;
20
+ context += `- **Type**: ${project.projectType || 'unknown'}\n`;
21
+ context += `- **Description**: ${project.description || 'No description'}\n`;
22
+ if (project.responsibilities && project.responsibilities.length > 0) {
23
+ context += `- **Responsibilities**:\n`;
24
+ for (const resp of project.responsibilities) {
25
+ context += ` - ${resp}\n`;
26
+ }
27
+ }
28
+ if (project.owns && project.owns.length > 0) {
29
+ context += `- **Owns (file patterns)**: ${project.owns.join(', ')}\n`;
30
+ }
31
+ if (project.exports) {
32
+ context += `- **Exports**: ${project.exports}\n`;
33
+ }
34
+ context += '\n';
35
+ }
36
+ return context;
37
+ }
38
+ /**
39
+ * The main routing prompt for the AI orchestrator
40
+ */
41
+ const ROUTING_PROMPT = `You are a Task Routing Orchestrator for a multi-project workspace. Your job is to analyze a task and determine which parts belong to which project based on their responsibilities.
42
+
43
+ {projectsContext}
44
+
45
+ ## Task to Analyze
46
+ **From Project**: {currentProject}
47
+ **Task Description**: {taskDescription}
48
+
49
+ ## Your Analysis
50
+
51
+ Analyze the task and determine:
52
+ 1. What components/code need to be created or modified?
53
+ 2. Which project is responsible for each component based on their declared responsibilities?
54
+ 3. If something doesn't exist in any project, it can be created by the requesting project
55
+ 4. If something SHOULD exist in another project (based on responsibilities), it must be delegated
56
+
57
+ ## Rules
58
+ - If a project is responsible for DTOs, ALL DTOs must be created there, not in the API
59
+ - If a project is responsible for services, shared services go there
60
+ - If a project is responsible for utils/common code, shared utilities go there
61
+ - The requesting project can ONLY implement what falls within its responsibilities
62
+ - When in doubt, check the "owns" patterns to see what file types belong where
63
+
64
+ ## Response Format
65
+ Respond with a JSON object:
66
+ {
67
+ "action": "proceed" | "delegate" | "mixed",
68
+ "myResponsibilities": ["List of things the requesting project should implement"],
69
+ "delegations": [
70
+ {
71
+ "targetProject": "project-id",
72
+ "taskTitle": "Short title for the task",
73
+ "taskDescription": "Detailed description of what to create",
74
+ "reasoning": "Why this belongs to this project"
75
+ }
76
+ ],
77
+ "suggestedImports": ["packages or modules to import after delegations complete"],
78
+ "architectureNotes": "Explanation of the distribution decision",
79
+ "warning": "Optional warning if something seems off"
80
+ }
81
+
82
+ IMPORTANT:
83
+ - "action" is "proceed" if everything can be done by the requesting project
84
+ - "action" is "delegate" if everything needs to go to other projects
85
+ - "action" is "mixed" if some work is local and some needs delegation
86
+ - Be specific in taskDescription so the receiving project knows exactly what to create
87
+ - Always explain the reasoning based on project responsibilities
88
+
89
+ Respond ONLY with the JSON object.`;
90
+ /**
91
+ * Routes a task to the appropriate project(s) based on responsibilities
92
+ */
93
+ export async function routeTaskTool(params) {
94
+ const { projectId, taskDescription } = params;
95
+ if (!taskDescription || taskDescription.trim() === '') {
96
+ return {
97
+ success: false,
98
+ action: 'proceed',
99
+ myResponsibilities: [],
100
+ delegations: [],
101
+ suggestedImports: [],
102
+ architectureNotes: 'No task description provided.',
103
+ warning: 'Please provide a task description to analyze.',
104
+ };
105
+ }
106
+ console.error(`\n=== Task Routing Orchestrator ===`);
107
+ console.error(`Project: ${projectId}`);
108
+ console.error(`Task: ${taskDescription.slice(0, 100)}...`);
109
+ // Get all projects with their responsibilities
110
+ const registryManager = new RegistryManager();
111
+ const allProjects = await registryManager.getAllProjects();
112
+ console.error(`Found ${allProjects.length} projects in registry`);
113
+ // Check if we have any projects with responsibilities defined
114
+ const projectsWithResponsibilities = allProjects.filter(p => p.responsibilities && p.responsibilities.length > 0);
115
+ if (projectsWithResponsibilities.length === 0) {
116
+ console.error(`Warning: No projects have responsibilities defined`);
117
+ return {
118
+ success: true,
119
+ action: 'proceed',
120
+ myResponsibilities: ['All components (no other project responsibilities defined)'],
121
+ delegations: [],
122
+ suggestedImports: [],
123
+ architectureNotes: 'No other projects have responsibilities defined. You can proceed with the full implementation. Consider running memorybank_generate_project_docs on other projects to define their responsibilities.',
124
+ warning: 'No project responsibilities found in workspace. Run generate_project_docs on all projects first.',
125
+ };
126
+ }
127
+ // Build context for the AI
128
+ const projectsContext = buildProjectsContext(allProjects, projectId);
129
+ // Call AI to analyze and route
130
+ const apiKey = process.env.OPENAI_API_KEY;
131
+ if (!apiKey) {
132
+ return {
133
+ success: false,
134
+ action: 'proceed',
135
+ myResponsibilities: [],
136
+ delegations: [],
137
+ suggestedImports: [],
138
+ architectureNotes: 'OPENAI_API_KEY not configured. Cannot analyze task routing.',
139
+ warning: 'AI routing unavailable. Proceeding without validation.',
140
+ };
141
+ }
142
+ try {
143
+ const client = new OpenAI({ apiKey });
144
+ const prompt = ROUTING_PROMPT
145
+ .replace('{projectsContext}', projectsContext)
146
+ .replace('{currentProject}', projectId)
147
+ .replace('{taskDescription}', taskDescription);
148
+ console.error(`Calling AI orchestrator...`);
149
+ // Use reasoning model for better analysis
150
+ const model = process.env.MEMORYBANK_REASONING_MODEL || "gpt-5-mini";
151
+ const response = await client.responses.create({
152
+ model,
153
+ reasoning: {
154
+ effort: "medium",
155
+ },
156
+ input: [
157
+ {
158
+ role: "user",
159
+ content: prompt,
160
+ },
161
+ ],
162
+ max_output_tokens: 4000,
163
+ });
164
+ // Extract content from response
165
+ let content = "";
166
+ for (const item of response.output || []) {
167
+ if (item.type === "message" && item.content) {
168
+ for (const contentItem of item.content) {
169
+ if (contentItem.type === "output_text") {
170
+ content += contentItem.text;
171
+ }
172
+ }
173
+ }
174
+ }
175
+ // Parse JSON response
176
+ const jsonMatch = content.match(/\{[\s\S]*\}/);
177
+ if (!jsonMatch) {
178
+ console.error(`Failed to parse orchestrator response`);
179
+ return {
180
+ success: false,
181
+ action: 'proceed',
182
+ myResponsibilities: [],
183
+ delegations: [],
184
+ suggestedImports: [],
185
+ architectureNotes: 'Failed to parse AI response.',
186
+ warning: 'Orchestrator analysis failed. Review task manually.',
187
+ };
188
+ }
189
+ const result = JSON.parse(jsonMatch[0]);
190
+ console.error(`\nOrchestrator Decision:`);
191
+ console.error(` Action: ${result.action}`);
192
+ console.error(` My responsibilities: ${result.myResponsibilities?.length || 0}`);
193
+ console.error(` Delegations: ${result.delegations?.length || 0}`);
194
+ if (result.delegations && result.delegations.length > 0) {
195
+ console.error(`\n Delegations:`);
196
+ for (const d of result.delegations) {
197
+ console.error(` → ${d.targetProject}: ${d.taskTitle}`);
198
+ }
199
+ }
200
+ return {
201
+ success: true,
202
+ action: result.action || 'proceed',
203
+ myResponsibilities: result.myResponsibilities || [],
204
+ delegations: result.delegations || [],
205
+ suggestedImports: result.suggestedImports || [],
206
+ architectureNotes: result.architectureNotes || '',
207
+ warning: result.warning,
208
+ };
209
+ }
210
+ catch (error) {
211
+ console.error(`Error in task routing: ${error.message}`);
212
+ // Fallback to chat completions if responses API fails
213
+ if (error?.status === 404 || error?.code === "model_not_found") {
214
+ try {
215
+ const client = new OpenAI({ apiKey });
216
+ const prompt = ROUTING_PROMPT
217
+ .replace('{projectsContext}', projectsContext)
218
+ .replace('{currentProject}', projectId)
219
+ .replace('{taskDescription}', taskDescription);
220
+ const response = await client.chat.completions.create({
221
+ model: "gpt-4o",
222
+ messages: [
223
+ {
224
+ role: "system",
225
+ content: "You are a Task Routing Orchestrator. Analyze tasks and route them to the appropriate projects based on responsibilities. Respond with JSON only.",
226
+ },
227
+ {
228
+ role: "user",
229
+ content: prompt,
230
+ },
231
+ ],
232
+ max_tokens: 4000,
233
+ });
234
+ const content = response.choices[0]?.message?.content || "";
235
+ const jsonMatch = content.match(/\{[\s\S]*\}/);
236
+ if (jsonMatch) {
237
+ const result = JSON.parse(jsonMatch[0]);
238
+ return {
239
+ success: true,
240
+ action: result.action || 'proceed',
241
+ myResponsibilities: result.myResponsibilities || [],
242
+ delegations: result.delegations || [],
243
+ suggestedImports: result.suggestedImports || [],
244
+ architectureNotes: result.architectureNotes || '',
245
+ warning: result.warning,
246
+ };
247
+ }
248
+ }
249
+ catch (fallbackError) {
250
+ console.error(`Fallback also failed: ${fallbackError}`);
251
+ }
252
+ }
253
+ return {
254
+ success: false,
255
+ action: 'proceed',
256
+ myResponsibilities: [],
257
+ delegations: [],
258
+ suggestedImports: [],
259
+ architectureNotes: `Error analyzing task: ${error.message}`,
260
+ warning: 'Orchestrator failed. Review task distribution manually.',
261
+ };
262
+ }
263
+ }
264
+ /**
265
+ * Tool definition for MCP
266
+ */
267
+ export const routeTaskToolDefinition = {
268
+ name: "memorybank_route_task",
269
+ description: `🚨 OBLIGATORIO antes de implementar cualquier código.
270
+
271
+ Analiza una tarea y determina qué partes corresponden a qué proyecto según sus responsabilidades.
272
+
273
+ El orquestador:
274
+ 1. Lee las responsabilidades de TODOS los proyectos del workspace
275
+ 2. Analiza qué componentes necesita la tarea (DTOs, services, controllers, etc.)
276
+ 3. Asigna cada componente al proyecto responsable
277
+ 4. Devuelve un plan de acción con delegaciones
278
+
279
+ DEBES llamar esta herramienta ANTES de escribir código para evitar:
280
+ - Crear DTOs en un API cuando existe una lib-dtos
281
+ - Duplicar services que ya existen en otro proyecto
282
+ - Violar la separación de responsabilidades
283
+
284
+ La respuesta incluye:
285
+ - myResponsibilities: Lo que TÚ debes implementar
286
+ - delegations: Tareas a delegar a otros proyectos
287
+ - suggestedImports: Dependencias a usar tras las delegaciones`,
288
+ inputSchema: {
289
+ type: "object",
290
+ properties: {
291
+ projectId: {
292
+ type: "string",
293
+ description: "ID del proyecto que solicita el análisis (tu proyecto actual)",
294
+ },
295
+ taskDescription: {
296
+ type: "string",
297
+ description: "Descripción completa de la tarea a realizar",
298
+ },
299
+ },
300
+ required: ["projectId", "taskDescription"],
301
+ },
302
+ };
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@grec0/memory-bank-mcp",
3
- "version": "0.1.41",
3
+ "version": "0.2.1",
4
4
  "description": "MCP server for semantic code indexing with Memory Bank - AI-powered codebase understanding",
5
5
  "license": "MIT",
6
6
  "author": "@grec0",