Registry Database

registry-database Skill

Teaches how to work with the Registry platform - a Convex-backed database that serves as single source of truth for all data (tasks, articles, courses, emails, coding projects, etc.).

When to Use This Skill

Use this skill when the user mentions:

• registry, convex, database, schema
• tasks, todos, task tracking, high-level tasks
• articles, article status, article metadata
• courses, quizzes, students, teaching data
• emails, outlook sync, email management
• MCP server tools, registry tools
• coding projects, session tracking
• morning review, unified dashboard

Core Concepts

Registry Platform Architecture

Purpose: Single source of truth for all personal and professional data.

Tech Stack:

• Convex: Real-time database with TypeScript schema
• Bun: Runtime for MCP server
• React: Web dashboard (planned)
• MCP: Model Context Protocol server for OpenCode integration

Location: ~/Code/github.com/theslyprofessor/registry/

Database Schema (10 Tables)

Article Slug as ID

Critical concept: Article slugs ARE the articleId in registry.

// Vault file: midimaze/Music Theory/Intervals.md
// Frontmatter: slug: intervals
// Registry: articleId = "intervals"

Why:

• Slugs are unique, human-readable identifiers
• Used in web URLs for midimaze 2.0
• Natural link between vault and database

Article Status (6 States)

draft → in_progress → needs_review → complete → published → archived

NOT the same as HIL:

• article_status: Content workflow state
• hil: Human-in-loop decision tracking

Task Status & Priority

Status: pending, in_progress, completed, cancelled

Priority: low, medium, high

Purpose: Registry tracks high-level tasks and ideas across all domains (teaching, coding, personal). For AI coding tasks that need multi-session work, use Beads (separate system) or OpenSpec change proposals.

MCP Server Tools

Location: registry/packages/mcp/src/index.ts

Core Tools (Available Now)

Planned Tools

• Email sync/management (Outlook → Registry)
• Course/quiz creation and management
• Article sync from vault (frontmatter → registry)
• Morning review dashboard queries
• Recording session tracking
• Coding project management

Common Workflows

1. Create a Registry

// User: "Create a new task registry for the NNT compiler project"

registry_create_registry({
  name: "nnt-compiler-tasks",
  type: "tasks",
  username: "ntiruviluamala",
  visibility: "private"
})

2. Add Tasks

// User: "Add a task to implement the lexer"

registry_create_task({
  content: "Implement lexer for NNT notation",
  registryId: "<registry-id>",
  priority: "high",
  projectContext: "nnt-compiler",
  tags: ["parser", "core"]
})

3. Sync OpenCode Todos

// User: "Sync my current todos to registry"

registry_sync_todos({
  todos: [...], // From todoRead
  registryId: "<registry-id>",
  projectContext: "current-project"
})

4. Query Tasks

// User: "Show me all in-progress tasks for the registry project"

registry_get_tasks({
  registryId: "<registry-id>",
  status: "in_progress",
  projectContext: "registry"
})

Development Workflows

Start Convex Dev Server

cd ~/Code/github.com/theslyprofessor/registry
bun convex dev

What happens:

• Connects to Convex cloud project
• Watches convex/schema.ts for changes
• Auto-deploys on save
• Provides dashboard URL

Modify Schema

File: convex/schema.ts

// Add new field to tasks table
tasks: defineTable({
  // ... existing fields
  estimatedHours: v.optional(v.number()),
})

Deploy: Just save - auto-deploys in dev mode

No migrations needed: Optional fields are graceful

Test MCP Server Locally

cd ~/Code/github.com/theslyprofessor/registry/packages/mcp
bun run src/index.ts
# Runs in stdio mode (for MCP protocol)

Add New MCP Tool

1. Define tool in schema:

   // packages/mcp/src/index.ts
   server.tool({
     name: "create_article",
     description: "Create article metadata in registry",
     input: z.object({
       slug: z.string(),
       title: z.string(),
       // ...
     }),
     execute: async (params) => {
       // Implementation
     }
   })
   

2. Create Convex mutation:

   // convex/articles.ts
   export const create = mutation({
     args: { /* ... */ },
     handler: async (ctx, args) => {
       // Database logic
     }
   })
   

3. Test in OpenCode: Tools auto-discover on restart

Data Flow Architecture

Vault (Obsidian markdown)
    ↓ frontmatter sync (planned)
Registry (Convex database)
    ↓ MCP tools
OpenCode (AI coding assistant)
    ↓ queries
Morning Review Dashboard (planned)

External Integrations:

• Outlook: Email sync via outlook-sqlite-mcp (planned)
• Canvas LMS: Course/quiz data (planned)
• GitHub: Coding project metadata (planned)
• Vault: Article frontmatter sync (planned)

Schema Evolution Guidelines

Adding Optional Fields

✅ Safe: Just add to schema, deploy

tasks: defineTable({
  // ... existing
  newField: v.optional(v.string()),
})

Adding Required Fields

⚠️ Requires migration:

// 1. Add as optional first
newField: v.optional(v.string()),

// 2. Write migration mutation
export const backfillNewField = mutation({
  handler: async (ctx) => {
    // Populate existing records
  }
})

// 3. Make required after backfill
newField: v.string(),

Adding Tables

✅ Safe: Add to schema, start using

gear_inventory: defineTable({
  model: v.string(),
  category: v.string(),
  // ...
})

Common Issues

Convex Connection Errors

Problem: "Failed to connect to Convex"

Solutions:

• Check CONVEX_URL in .env.local
• Verify bun convex dev is running
• Check internet connection
• Re-authenticate: bunx convex login

MCP Server Not Discoverable

Problem: Tools don't appear in OpenCode

Solutions:

• Check ~/.config/opencode/opencode.json has registry config
• Restart OpenCode to reload MCP servers
• Test server directly: bun run packages/mcp/src/index.ts
• Check server logs for errors

Schema Validation Errors

Problem: "Document doesn't match schema"

Solutions:

• Check all required fields are provided
• Verify field types match schema (string vs number)
• Use v.optional() for nullable fields
• Check array types: v.array(v.string())

Task Creation Failures

Problem: "Failed to create task"

Solutions:

• Verify registryId exists (use list_registries first)
• Check required fields are provided (content, status, priority)
• Verify Convex connection is active
• Check projectContext is a valid string

Integration with Other Systems

OpenCode Sessions

Purpose: Track conversation history, link to tasks

Flow:

OpenCode starts → Create session record
Task created → Link to current sessionId
Session ends → Update stats (duration, tokens, files)

Vault Articles

Purpose: Sync frontmatter metadata to registry

Flow (Planned):

Article saved → Extract frontmatter
Check slug exists → Update or create
Sync article_status → Track workflow
Link to tasks/courses → Bidirectional references

Outlook Email

Purpose: Track correspondence, link to courses/articles

Flow (Planned):

Email received → Create registry record
Parse subject → Detect course/student
Link to courseId → Organize by context
Flag for follow-up → Task creation

Configuration

MCP Server Config: ~/.config/opencode/opencode.json

{
  "mcp": {
    "registry": {
      "type": "local",
      "command": [
        "bun",
        "run",
        "/Users/ntiruviluamala/Code/github.com/theslyprofessor/registry/packages/mcp/src/index.ts"
      ],
      "enabled": true,
      "env": {
        "CONVEX_URL": "https://..."
      }
    }
  }
}

Convex Config: registry/convex.json

{
  "project": "registry-project",
  "team": "theslyprofessor",
  "prodUrl": "https://..."
}

Best Practices

Task Management

1. Use projectContext: Group related tasks (e.g., "nnt-compiler", "registry", "swc-teaching")
2. Set priority appropriately: Reserve "high" for urgent work
3. Link to articles: Use articleId (slug) for documentation context
4. Update status frequently: Keep registry current with task progress
5. High-level only: Registry is for ideas and high-level tasks, not detailed AI coding tasks (use Beads/OpenSpec for those)

Article Tracking

1. Use slug as ID: Never change slugs (breaks links)
2. Track status progression: draft → ... → published
3. Set visibility correctly: Private for _ folders
4. Link related content: Tasks, courses, emails

Schema Design

1. Start with optional fields: Easier to evolve
2. Use references (IDs): Link between tables
3. Index query fields: Add .index() for common filters
4. Document relationships: Comment foreign keys

MCP Tool Design

1. Clear descriptions: Help OpenCode choose right tool
2. Validate inputs: Use Zod schemas strictly
3. Return useful data: Include IDs for follow-up actions
4. Handle errors gracefully: Return error messages, not exceptions

Project Structure

registry/
├── convex/                    # Convex backend
│   ├── schema.ts             # Database schema (10 tables)
│   ├── tasks.ts              # Task mutations/queries
│   ├── articles.ts           # Article operations
│   └── ...                   # Other table logic
├── packages/
│   └── mcp/                  # MCP server
│       └── src/
│           └── index.ts      # Tool definitions
├── web/                      # React dashboard (planned)
├── convex.json               # Convex project config
└── package.json              # Dependencies

See Also

• Related Skills:

  • obsidian-workflows - Vault management and article creation
  • swc-teaching - Course/quiz workflows
  • (Note: For AI coding tasks, see Beads or OpenSpec - separate systems)

• Related AGENTS.md:

  • ~/Code/github.com/theslyprofessor/registry/AGENTS.md - Full project docs

• External Resources:

  • Convex Documentation
  • MCP Specification

Quick Reference

Essential Commands

# Start development
cd ~/Code/github.com/theslyprofessor/registry
bun convex dev

# Test MCP server
cd packages/mcp && bun run src/index.ts

# Deploy schema changes
# (Auto-deploys on save in dev mode)

# Re-authenticate
bunx convex login

Key Concepts

• Slug = Article ID (always)
• article_status ≠ HIL (different workflows)
• Tasks link to articles (via articleId/slug)
• High-level task tracking (detailed AI tasks → Beads/OpenSpec)
• Optional fields are safe (no migrations needed)
• MCP tools auto-discover (restart OpenCode)