AI Config

Personal Cursor configuration with skills, agents, and rules for structured planning and implementation workflows.

Installation

Option 1: Global (Recommended)

Install once, available in all projects:

cd /Users/hardik/Projects/ai-config

# Copy everything to global Cursor config
cp -r .cursor/* ~/.cursor/

# Verify
ls -la ~/.cursor/
# Should see: skills/, agents/, rules/

Option 2: Project-Specific

Install in individual projects:

cd /Users/hardik/Projects/ai-config

# Copy to your project
cp -r .cursor/* /path/to/your-project/.cursor/

Post-Installation

Create required folders in each project where you'll use the skills:

cd /path/to/your-project
mkdir -p thoughts/shared/plans thoughts/shared/specs thoughts/shared/handoffs thoughts/ledgers

Open your project in Cursor and type / in chat -- you should see the available skills.

---

Quick Start

1. Create a Plan

In Cursor chat:

/create_plan

Or with a file reference:

/create_plan thoughts/tickets/feature-123.md

What happens:

1. AI asks what you're building
2. Spawns agents to research your codebase
3. Asks clarifying questions
4. Creates detailed plan at thoughts/shared/plans/YYYY-MM-DD-description.md

2. Implement the Plan

/implement_plan thoughts/shared/plans/YYYY-MM-DD-description.md

What happens:

1. Reads the plan and creates continuity ledger
2. For each task:
   • implement_task agent implements the work
   • task-review-agent reviews the code
   • Fixes issues if needed (max 3 iterations)
   • Creates handoff document
3. Moves to next task after approval
4. Survives context compaction (handoffs persist on disk)

Two modes:

• Direct (1-3 tasks): Implements in main context
• Agent Orchestration (4+ tasks): Agents handle each task

---

What's Included

Skills (8)

Skill	Description
/create_plan	Create detailed implementation plans through interactive research
/implement_plan	Execute plans with task tracking and review loops
/implement_task	Sub-skill used by implement_plan for individual tasks
/create_new_project_vercel_nextjs_plan	Plan greenfield Next.js + Vercel web apps
/implement_new_project_vercel_nextjs_plan	Execute Next.js + Vercel plans with architecture validation
/plan_python_vercel_backend_architect	Plan Python Flask + Vercel backend APIs
/implement_python_vercel_backend_architect	Execute Python Flask + Vercel backend plans
/debug_plan	Systematic debugging through structured investigation

Agents (8)

Planning & exploration:

• codebase-locator - Find files and components related to features
• codebase-analyzer - Deep dive into implementation details
• codebase-pattern-finder - Find similar implementations and patterns

Architecture specialists:

• nextjs-vercel-architect - Next.js + Vercel architecture validation and optimization
• python-vercel-backend-architect - Python Flask + Vercel backend architecture validation

Quality & validation:

• validate-agent - Validate plans against best practices
• review-agent - Review entire implementation (plan vs reality)
• task-review-agent - Review individual task implementations

Rules (3)

Rule	Applies To
react-native-standards.mdc	**/*.tsx, **/*.ts, **/*.swift, **/*.kt
nextjs-vercel-standards.mdc	**/*.tsx, **/*.ts, **/*.jsx, **/*.js, next.config.*, middleware.ts
python-vercel-backend-standards.mdc	Python Flask + Vercel backend projects

---

Workflow

The Full Cycle

Feature Request
      |
  /create_plan  --> Research --> Questions --> Plan Document
      |
/implement_plan --> Task 1 --> Review --> Fix --> Approved
                    Task 2 --> Review --> Fix --> Approved
                    Task 3 --> Review --> Fix --> Approved
                    ...
      |
   Feature Done

Example: Planning Phase

You: /create_plan

AI: I'll help you create a detailed implementation plan.
    What are you building?

You: Add biometric authentication to the login screen

AI: [Spawns codebase research agents...]
    I found the current auth implementation at src/auth/...

    Questions:
    - Should we support both Face ID and Touch ID?
    - What happens if biometric auth fails?
    - Should we keep password login as fallback?

You: [Answer questions...]

AI: [Creates detailed plan with phases, code examples, success criteria]
    Plan created at: thoughts/shared/plans/2026-02-11-biometric-auth.md

Example: Implementation Phase

You: /implement_plan thoughts/shared/plans/2026-02-11-biometric-auth.md

AI: I'll use agent orchestration for this plan (5 tasks).

    Creating continuity ledger: thoughts/ledgers/biometric-auth.md
    Setting up handoffs: thoughts/shared/handoffs/biometric-auth/

    ─────────────────────────────────────
    Task 1 of 5: Create biometric auth module
    ─────────────────────────────────────

    Implementing...
    [Spawns implement_task agent]
    [Creates handoff: task-01-biometric-module.md]

    Reviewing (iteration 1)...
    [Spawns task-review-agent]
    [Review: APPROVED]

    Task 1 approved. Moving to Task 2...

    [Continues for all 5 tasks...]

Behind the Scenes

During /create_plan:

1. codebase-locator finds relevant files
2. codebase-analyzer understands current implementation
3. codebase-pattern-finder finds similar patterns
4. AI asks clarifying questions
5. Creates detailed plan with phases, code examples, and success criteria

During /implement_plan:

1. Creates continuity ledger and handoff directory
2. For each task: Implement -> Review -> Fix (if needed) -> Approve -> Next
3. Each task creates a handoff document that persists on disk
4. Ledger tracks progress across sessions

Generated Artifacts

thoughts/
├── shared/
│   ├── plans/
│   │   └── 2026-02-11-push-notifications.md
│   └── handoffs/
│       └── push-notifications/
│           ├── task-01-ios-apns-setup.md
│           ├── task-02-android-fcm-setup.md
│           └── ...
└── ledgers/
    └── push-notifications.md

---

Next.js + Vercel Projects

For greenfield Next.js websites deployed on Vercel:

Plan

/create_new_project_vercel_nextjs_plan

What happens:

1. Analyzes your Next.js template structure
2. Uses nextjs-vercel-architect agent for validation
3. Deep requirements interview (content, UX, performance, SEO)
4. Creates comprehensive 5-phase plan optimized for Vercel

Implement

/implement_new_project_vercel_nextjs_plan thoughts/shared/plans/2026-02-11-nextjs-project.md

Additional checks beyond the standard workflow:

• TypeScript strict mode (npx tsc --noEmit)
• ESLint verification (npm run lint)
• Build verification (npm run build)
• Lighthouse audit (90+ required)
• Core Web Vitals check
• Bundle size verification (< 100kb)
• SEO completeness check
• Vercel deployment readiness

See .cursor/skills/create_new_project_vercel_nextjs_plan/README.md for detailed usage.

---

Python Flask + Vercel Backend

For greenfield Python Flask APIs deployed on Vercel:

Plan

/plan_python_vercel_backend_architect

Implement

/implement_python_vercel_backend_architect thoughts/shared/plans/YYYY-MM-DD-project.md

See .cursor/skills/plan_python_vercel_backend_architect/README.md for detailed usage.

---

Debugging

For systematic bug investigation:

/debug_plan

Walks through structured investigation, root cause analysis, and fix planning.

---

Tips

Better Plans

• Provide ticket files or detailed descriptions
• Answer questions thoroughly
• Mention constraints upfront (deadlines, platform differences)

Smooth Implementation

• Trust the process -- let agents do their work
• Read handoffs if you need to resume (thoughts/shared/handoffs/)
• Check the ledger for current state (thoughts/ledgers/)

Context Limits

If context clears mid-implementation, just run the same command again:

/implement_plan thoughts/shared/plans/YYYY-MM-DD-description.md

AI reads the ledger and handoffs, continues from where it stopped.

---

Troubleshooting

Skill not showing up

1. Check file structure: ls -la ~/.cursor/skills/create_plan/ (should contain SKILL.md)
2. Restart Cursor completely
3. Check frontmatter in SKILL.md has description: field

Agents not working

1. Verify agents exist: ls -la ~/.cursor/agents/
2. Check agent files have proper YAML frontmatter with name: and description:

Folder not found errors

mkdir -p thoughts/shared/plans thoughts/shared/specs thoughts/shared/handoffs thoughts/ledgers

Review keeps rejecting

After 3 iterations, you'll be asked to continue with issues, fix manually, or skip task.

---

Folder Structure

ai-config/
├── readme.md
└── .cursor/
    ├── skills/
    │   ├── create_plan/
    │   ├── create_new_project_vercel_nextjs_plan/
    │   ├── implement_plan/
    │   ├── implement_new_project_vercel_nextjs_plan/
    │   ├── implement_task/
    │   ├── plan_python_vercel_backend_architect/
    │   ├── implement_python_vercel_backend_architect/
    │   └── debug_plan/
    ├── agents/
    │   ├── codebase-locator.md
    │   ├── codebase-analyzer.md
    │   ├── codebase-pattern-finder.md
    │   ├── nextjs-vercel-architect.md
    │   ├── python-vercel-backend-architect.md
    │   ├── validate-agent.md
    │   ├── review-agent.md
    │   └── task-review-agent.md
    └── rules/
        ├── react-native-standards.mdc
        ├── nextjs-vercel-standards.mdc
        └── python-vercel-backend-standards.mdc

---

Updating

To get latest changes after pulling this repo:

cd /Users/hardik/Projects/ai-config

# Reinstall globally
cp -r .cursor/* ~/.cursor/

# Or to a specific project
cp -r .cursor/* /path/to/your-project/.cursor/

Adding Custom Rules

Create additional rules following this pattern:

---
description: Your custom standards
globs: ["**/*.tsx", "**/*.ts"]
alwaysApply: false
---

# Your Standards

Your conventions and patterns here...