WebSmartTeam

feature-dev

0
0
# Install this skill:
npx skills add WebSmartTeam/COR-CODE --skill "feature-dev"

Install specific skill from multi-skill repository

# Description

Implement new features in existing projects using 7-phase workflow (Discovery → Explore → Clarify → Design → Implement → Review → Summary). Use when adding functionality, building new components, creating API endpoints, or extending existing systems. Triggers: new feature, implement, add functionality, build feature, extend, create endpoint, add component.

# SKILL.md


name: feature-dev
description: Implement new features in existing projects using 7-phase workflow (Discovery → Explore → Clarify → Design → Implement → Review → Summary). Use when adding functionality, building new components, creating API endpoints, or extending existing systems. Triggers: new feature, implement, add functionality, build feature, extend, create endpoint, add component.
updated: 2025-01-18
user-invocable: true
allowed-tools:
- Read
- Write
- Edit
- MultiEdit
- Bash
- Glob
- Grep
- TodoWrite
- Task
- AskUserQuestion
- mcp__supabase__
- mcp__context7__

- mcp__sequential-thinking__
- mcp__firecrawl__

- mcp__magic__*


Feature Development for Existing Projects

Purpose: Add new features to established projects using a structured workflow. No tech stack assumptions - reads from CLAUDE.md and existing code.

Why No context: fork

This skill deliberately keeps the main conversation context because:
- You already know the codebase from previous work
- Forking loses that accumulated understanding
- Sub-agents via Task tool can explore and report back
- Main conversation retains decision history

7-Phase Workflow

Phase 1: Discovery

Goal: Understand what the user wants to build

  • Listen to the feature request
  • Identify core requirements vs nice-to-haves
  • Note any constraints mentioned (performance, security, but timeline is nonsense,usual developer cost for this work to charge client)
  • Tech stack will be discovered in Phase 2 from CLAUDE.md and codebase

Output: Clear feature scope statement

Phase 2: Codebase Exploration

Goal: Understand the existing system and its tech stack

First: Read CLAUDE.md and discover the project's actual setup:
- Framework (Next.js, Remix, Nuxt, plain React, etc.)
- Database (Supabase, Prisma, Drizzle, raw SQL, etc.)
- Hosting (Vercel, AWS, Netlify, etc.)
- Auth provider (Supabase Auth, NextAuth, Clerk, etc.)
- Styling (Tailwind, CSS modules, styled-components, etc.)

Then: Use Task tool sub-agents in parallel to explore relevant areas:

Task (Explore): "Find authentication patterns - how does this project handle auth?"
Task (Explore): "Identify database schema - what ORM/client, what tables exist?"
Task (Explore): "Map API routes - REST, tRPC, Server Actions, GraphQL?"
Task (Explore): "Find UI patterns - component library, design tokens, layouts"

Understand:
- How existing features are structured
- Naming conventions and file organisation
- Error handling patterns
- State management approach

Output: Clear picture of project tech stack and patterns to follow

Phase 3: Clarifying Questions

Goal: Fill gaps before designing

Ask only what you genuinely don't know:
- Business logic ambiguities
- Edge cases and error handling preferences
- Integration priorities
- Performance requirements

Rule: Don't ask about tech choices already in CLAUDE.md or codebase

Output: Complete understanding to design solution

Phase 4: Architecture Design

Goal: Design the implementation approach

Document in TodoWrite:
- Database changes (if any)
- API endpoints needed
- UI components to create/modify
- Integration points with existing code
- Migration strategy (if schema changes)

Consider:
- Follows existing patterns in codebase
- Maintains consistency with current architecture
- Handles errors like existing code does
- Security model matches project standards

Output: Clear implementation plan in todos

Phase 5: Implementation

Goal: Build the feature

Execute todos systematically:
1. Database/schema changes first (if any)
2. Backend/API layer
3. Frontend/UI layer
4. Integration and wiring
5. Tests (matching project's test patterns)

Rules:
- One logical change at a time
- Commit after each working milestone
- Follow existing code patterns exactly
- Use project's established libraries

Phase 6: Quality Review

Goal: Verify the implementation

Check against:
- Original requirements from Phase 1
- Existing patterns (does it match the codebase style?)
- Error handling (does it follow project conventions?)
- Security (authentication, authorisation, input validation)
- Performance (no obvious N+1 queries, proper caching)

If visual changes, use available tools:
- Firecrawl: scrape with screenshot format for quick captures
- Playwright: E2E testing and detailed visual regression
- Chrome tabs: colour picker, console, network (requires claude --chrome session)

Output: Verified, working feature

Phase 7: Summary

Goal: Document what was done

Provide:
- What was built (feature summary)
- Files created/modified
- Database changes made
- Any new environment variables needed
- Testing instructions
- Follow-up improvements (if any)

Usage

Invoke with feature description:

/feature-dev Add a user notification system with email and in-app alerts

Or naturally:

"I need to add a new dashboard section for analytics - can you use the feature-dev workflow?"

Key Principles

  1. Discover, don't assume - Read CLAUDE.md and explore codebase to understand actual tech stack
  2. Adapt to project - Use whatever framework, database, styling the project already has
  3. Keep context - Main conversation remembers everything, sub-agents report back
  4. Follow patterns - New code should look like it belongs in the existing codebase
  5. Ask smart questions - Only what you can't determine from existing code
  6. Incremental commits - Working milestones, not one big commit

# Supported AI Coding Agents

This skill is compatible with the SKILL.md standard and works with all major AI coding agents:

Learn more about the SKILL.md standard and how to use these skills with your preferred AI coding agent.