Implement

Execute the agreed implementation plan with task management, sub-agent orchestration, and verification.

When to Use

• User calls /implement after finalizing approach
• Transition from planning/discussion to coding phase
• User says "implement this", "build this", "let's do it"

Workflow

Phase 1: Extract Plan

Identify the implementation scope from:

1. Current conversation context
2. Linked plan file (if user provided)
3. Previous discussion in same session

Output a brief summary of what will be implemented. Confirm with user if scope is ambiguous.

Phase 2: Task Breakdown

Create tasks using TodoWrite tool:

1. Analyze scope and break into discrete tasks
2. Order tasks by dependency (what must finish before what)
3. Each task should be:
   • Specific and actionable
   • Completable in one focused effort
   • Testable/verifiable

Example task breakdown:

1. Create data model for X
2. Implement API endpoint for Y
3. Add validation logic
4. Write component Z
5. Connect frontend to API
6. Test full flow

Phase 3: Execute

For each task:

1. Mark task in_progress
2. Implement the task
3. Verify it works (run, build, or quick test)
4. Mark task completed
5. Move to next task

When to use sub-agents:

• Complex file exploration: use Explore agent
• Multi-step bash operations: use Bash agent
• Research across codebase: use general-purpose agent

When to load domain skills:

If a task requires specialized knowledge, read the corresponding skill from D:\.claude\skills\ before executing:

Domain	Skill	When to load
Business analytics	analyst	Analyzing sales/revenue data, finding weak areas
Data visualization	csv	Creating charts, dashboards from CSV
KPI dashboards	dashboard	Designing metrics, monitoring layouts
Design principles	design	Architecture decisions, file structure, UI patterns
Word documents	docx	Creating/editing .docx files
Frontend/UI	frontend-design	Building web components, pages, interfaces
Document conversion	markdown	Converting files to markdown
Obsidian notes	obsidian	Working with Obsidian markdown syntax
PDF processing	pdf	Extract, merge, split, fill forms
PowerPoint	pptx	Creating/editing presentations
Data storytelling	story	Presenting analytics, executive reports
QA/Testing	test	Test strategies, test cases, bug tracking
Excel/Spreadsheets	xlsx	Creating/editing spreadsheets with formulas

How to use:

1. Identify if task matches a skill domain
2. Read D:\.claude\skills\{skill-name}\SKILL.md
3. Follow skill's workflow/instructions for that specific task
4. Continue with implementation

When task fails:

• Diagnose the issue
• Fix and retry
• If blocked, note blocker and ask user

Phase 4: Quality Check

After all tasks complete, run appropriate checks based on project:

1. Build - Ensure project compiles/builds
2. Lint/Format - If project has linter configured
3. Type check - If TypeScript/typed language
4. Test - Run existing tests if available

Fix any failures before proceeding.

Phase 5: Verify Features

Test that implemented features work as originally discussed:

1. Run the application/feature
2. Verify expected behavior matches requirements
3. Check edge cases mentioned in discussion
4. If something doesn't work, fix it

Phase 6: Complete

When all features work:

1. Clear completed todos
2. Summarize what was implemented
3. Note any follow-up items discovered

State Management

When conversation gets long or context is filling up:

Save state to file (AI chooses location, typically project root or scratchpad):

# Implementation State

## Scope
[What we're implementing]

## Tasks
- [x] Task 1
- [x] Task 2
- [ ] Task 3 (in progress)
- [ ] Task 4

## Current Status
[Where we are, what's next]

## Notes
[Any blockers, decisions made, etc.]

To resume: Read state file and continue from current task.

Principles

• Linh dong: This is a flexible framework, not rigid checklist
• AI tu quyet dinh: AI chooses tools, agents, approach based on context
• Fail thi fix: If something breaks, fix it, don't skip
• Done = works: Implementation complete when features work as discussed, not just code written

Error Handling

Situation	Action
Unclear requirement	Ask user to clarify
Build fails	Fix build errors
Test fails	Fix failing tests
Blocked by external dependency	Note blocker, ask user
Context running low	Save state, inform user

Examples

Simple feature:

User: [discusses adding login button]
User: /implement

AI:
1. Extracts: "Add login button to header"
2. Creates 3 tasks: component, styling, click handler
3. Implements each task
4. Verifies button appears and works
5. Done

Complex feature:

User: [discusses new API with multiple endpoints]
User: /implement

AI:
1. Extracts full API spec from discussion
2. Creates 8 tasks across routes, models, validation
3. Uses sub-agents for exploration when needed
4. Runs tests after each endpoint
5. Final integration test
6. Done