Ever started a feature, gotten halfway through, then realized you forgot a critical dependency? Or had Claude Code build something, only to realize the tasks were done in the wrong order? I got tired of managing complex features in my head, so I built a plugin that brings structured planning directly into Claude Code.
The Problem
When building features with AI assistance:
- You explain the feature, Claude starts coding, but forgets earlier context
- Multi-step tasks get jumbled—tests written before the code they test
- You lose track of what's done vs. what's pending
- Dependencies between tasks aren't respected (middleware before routes, etc.)
- Context switching kills momentum—you forget where you left off
The Solution: Plan-Mode
What if you could define your entire feature as a dependency graph, and have Claude execute it step-by-step?
You: /plan JWT Authentication
Claude: Creates .claude/plans/jwt-authentication.md with 6 tasks
You: /plan-execute jwt-authentication
Claude: Executes next available task respecting dependencies
That's it. Structured planning meets AI execution.
How It Works
-
Create a plan -
/plan <feature>generates a YAML-based plan with tasks and dependencies -
Track progress -
/planshows all plans with completion metrics -
Execute systematically -
/plan-execute <name>runs the next available task -
Manual control -
/plan-update <name> [id] [status]lets you adjust as needed
Plans live in your repo at .claude/plans/, versioned with your code.
Get Started in 30 Seconds
/plugin marketplace add quochuydev/cc-plan-mode
/plugin install plan-mode@cc-plan-mode
This adds:
-
/plancommand - View and create structured plans -
/plan-executecommand - Run tasks sequentially with dependency awareness -
/plan-updatecommand - Manually adjust task status
Commands
| Command | Purpose | Example |
|---|---|---|
/plan |
List all plans with progress | Shows completion % per plan |
/plan <desc> |
Create new plan | /plan user authentication |
/plan-execute <name> |
Execute next task | /plan-execute auth-system |
/plan-update <name> [id] [status] |
Update task status | /plan-update auth-system 3 completed |
Plan Format
Plans use YAML frontmatter for clean structure:
---
name: jwt-authentication
overview: Add JWT-based authentication to the API
todos:
- id: 1
content: Install jsonwebtoken and bcrypt dependencies
status: pending
- id: 2
content: Create User model with password hashing
status: pending
dependencies: [1]
- id: 3
content: Implement auth middleware
status: pending
dependencies: [2]
---
Why This Works
- Dependency awareness - Tasks execute in the right order, always
- Persistent state - Plans survive across sessions, stored in your repo
- Clear visibility - See exactly what's done, what's next, what's blocked
- AI-native execution - Claude understands context from the plan file
Try It
If you're building multi-step features with Claude Code and want more structure:
/plugin marketplace add quochuydev/cc-plan-mode
/plugin install plan-mode@cc-plan-mode
Then run /plan my-feature and watch it break down your feature into manageable, ordered tasks.
GitHub: quochuydev/cc-plan-mode
How do you currently manage complex, multi-step features when working with AI assistants? I'd love to hear what workflows have worked for your team.
Top comments (0)