Feature Development Flow

How Hypr's planning and execute agents collaborate to build features end-to-end.

Core Principle

Different types of work require different agents with specialized approaches. Never mix planning with execution. The Plan Agent designs feature documents; Execute Agents implement chunks from those documents.

Agent Types and When to Use

1. Plan Agent — feature-level work

Use when: a user requests a new feature, enhancement, or complex change.

  • Follows the planning process from planning-agent.md (created via /hypr:assess-planning).
  • Creates comprehensive feature documents with chunk breakdown.
  • Never implements code — only plans and documents.

2. Frontend Execute Agent — single chunk

Use when: working on a specific frontend chunk from a feature plan.

  • Follows the 8-step per-file implementation loop from frontend-agent.md.
  • References the feature document for context.
  • Only works on the assigned chunk, not the entire feature.

3. Backend Execute Agent — single chunk

Use when: working on a specific backend chunk from a feature plan.

  • Follows the 7-step per-file implementation loop from backend-agent.md.
  • References the feature document for context.
  • Only works on the assigned chunk, not the entire feature.

Decision Tree

User Request → Determine Agent Type:

Is this a new feature or complex change?
├── YES → Use Plan Agent
│   ├── Create feature document
│   ├── Break into chunks
│   └── Hand off chunks to Execute Agents
└── NO → Is it frontend or backend work?
    ├── Frontend → Use Frontend Execute Agent
    └── Backend → Use Backend Execute Agent

Feature Document Template

Plan Agents create feature documents at features/feature-XXX.md following this shape: 

# Feature XXX: [Feature Name]

## Requirements
- [List original user requirements]

## Architecture Design
- [How this feature fits into existing app patterns]
- [What components/services will be created/modified]
- [Integration points with existing systems]

## Implementation Chunks

### Chunk 1: [Name]
**Type:** Frontend/Backend/Both
**Dependencies:** None / Chunk X must be completed first
**Files to create/modify:**
- path/to/file1.tsx
**Tests required:** Yes/No - [when]
**Acceptance criteria:**
- [ ] Specific outcome 1

[Continue for all chunks...]

## Testing Strategy
- Unit tests: [when and what to test]
- Integration tests: [when and what to test]

## Rollback Plan
- [How to undo this feature if needed]

Critical Rules

  1. Plan Agent never executes code — only creates feature documents.
  2. Execute Agents never plan features — only implement assigned chunks.
  3. Execute Agents must reference the feature document for context.
  4. Execute Agents must mark chunks complete in the feature document when done.
  5. Plan Agent creates realistic chunks — small enough for a single execute session.

Related