Feature Development Workflow SOP

Version: 1.0.0 Last Updated: 2026-01-11 Status: Active

Overview

Purpose

This workflow enables complex, multi-session feature development using AI agents with persistent documentation. It ensures continuity across sessions, tracks progress systematically, and allows any agent (or human) to pick up work exactly where the previous session left off.

When to Use

ALWAYS: Multi-day features, architecture changes, projects requiring research + design + implementation phases, work spanning multiple sessions SKIP: Tasks completable in a single session, simple bug fixes, minor updates

Process Workflow

Flow Diagram

[New Feature Request]
        |
[Phase 1: Setup] --> Create feature folder + INDEX.md
        |
[Phase 2: Research] --> Spawn Explore agent --> research/*.md
        |
[Phase 3: Design] --> design/DESIGN_##_*.md
        |
[Phase 4: Planning] --> development/EXECUTION_##_*.md
        |
[Phase 5: Execute] --> Spawn sub-agents --> Implement
        |
[Phase 6: Validate] --> TypeScript + Review + Tests
        |
[Phase 7: Handoff] --> Update SESSION_HANDOFF.md

Phase Summary

Quick Start

Starting a New Feature

1. Create Feature Folder

   /docs/ignored/{feature-name}/
   ├── INDEX.md                    # Central navigation + status
   ├── SESSION_HANDOFF.md          # Session continuity
   ├── research/                   # Research and analysis
   ├── design/                     # Design documents
   └── development/                # Execution plans
   

2. Create INDEX.md using INDEX_TEMPLATE.md

3. Create SESSION_HANDOFF.md using SESSION_HANDOFF_TEMPLATE.md

4. Begin Research Phase - Invoke

   /design-research

   skill

Resuming a Feature

1. Read

   INDEX.md

   for overall project status
2. Read

   SESSION_HANDOFF.md

   for last session context
3. Check dependency graph for next available task
4. Read relevant DESIGN + EXECUTION docs
5. Continue execution

Key Principles

1. Index as Single Source of Truth - INDEX.md is the authoritative reference for project status
2. Session Handoff Continuity - Every session ends with updated SESSION_HANDOFF.md
3. Atomic Execution Documents - Each execution is self-contained with clear inputs/outputs
4. Agent Specialization - Use Explore for research, general-purpose for implementation
5. Documentation-Driven Progress - Work isn't "done" until docs reflect completion

Detailed Workflow

See WORKFLOW.md for complete phase-by-phase instructions including:

• Step-by-step procedures for each phase
• Agent spawning patterns
• Validation checklists
• Troubleshooting guidance

Templates

Agent Spawn Patterns

Research Phase (Explore Agent)

Use Task tool with subagent_type: "Explore"
- Codebase questions
- Finding files/patterns
- Understanding architecture

Implementation Phase (General-Purpose Agent)

Use Task tool with subagent_type: "general-purpose"
- Creating files
- Modifying code
- Running validations

Quick Reference

Folder Structure

/docs/ignored/{feature}/

/docs/ignored/{feature}/INDEX.md

/docs/ignored/{feature}/SESSION_HANDOFF.md

/docs/ignored/{feature}/research/

/docs/ignored/{feature}/design/

/docs/ignored/{feature}/development/

Validation Commands

# TypeScript check
npx tsc --noEmit

# Format and lint
task format-all

# Run tests
task test

Status Indicators

Troubleshooting

Related Skills

/design-research

/review-code

/test

/git-pr

Note: Skill paths (

/skill-name

) work after deployment. In the template repo, skills are in domain folders.

End of SOP