Document-driven development skills with Multi-Agent Debate for AI coding tools
Archflow is a skill-based plugin for AI coding tools (Cursor, Claude Code, OpenCode) that enables systematic, document-driven development workflows. It leverages Multi-Agent Debate to produce high-quality architectural designs and maintains full traceability from requirements to implementation.
┌─────────────────────────────────────────────────────────────────────────────┐
│ Backend Development │
│ │
│ /spec → /arch → /check → /pre-build → /build → /test │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────────┐
│ Frontend Development │
│ │
│ /spec → /arch(BE) → /check → /ui → /arch(FE) → /check → /pre-build │
│ → /build → /test │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────────┐
│ Bug Fix │
│ │
│ /debug → /trace → /sync │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────────┐
│ Enhancement │
│ │
│ /reinforce → /arch → /check → /pre-build → /build → /test │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
┌─────────────────────────────────────────────────────────────────────────────┐
│ Legacy Documentation │
│ │
│ /reverse → /reinforce (optional) → /check │
│ │
└─────────────────────────────────────────────────────────────────────────────┘
SSOT Principle: All changes start from spec.md. arch.md is a derived document.
- Document-Driven Development: Maintain clear requirements, designs, and changelogs throughout the development lifecycle
- Multi-Agent Debate: Two specialized agents (Domain Architect and Best Practice Advisor) collaborate to produce optimal designs
- Systematic Workflows: 15 pre-built skills for common development tasks
- Full Traceability: Keep documentation synchronized with code changes
Open docs/launchpad.md in your AI tool and tell the AI:
"Follow this launchpad to install archflow"
The AI will clone the repository and copy the necessary files automatically.
git clone https://github.com/samdae/archflow.git .archflow-tmp
cp -r .archflow-tmp/skills/ .cursor/skills/
cp -r .archflow-tmp/rules/ .cursor/rules/
cp -r .archflow-tmp/agents/ .cursor/agents/
rm -rf .archflow-tmpInside Claude Code, run these commands:
/plugin marketplace add samdae/archflow
/plugin install archflow@samdae
/plugin enable archflow@samdae
git clone https://github.com/samdae/archflow.git .archflow-tmp
cp -r .archflow-tmp/skills/ .opencode/skills/
cp -r .archflow-tmp/commands/ .opencode/commands/
cp -r .archflow-tmp/agents/ .opencode/agents/
cp -r .archflow-tmp/rules/ .opencode/rules/
rm -rf .archflow-tmpThen add to opencode.json in the project root (create if absent):
{
"$schema": "https://opencode.ai/config.json",
"instructions": [".opencode/rules/archflow-rules.md"]
}If
opencode.jsonalready exists, append to theinstructionsarray instead of overwriting.
| Tool | Installation |
|---|---|
| Cursor | Clone + copy to .cursor/skills/, .cursor/rules/, .cursor/agents/ |
| Claude Code | /plugin marketplace add samdae/archflow → install → enable |
| OpenCode | Clone + copy to .opencode/skills/, .opencode/commands/, .opencode/agents/ + opencode.json |
After installation, use slash commands:
| Command | Description |
|---|---|
/archflow |
Show all available skills |
/spec |
Transform materials into spec.md |
/arch |
Multi-agent debate → arch-be.md or arch-fe.md |
/ui |
Generate UI specification from API endpoints |
/check |
Verify design completeness |
/pre-build |
Verify environment readiness (deps, secrets, infra) |
/build |
Automated implementation from design |
/test |
Generate and/or run tests (scoped) |
/debug |
Systematic bug fixing with direct execution |
/trace |
Record changes to trace.md |
/sync |
Sync changes to arch.md |
/reinforce |
Add requirements to spec.md or fill gaps |
/reverse |
Reverse-engineer docs from existing code |
/overview |
Generate 1-page project overview |
/runbook |
Generate deployment documentation |
/spec → /arch → /check → /pre-build → /build → /test
/spec → /arch(BE) → /check → /ui → /arch(FE) → /check → /pre-build → /build → /test
/debug → /trace → /sync
/reinforce → /arch → /check → /pre-build → /build → /test
/reverse → /reinforce (optional) → /check
Note: All changes start from spec.md via
/specor/reinforce.
| Skill | Description | BE/FE | Recommended Model |
|---|---|---|---|
spec |
Refine raw requirements into structured docs | - | Opus |
arch |
Multi-Agent Debate for feature design | BE/FE | Opus |
ui |
Generate UI specification from API endpoints | - | Opus |
check |
Verify design completeness before implementation | BE/FE | Sonnet |
pre-build |
Verify environment readiness (deps, secrets, infra) | BE/FE | Sonnet |
build |
Automated implementation from design docs | BE/FE | Sonnet |
test |
Generate and/or run tests with scoped targeting | BE/FE | Sonnet |
debug |
Systematic debugging with direct code execution | - | Opus |
trace |
Generate structured changelogs | - | Sonnet |
sync |
Sync documentation after code changes | - | Sonnet |
reinforce |
Add requirements or fill documentation gaps | - | Sonnet |
reverse |
Generate documentation from code | BE/FE | Opus |
overview |
Generate 1-page project overview | - | Sonnet |
runbook |
Generate deployment documentation | - | Sonnet |
Skills marked with "BE/FE" support separate profiles for Backend and Frontend development.
| Agent | Description |
|---|---|
domain-architect |
Project-aware design agent with domain knowledge |
best-practice-advisor |
Context-free best practice advisor |
Skills that generate documentation follow this folder structure:
projectRoot/
└── docs/
└── {serviceName}/
├── spec.md # from /spec
├── arch-be.md # from /arch (Backend)
├── ui.md # from /ui (UI specification)
├── arch-fe.md # from /arch (Frontend)
├── trace.md # from /trace
├── overview.md # from /overview
└── runbook.md # from /runbook
Note: Frontend development requires
/uibefore/arch (FE). The ui.md defines screens, components, and interactions based on backend API endpoints.
All skills are written in English but automatically respond in the user's language.
Supported trigger keywords include:
- English:
arch,build,debug, etc. - Korean:
설계,구현,디버그, etc.
Apache License 2.0
samdae
https://github.com/samdae/archflow
Every skill file must begin with YAML front matter:
---
id: spec
name: Spec
description: |
Transform unstructured materials into refined requirements document.
Collects service name and input materials, creates Q&A loop.
Triggers: spec, specification, 요구사항 정의
user-invocable: true
version: 2.0.0
triggers:
- "spec"
- "specification"
- "requirements"
requires: []
recommended_model: opus
allowed-tools:
- Read
- Write
- Glob
- Grep
- LS
- AskQuestion
---| Field | Description |
|---|---|
id |
Unique skill identifier in kebab-case |
name |
Human-readable skill name |
description |
Brief description with triggers |
user-invocable |
Must be true for slash command registration |
version |
Semantic versioning |
allowed-tools |
List of tools this skill can use |
All skills include a global rules reference after the front matter:
> **Global Rules**: Adheres to rules/archflow-rules.mdEvery skill should handle unavailable tools:
| Tool | Fallback |
|---|---|
| Read/Grep | Request file path → ask for copy-paste |
| AskQuestion | "Please select: 1) OptionA 2) OptionB" format |
| Task | Step-by-step checklist or Self-debate pattern |
Skills producing implementation-ready documents must validate:
| Item | Validation Criteria |
|---|---|
| DB Schema | All tables have full column definitions |
| Code Mapping | All features mapped to file/class/method |
| API Spec | All endpoints have Request/Response defined |
| Error Policy | Main error scenarios defined |