Plagger for the Agent Era — 20 years later.
Plagger (2006) was a personal RSS aggregator that lived on your server.
adev (2026) is an autonomous development agent that lives in your YAML.
adev is a YAML-only configuration file that spawns an entire development team of AI agents.
Everything starts from chat.
Chat → YAML (auto-generated) → Specs → Code → Test → Deploy
↑ stack, style, platform
determined from conversation
You talk. Agents listen. YAML is created. Software appears.
- 💬 Chat — Your conversation becomes requirements + config
- 📝 YAML generation — Stack, style, deploy target auto-decided from chat
- 📋 Spec writing — Auto-generated from chat context
- 💻 Implementation — Code generation in your chosen stack
- ✅ Testing — TDD cycle automation
- 🚀 Deployment — One-command deploys to your platform
- 🔄 Iteration — Continuous improvement from feedback
No CLI. No complex setup. Just Chat → YAML → Software.
"The best interface is no interface."
20 years ago, Plagger let you configure RSS feeds in YAML.
Today, adev lets you configure software development in YAML.
| Era | Tool | Configuration | Output |
|---|---|---|---|
| 2006 | Plagger | config.yaml |
RSS Feed |
| 2026 | adev | adev.yaml |
Working Software |
Minimal:
stack:
language:
primary: tsFull (9 Layers):
metadata:
version: "1.0.0"
generated: "2026-03-19T00:00:00+09:00"
context:
chat:
url: https://gemini.google.com/app/your-session
stack:
language:
primary: ts
runtime:
name: node
version: "20"
style:
methodology: TDD
deploy:
platform: firebase
team:
workflow: chat-to-impl
structure:
root:
- src/
- tests/
constraints:
must:
- "すべてのコードは型付きであること"
quality:
metrics:
maintainability_index: 80# Anywhere, anytime
adev run# 対話型仕様選択
.\skills\adev-config-skill.ps1 -i
# チャットから自動抽出
.\skills\adev-config-skill.ps1 -chat "Firebase で TypeScript のアプリを作りたい"[adev] Spawning team...
[spec_writer] Reading chat context...
[spec_writer] Generated 3 user stories
[code_reviewer] Reviewing architecture...
[implementer] Creating src/functions.ts...
[tester] Running jest... 12/12 passed
[deploy] Deploying to firebase... ✓
See skills/LAYERS.md for the full 9-layer specification.
Quick reference:
| Layer | Purpose | Example |
|---|---|---|
| 0. METADATA | 設定のメタデータ | version, generator, confidence |
| 1. CONTEXT | 開発の文脈 | chat URL, project vision |
| 2. STACK | 技術スタック | language, runtime, framework |
| 3. STYLE | 開発スタイル | TDD, git workflow |
| 4. DEPLOY | デプロイ | firebase, vercel, aws |
| 5. TEAM | エージェント構成 | roles, responsibilities |
| 6. STRUCTURE | プロジェクト構造 | directories |
| 7. CONSTRAINTS | 制約条件 | must, should, must_not |
| 8. QUALITY | 品質基準 | metrics, gates |
Settings evolve as your project grows:
txt → md → yaml → json
│ │ │ │
│ │ │ └─ Validated, typed
│ │ └──────── Layered, semantic
│ └─────────────── Structured notes
└───────────────────── Chat log
Stage 0: TXT - "Firebase で TypeScript のアプリ作りたい" Stage 1: MD - Structured markdown notes Stage 2: YAML - 9-layer configuration Stage 3: JSON - Validated schema
See skills/EVOLUTION.md for details.
┌─────────────────┐
│ You Chat │
│ (natural lang) │
└────────┬────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ adev Agent: config_writer │
│ │
│ "I want a Firebase app with TypeScript, TDD style" │
│ ↓ │
│ Generates: adev.yaml │
│ - stack: ts │
│ - style: TDD │
│ - deploy: firebase │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ adev.yaml (config) │
│ context: │
│ chat: https://gemini.google.com/app/your-session │
│ dev: │
│ style: TDD │
│ stack: ts │
│ deploy: │
│ platform: firebase │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────────────────────────────────────────────┐
│ adev Agent Orchestrator │
│ │
│ [spec_writer] Reads chat → Generates specs │
│ [code_reviewer] Reviews architecture │
│ [implementer] Writes code │
│ [tester] Runs TDD cycle │
│ [deployer] Deploys to platform │
└─────────────────────────────────────────────────────────┘
│
▼
┌─────────────────┐
│ Working Software│
│ + Tests + Docs │
└─────────────────┘
Flow:
- You chat naturally about what you want
config_writeragent → generatesadev.yaml(stack, style, platform)spec_writeragent reads chat → generates formal specs- Team implements, tests, deploys automatically
- You review. Iterate via more chat.
YAML is auto-generated. You just chat.
If you can write YAML, you can use adev.
Local, CI, cloud — adev runs wherever your code lives.
Built for the age of AI agents, not CLI tools.
No GUI. No complex config. Just one file.
Natural language, Markdown, YAML, JSON — any format works:
# Natural Language
"Firebase で TypeScript のアプリを作りたい、TDD で"
# Markdown
## Stack: TypeScript
## Style: TDD
## Deploy: Firebase
# Keywords
ts TDD firebase jest github
All inputs are converted to adev.yaml automatically.
adev includes built-in skills for configuration:
# Interactive mode with recommendations
.\skills\adev-config-skill.ps1 -i
# Auto-extract with optimal stack suggestion
.\skills\adev-config-skill.ps1 -chat "Web アプリ作りたい"Features:
- 🎯 Optimal Stack Recommendation - Suggests best stack from project type
- 📊 Pro/Con Comparison - Shows trade-offs for each option
- 📝 ADR Generation - Records decisions with rationale
# Generate ADR during config
.\skills\adev-config-skill.ps1 -i
# → Answer "y" when asked about ADROutputs: docs/adr/YYYYMMDD-HHMMSS-stack-decision.md
ADR includes:
- Context
- Options considered (with comparison tables)
- Decision & rationale
- Consequences (positive/negative/risks)
See skills/ADR.md for details.
| Feature | Plagger (2006) | adev (2026) |
|---|---|---|
| Config | Manual YAML | Auto-generated from chat |
| Input | RSS feed URLs | Chat conversation |
| Specs | Manual config | Auto-generated from chat |
| Output | Aggregated feed | Working software |
| Plugins | Perl modules | AI agents |
| Automation | Cron | Event-driven |
| Portability | Server-bound | Anywhere |
- Config skill - 対話型仕様選択
- Plugin system for custom agents
- Multi-chat context support
- State persistence across sessions
- Agent memory & learning
- Team collaboration features
MIT — Just like Plagger was.
This project is inspired by Plagger, the legendary RSS aggregator created by miyagawa in 2006. Plagger proved that flexible, plugin-based architecture could be configured entirely in YAML.
20 years later, we're applying that same philosophy to AI-driven development.
miyagawa, thank you for showing us the power of YAML.
adev — Chat → YAML → Ship software.