nairon-factory
v0.1.2
Published
AI-augmented SDLC workflow - from stakeholder call to deployed feature
Downloads
289
Maintainers
Readme
Nairon Factory
AI-augmented SDLC workflow - from stakeholder call to deployed feature.
┌─────────────────────┐ ┌─────────────────────┐
│ Linear │◄───►│ Beads (bd sync) │
│ (kanban board) │ │ (source of truth) │
└─────────────────────┘ └──────────┬──────────┘
│
▼
┌─────────────────────┐
│ VPS Agents │
│ (OpenClaw) │
└──────────┬──────────┘
│
▼
┌─────────────────────┐
│ Telegram/Discord │
│ (async Q&A) │
└─────────────────────┘What is this?
Nairon Factory is the methodology we use to build startups for clients - open-sourced so others can adopt it.
Philosophy: "Speed of impact, not speed of output."
Key Innovation: Async human-in-the-loop. Agents work autonomously on a VPS, ping humans via Telegram/Discord when they need decisions. No babysitting.
SDLC Phases
| Phase | Time | What Happens | |-------|------|--------------| | Requirements | 25% | Stakeholder call → PRD | | Planning | 35% | PRD → Technical plans + tickets | | Implementation | 15% | Plans → Working code with atomic commits | | Review | 25% | Code → Tested, reviewed, deployed |
Components
1. nf-tools CLI
State management for the workflow. Runs on VPS.
nf-tools state init "Project Name"
nf-tools phase create "Authentication"
nf-tools plan create 1 --name "OAuth Setup"
nf-tools progress2. Agents
Specialized AI agents that do the actual work:
| Agent | Model | Purpose |
|-------|-------|---------|
| nf-interviewer | opus-4.6 | Clarifies requirements, generates PRDs |
| nf-planner | gpt-5.3-codex | Creates wave-based technical plans |
| nf-executor | gpt-5.3-codex (gemini-3.1-pro for frontend) | Implements plans with atomic commits |
| nf-reviewer | gpt-5.3-codex | Tests, reviews, verifies completion |
| nf-scout | gpt-5.3-codex-spark | Parallel codebase exploration (read-only) |
Model Strategy:
- opus-4.6: Interviewing (stakeholder empathy, ambiguity detection)
- gemini-3.1-pro: Frontend tasks (UI/UX intuition, component design)
- gpt-5.3-codex: Planning, implementation, review (superior code generation + technical reasoning)
- gpt-5.3-codex-spark: Scouting (fast parallel exploration)
3. Linear Integration
Beads syncs with Linear so you see tasks on a real kanban board.
4. Telegram Bot
Questions from agents appear in Telegram. Team discusses, answers with /answer.
Installation
npx nairon-factory@latestThis will prompt you to:
- Choose your runtime (Claude Code or OpenCode)
- Choose install location (global or local)
Manual Install
# Claude Code only
npx nairon-factory@latest --claude --global
# OpenCode only
npx nairon-factory@latest --opencode --global
# Both runtimes
npx nairon-factory@latest --all --global
# Local to current project
npx nairon-factory@latest --claude --localUninstall
# Uninstall from all runtimes (global)
npx nairon-factory@latest --uninstall --all --global
# Uninstall from specific runtime
npx nairon-factory@latest --uninstall --claude --global
npx nairon-factory@latest --uninstall --opencode --global
# Uninstall from local project
npx nairon-factory@latest --uninstall --claude --localConnect Services (Optional)
# Link to Linear
nf config set linear_api_key <key>
nf config set linear_team_id <team>
# Link to Telegram
nf config set telegram_bot_token <token>
nf config set telegram_chat_id <chat>Workflow
1. Start Project
Upload stakeholder call transcript → nf-interviewer clarifies → PRD generated
2. Plan Phases
nf-planner researches codebase → creates wave-based plans → tickets in Linear
3. Execute
nf-executor implements wave by wave → atomic commits → handles deviations:
- Tier 1-3: Auto-fix
- Tier 4: Asks human via Telegram
4. Review
nf-reviewer runs tests → AI code review → creates verification report
5. Deploy
If gates pass → merge PR → deploy
Directory Structure
.nf/
├── STATE.md # Current project state
├── PRD.md # Product requirements
├── config.json # Linear/Telegram keys
├── phases/
│ ├── 1-authentication/
│ │ ├── CONTEXT.md # Phase goal, scope, criteria
│ │ ├── PLAN-1.md # Wave-based task breakdown
│ │ ├── SUMMARY-1.md # Execution results
│ │ └── VERIFICATION.md
│ └── 2-dashboard/
└── questions/
├── q-123.json # Pending questions
└── answered/Deviation Handling
When agents encounter unexpected situations:
| Tier | Severity | Action | |------|----------|--------| | 1 | Trivial | Auto-fix silently (typos, formatting) | | 2 | Minor | Auto-fix + note in summary | | 3 | Significant | Auto-fix + add blocker | | 4 | Major | STOP, ask human via Telegram |
Git Worktrees
For parallel phase execution:
# Create worktree for phase 2
git worktree add ../project-phase-2 -b phase/2-dashboard
# Agents work in separate worktrees
# Merge back when phase completeDevelopment
# Install dependencies
bun install
# Run CLI locally
bun run packages/cli/src/index.ts
# Build
bun run build
# Test
bun testLicense
MIT
