swe-atlas

Turn Claude Code into a senior software engineer with persistent identity, real engineering principles, and production-grade tooling.

Quick Start

npx swe-atlas@latest new-project "project-folder-name"

If no folder name is provided, the current directory is used.

Installation

# Or install globally
npm install -g swe-atlas
swe-atlas new-project

Update

# Always use the latest version
npx swe-atlas@latest new-project

# Or update global install
npm update -g swe-atlas

What It Does

The CLI interactively asks:

1. Your name — ATLAS refers to you as "Boss"
2. Project type — Single repo or Multi repo (workspace with repos/ folder)
3. Context templates — Which conventions to activate (all are copied, you pick what's active)
4. MCP servers — Playwright (browser automation), PostgreSQL (database access)

Then scaffolds the full ATLAS setup:

your-project/
├── CLAUDE.md                # ATLAS identity & entry point
├── IMPORTANT_NOTES.md        # Critical lessons & warnings
├── .claude/
│   ├── skills/              # 11 specialized skills
│   ├── agents/              # 6 task-specific agents
│   ├── commands/            # 5 slash commands
│   ├── hooks/               # Task completion hooks
│   ├── settings.json        # Permissions config
│   └── settings.local.json  # Local overrides
├── self/ (or atlas/self/)    # Identity & engineering principles
├── context-templates/        # All convention templates
├── development-context/      # Active conventions for this project
├── external_information/     # Git submodules (plugins, skills)
├── automation_tests/         # QA test cases and results
├── docs/                     # Project documentation
├── misc/prompts/             # Prompt templates
└── .mcp.json                 # MCP server configuration

Single vs Multi Repo

| | Single Repo | Multi Repo | |---|---|---| | Use case | One project | Workspace with multiple projects | | Identity files | atlas/self/ | self/ | | Projects | Your existing code stays as-is | repos/backend/, repos/frontend/ | | Command | npx swe-atlas new-project | npx swe-atlas new-project my-workspace |

What You Get

11 Skills

| Skill | What It Does | |-------|--------------| | /abstraction-power | Pattern recognition — identify and extract reusable abstractions | | /frontend-design | Production-grade web UI with anti-AI-slop methodology | | /human-writing | Content indistinguishable from skilled human writers | | /pdf | Read, merge, split, watermark, OCR, fill forms | | /pptx | Create and edit slide decks | | /docx | Create and edit Word documents | | /xlsx | Spreadsheets with formulas, charts, data cleaning | | /canvas-design | Visual art and posters as PNG/PDF | | /algorithmic-art | Generative art using p5.js | | /theme-factory | 10 professional themes for any artifact | | /mcp-builder | Guide for creating MCP servers |

6 Agents

| Agent | What It Does | |-------|--------------| | code-architect | Feature architecture with implementation blueprints | | code-explorer | Trace execution paths, map architecture layers | | code-review | Multi-agent PR review with confidence scoring | | code-simplifier | Refine code for clarity and maintainability | | qa-manual-tester | Browser-based QA testing via Playwright | | commit | Git commits following ATLAS convention |

5 Commands

| Command | What It Does | |---------|--------------| | /swe-atlas:new-project-context | Initialize project context, configure conventions, set up workspace | | /swe-atlas:create-phase "phase-01-mvp" | Create a development phase with specs, test cases, and structure | | /feature-dev | Guided feature development with codebase exploration | | /run-be-fe | Run backend and frontend in background | | /qa-manual-test-run | Execute QA test cases | | /commit | Commit using sub-agent |

After Setup

Open your project in Claude Code and:

/swe-atlas:new-project-context  # Configure ATLAS for your project
/feature-dev                    # Guided feature development
"Add user authentication"       # Or just describe what you need

The Neuron Activation Problem

AI coding assistants have deep engineering capabilities locked behind generic defaults. The difference between "write a function" and a properly activated AI Software Engineer is like night and day.

Context rot — models advertise 200k-1M token windows, but performance degrades long before the limit. The 10,000th token is not as trustworthy as the 10th.

Vibecoding hangovers — without engineering principles, AI-generated code becomes unmaintainable fast.

Repetitive setup — every session means re-explaining project structure, conventions, and decisions from scratch.

ATLAS solves all three. Persistent identity. Structured context. Engineering principles that load automatically. Just open the project and build.

Links

• GitHub Repository
• Claude Code

License

MIT