SWE-ATLAS
Help Claude Code decide what to build through visual exploration and engineering judgment before implementation begins.
- Stars
- ★ 402
- Last updated
- 3mo ago
- Primary language
- Python
- FA score
- 42/100 · Major gaps
At a glance
- How it runs
- Works with
- Platform-specificClaude Code
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- A solo developer using Claude Code wants to validate an idea with a wireframe or clickable prototype before generating implementation code.
- Not a fit if
- Teams that need to use the toolkit outside Claude Code
- Projects that require Markdown-only planning artifacts
- Source review
- 42/100 · Major gaps 1 safety controls not found
What does this agent do, and when should you use it?
SWE-ATLAS is an open-source template for Claude Code that scaffolds skills, subagents, slash commands, and engineering conventions into a project. It offers Vanilla, Collaborative, and Autonomous CLAUDE.md modes, with Collaborative as the default. The workflow explores an idea with an HTML wireframe, clickable prototype, or visual plan, clarifies important unknowns through Q&A, and captures the phase in a browser-ready HTML document. Claude Code’s native plan mode, /goal, and workflows then handle implementation, while ATLAS supplies QA skills and agents. It is built for the Claude Code ecosystem; Playwright browser automation and PostgreSQL MCP are optional setup choices.
Run npx swe-atlas@latest new-project to interactively configure and scaffold the current directory or a new one, choosing a CLAUDE.md mode, skills, DESIGN.md template, browser automation setup, and PostgreSQL option. In Claude Code, /atlas:get-to-know initializes project context; /brainstorm:wireframe creates a standalone HTML wireframe, /brainstorm:prototype creates a multi-screen React prototype, and /plan:create-phase clarifies key issues through Q&A and outputs a self-contained phase.html. /plan:visual produces a visual HTML plan for a refactor, migration, or architecture decision. The user then hands the document to Claude Code’s native plan mode, /goal, and workflow features for implementation. QA commands author manual test cases and can run them through a configured Playwright integration.
- A solo developer using Claude Code wants to validate an idea with a wireframe or clickable prototype before generating implementation code.
- A small product team needs targeted Q&A to clarify key requirements and a browser-readable HTML phase plan to share.
- A tech lead is planning a refactor, migration, or architecture change and wants Mermaid diagrams and decision matrices in the plan.
- A team using Claude Code’s native plan and workflow features wants project conventions, engineering skills, and decision records alongside them.
- A project team needs browser-based manual QA with a configured Playwright integration.
How do you install or deploy this agent?
You need Claude Code and a working Node.js/npm environment. The CLI quick start is:
npx swe-atlas@latest new-projectTo scaffold a new directory instead:
npx swe-atlas@latest new-project my-workspaceFor manual setup, clone the repository with its submodules:
git clone --recurse-submodules https://github.com/syahiidkamil/Software-Engineer-AI-Agent-Atlas
cd Software-Engineer-AI-Agent-AtlasThe scaffolder asks about CLAUDE.md mode, skills, design template, and optional integrations. The README does not specify a Node.js version or explain how to install Claude Code.
How do you use this agent?
In Claude Code, start with:
/atlas:get-to-knowA typical workflow is:
/design:create-design-md
/brainstorm:prototype
/plan:create-phase "phase-01-mvp"Then switch to Claude Code plan mode and build from the generated phase.html. For a non-UI change, use /plan:visual "what to plan". If you choose Playwright CLI, the README gives this installation command:
npm install -g @playwright/cli@latestWhat are this agent's strengths and limitations?
- Provides HTML wireframes, clickable React prototypes, and visual plans for reviewing an idea before implementation.
- Uses targeted Q&A to resolve important unknowns and captures wireframes, data flow, and decision matrices in one HTML phase document.
- Includes three CLAUDE.md modes, 17 skills, 6 agents, and domain-organized slash commands.
- In Autonomous mode, the
free-willskill records important decisions, alternatives, and rationale for later review.
- The toolkit targets Claude Code; the source does not establish compatibility with other model providers or agent platforms.
- Users need Node.js/npm, terminal and filesystem access, and familiarity with Claude Code commands.
- Optional Playwright MCP, Playwright CLI, and PostgreSQL MCP integrations add setup work when selected.
- The README does not state a Node.js version, provide license details, or specify usage costs.
How does this agent compare with similar options?
Compared with spec-driven development frameworks named in the README, including spec-kit, BMAD, and Get Shit Done, ATLAS uses wireframes, prototypes, and a single HTML phase document for early exploration and planning, then hands execution to Claude Code’s native plan, goal, and workflow features. It aims to reduce multi-file Markdown planning and multi-phase process overhead.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| SWE-ATLAS This agent | 42 · Major gaps | CLIFree + model costs | ★ 402 | 3mo ago | Python | Claude Code |
| InPlan Interactive Planning Editor | 81 · Good | CLIFreemium | ★ 24 | 1mo ago | TypeScript | Claude Code |
| Okto Pulse | 80 · Good | CLIFree | ★ 51 | 2d ago | Python | Codex · Claude Code |
| Smart Ralph | 73 · Some gaps | Agent plugin / skillFree + model costs | ★ 559 | 1mo ago | Shell | Codex · Claude Code |
How does FollowAgents rate this agent?
Why each dimension lost points
The evidence shows that scaffolded skills are unselected by default, PostgreSQL MCP is opt-in, browser automation offers isolated or persistent profiles, and collaborative mode retains a review-and-commit loop. These support user control, so confirmation scores 2. Deductions reflect the autonomous mode, which removes the approval loop and permits unattended decisions, plus database and browser integrations without documented permission scope, data-flow detail, sensitive-data safeguards, rollback, or undo guidance. Dependency security is supported only by a Playwright CLI installation note; no audit or mitigation process is shown.
The README describes the workflow and repository structure in detail, but it both emphasizes Claude Code's native planning and workflows and presents ATLAS as providing execution and QA capabilities; the role boundary is somewhat unclear, so self-consistency scores 1. It lists Node.js 18+ and external MCP, Playwright CLI, and submodule prerequisites, but gives no failure messages or recovery paths.
The material targets Claude Code users and describes wireframing, prototyping, planning, and QA scenarios, with three CLAUDE.md modes and skills installed à la carte. Trigger descriptions for autonomous judgment include some mechanical conditions, but terms such as medium-to-high stakes remain broad. Environment fit is limited by dependencies on Node, Claude Code, MCP, PostgreSQL, and browser tools without complete setup or compatibility guidance.
The README organizes information through repository structure, commands, integrations, and quick-start guidance, and provides an example workflow and comparison-style FAQ information. Domain-based command names and the skill table support stable naming. package.json declares MIT, an author, and a version, supporting a license score of 2. Deductions reflect the lack of explicit known limitations, a changelog, and clear maintenance responsibility; an author field alone does not establish responsibility or an update path.
Wireframes, clickable prototypes, and single-file HTML phase documents are concrete outputs that can feed Claude Code's subsequent workflow, and optional skill selection may avoid unnecessary installation. Claims about token savings and quality gains are largely product assertions without comparative evidence. The README also acknowledges that HTML costs more to generate, while unattended operation may add review costs, so cost-benefit evidence is limited.
The README and package.json support some checkable facts, including commands, version, Node requirement, and MIT metadata. Most claims about utility and quality are assertions without tests or independent corroboration, so cross-source corroboration is weak. Some feature descriptions are clearly presented as product behavior, but the material does not consistently separate demonstrated facts from inference and promotional claims.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- Autonomous mode removes the approval loop. Before unattended use, carefully review decision logs and the permissions available to tools and integrations.
- The material does not explain credential, persistent browser profile, or database-data handling and cleanup, or how to recover from failed operations.
- Claims about token savings, quality gains, and one-shot app creation are not verified by comparative data or independent evidence in the supplied material.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
Full Changelog: https://github.com/syahiidkamil/Software-Engineer-AI-Agent-Atlas/compare/v2.6.3...v2.6.4View on GitHub ↗ -
Release notes excerpt
Full Changelog: https://github.com/syahiidkamil/Software-Engineer-AI-Agent-Atlas/compare/v2.6.2...v2.6.3View on GitHub ↗ -
Release notes excerpt
Full Changelog: https://github.com/syahiidkamil/Software-Engineer-AI-Agent-Atlas/compare/v2.6.1...v2.6.2View on GitHub ↗
FAQ
Can I use it with Codex or another coding assistant?
What services or credentials do I need to configure?
Can it generate an app unattended?
/atlas:oneshotapp "idea", with automated decisions, a build loop, and review steps. This depends on Claude Code, and Autonomous mode removes the approval loop.What does phase planning produce?
/plan:create-phase outputs a self-contained phase.html that can contain a low-fidelity wireframe, clarifications, and decision matrices for Claude Code plan mode.