SWE-ATLAS: a senior-engineer layer for Claude Code
SWE-ATLAS
A Claude Code template: wireframe, prototype and an HTML spec before the code, then a build with the native plan and goal modes
- 404 ★ on the repository
- Last commit
Extended access
Why this level
- The installer writes CLAUDE.md, a .claude folder with agents, commands and skills, and supporting documents into the project
- The autonomous flavor removes human approval, so the agent decides and builds the app by itself
- Browser automation through Playwright, if chosen, gives the agent control of web pages
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add workflows/swe-atlasDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- The installer writes CLAUDE.md, a .claude folder with agents, commands and skills, and supporting documents into the project
- The autonomous flavor removes human approval, so the agent decides and builds the app by itself
- Browser automation through Playwright, if chosen, gives the agent control of web pages
This is third-party code. Review the repository files before installing.
What it does
SWE-ATLAS lays out skills, subagents, commands and engineering principles in a Claude Code project. The authors' idea is to show cheaply what is being built first, then hand a finished document to Claude Code's native loop of plan mode, the /goal command and dynamic workflows. They consider heavy specification chains of many files unnecessary. At the start the agent explores the idea. /brainstorm:wireframe makes an interface wireframe as one HTML file, /brainstorm:prototype builds a clickable React prototype, and /design:create-design-md shows three design variants in the browser and locks the chosen one in a DESIGN.md file. For a change with no interface, such as a migration, /plan:visual draws a visual plan with Mermaid diagrams. Then /plan:create-phase asks only about what could derail the build and saves the phase as one self-contained HTML document. Plan mode starts from it. There is also a shortcut: /atlas:oneshotapp builds an app from an idea without stops. In the autonomous flavor of the template the free-will skill decides contested forks and records them in docs/decision_logs/.
Who it is for. For developers and product engineers on Claude Code who want to agree with the agent on what is built from a picture and a document, without long requirement texts.
Good fit when
- A new feature or screen: you need a wireframe or a clickable prototype to see the shape before generating code
- The product's look is undecided: three variants in the browser and the chosen one locked in DESIGN.md
- A refactor, migration or architecture call needs a visual plan with diagrams and a decision matrix
- You want the agent to work autonomously for a long stretch and keep a decision log for later review
- A team needs living test cases in Markdown and a run through a browser agent with Playwright
Not a fit when
- You need a strict written process with a constitution, a spec and a task ledger: here it is intentionally replaced by a wireframe and one phase document
- The work is not in Claude Code: the template is built around its commands, skills and subagents
- The project already has a carefully hand-tuned CLAUDE.md and you do not want an installer to offer its own
- You need Word, Excel and PowerPoint document skills in the bundle: the heavy document skills were removed from the set
Example request
/brainstorm:wireframe a customer account page with order history and a status filterLimitations
The template works inside Claude Code, and the Anthropic API is unavailable from Russia. The installer needs Node.js for npx. Skills are installed selectively, with none selected by default. The autonomous flavor removes the human approval loop and adds three skills that stand in for it.
The authors admit that HTML costs more tokens and time than Markdown. They treat the phase document as one needed file instead of a set of drifting ones. Claude Code's auto mode and dynamic workflows were in research preview at the time of writing. Word, Excel and PowerPoint skills are not part of the template: in an installed project they sit in the anthropics/skills submodule.
How to disable. Delete the .claude folder (if it holds none of your own files) and the CLAUDE.md file the installer added from the project, plus NOTES.md and docs/decision_logs if you do not need them. If the project already had its own CLAUDE.md, restore it from Git.
FAQ
How do I install SWE-ATLAS into a Claude Code project?
In the project folder run npx swe-atlas@latest new-project. To scaffold into a new folder, add its name as the last argument. The installer asks for the CLAUDE.md flavor, your name, project type, the skills you need, a DESIGN.md template, the browser automation method and PostgreSQL.
How does SWE-ATLAS differ from spec-driven development such as Spec Kit?
Spec chains first write a constitution, a specification, a plan and a task breakdown. SWE-ATLAS limits itself to a wireframe or prototype, answers to the questions that would derail the build, and one HTML phase document. The planning and execution loop itself stays native to Claude Code.
Which CLAUDE.md flavor should I choose?
Vanilla installs the minimum, only NOTES.md and a decision log. Collaborative is the default: the full engineer identity and a review-and-commit loop, with you in the driver's seat. Autonomous removes the approval step, adds the free-will, super-product-owner and super-ui-ux-design skills and is built for unattended runs.
What does the /plan:create-phase command do?
It asks only about the unknowns that would derail the build and puts the wireframe, answers and decision matrices into one phase.html file. Then you switch to plan mode and Claude Code builds from that file.
What is the free-will skill for?
It fires on contested forks: picking a stack, a database schema, retiring something others depend on. The skill holds several alternatives, grounds each in evidence from the code or docs, tries to refute the winner and records the outcome in docs/decision_logs/. It also fires on its own, for example after two failed fixes.
Discussion
No questions or reviews yet.
Related
WorkflowsA starter kit for Claude Code: a CLAUDE.md memory-bank template with ready hooks, subagents and skills you copy into your project
A skill that maps a repository into meaningful blocks, walks through the code on that map step by step and highlights broken spots
A ready harness for Claude Code and Codex: spec-driven development, TDD, memory between sessions, quality gates and human checkpoints
A Claude Code project template for game development: a studio hierarchy of agents, skills for every phase, hooks, rules and doc templates
edt1c-ai-template: 1C EDT project template for Claude Code
edt1c-ai-template
A repository template for 1C EDT development with Claude Code: feature specs, BSL standards and skills for query builders and spreadsheet documents
A dev process built from composable skills for Claude Code and Codex: plan, implement, review, test, commit and PR
Comparison with similar tools
Entries that share the main search query, and how they differ.
README in short
SWE-ATLAS installs skills, subagents, commands and a CLAUDE.md file into a Claude Code project and leaves execution to the native plan loop. Anthropic's document skills are attached as a submodule and are not part of the default set.