Skip to content
Conduit

skill

appatlas

AppAtlas — map every screen of any app. Claude Code skill + plugin for end-to-end app walkthroughs (web + iOS/Android): screenshots, flow recordings, behavior specs, and a Mobbin-style browsable atlas

sleepyw33kday/appatlascommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

The download comes from github.com/sleepyw33kday/appatlas itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 17, 2026

🔭 AppAtlas

Map every screen of any app — automatically.

AppAtlas is a Claude Code skill + plugin that walks an entire application end to end and produces a complete atlas: every screen, sub-tab, setting, and flow — captured as screenshots, flow recordings, and behavior specs, then assembled into a Mobbin-style browsable index.

Works on web apps (via the Claude-in-Chrome tools) and iOS / Android apps (via the Xcode simulator or Android emulator).

▶ Explore a live atlas  ·  Landing page  ·  Install  ·  How it works

The live atlas above is a real, explorable AppAtlas output: 350 screens of a production SaaS, with the collapsible app-structure tree, live filtering, search, and per-section behavior specs.

▶ Explore the live atlas — 350 real screens, tagged and linked.


Why

Documenting a product's full surface — for UX research, competitive teardowns, onboarding a team, or QA regression baselines — is slow, manual, and always goes stale. AppAtlas turns "spend two days clicking through and screenshotting everything" into one instruction. Claude orchestrates a fleet of subagents that crawl the app systematically, write down exactly what each control does, and hand you a single self-contained page you can filter, search, and share.

What you get

  • A folder tree — one directory per app section, each with a spec.md behavior spec and a screenshots/ set.
  • Behavior specs — every button, dropdown, filter, toggle, modal, empty state, and gated/upgrade screen, described in plain language.
  • Flow recordings — short GIFs of the key product moments (a record drill-down, a generation completing).
  • A browsable atlas — a single self-contained HTML page with four ways in, all data-driven:
    • an auto-built tree (nav group → page → sub-tab → modal)
    • a faceted, Mobbin-style tag rail (page types, flows, UI patterns, states) that filters across the whole atlas and uses a controlled vocabulary so tags stay consistent as you crawl more apps
    • a structure panel on every screen (page type, features, element inventory, states, tags)
    • a Flows graph that wires screens together by semantic links (what opens a modal, navigates, switches a tab), laid out per product journey Light + dark, no external dependencies.

A recorded flow, captured automatically

How it works

AppAtlas encodes a battle-tested orchestration pattern, not just a prompt:

  1. Recon — Claude signs in, walks onboarding capturing every step, maps the full navigation, and writes an agent briefing + folder skeleton.
  2. Crawl — one Sonnet subagent per app section, strictly sequential per UI resource (a browser tab group or a simulator is a single shared surface — parallel agents fight over it). Each captures screenshots and writes its spec.
  3. Review → fix → coverage audit — parallel QA agents verify every screenshot against its spec, then a feature-level audit classifies each control as walked / shallow / skipped / missed and fills the gaps (the ones that hide in avatar menus, nested dialogs, and org-vs-project route twins).
  4. Atlas — a generator assembles the self-contained browsable index.

It ships with a blocker playbook for the things that derail an automated crawl: client-side paywall overlays, trial/checkout hand-off (it never enters payment data — it hands the checkout to you), quota budgeting, blank-shell/renderer recovery, and mid-crawl subagent deaths.

Built and verified with a TDD-for-skills process on a real 13-section, 350-screenshot teardown of a live SaaS.

Install

As a plugin (Claude Code marketplace)

/plugin marketplace add sleepyw33kday/appatlas
/plugin install appatlas@appatlas

As a skill (manual)

git clone https://github.com/sleepyw33kday/appatlas
cp -r appatlas/skills/appatlas ~/.claude/skills/appatlas

Use

Ask Claude Code for a walkthrough of any app:

Do an appatlas of https://app.example.com — document every screen and setting, screenshots + specs, and give me a browsable index.

Or point it at a mobile build:

AppAtlas our iOS app in the simulator — every tab, every setting, and a browsable atlas at the end.

Requirements

  • Claude Code with the Claude-in-Chrome browser tools (for web apps).
  • For mobile: Xcode + simulator with idb (iOS), or Android SDK + emulator with adb.
  • An account on the target app you're authorized to document. AppAtlas never enters payment credentials, never performs destructive or irreversible actions, and hands paid checkouts back to you.

What's in this repo

appatlas/
├── .claude-plugin/
│   ├── plugin.json          # plugin manifest
│   └── marketplace.json     # marketplace manifest (this repo is its own marketplace)
├── skills/appatlas/
│   ├── SKILL.md             # the skill: phases, iron rules, blocker table, mobile lane
│   └── references/
│       ├── playbook.md      # briefing template, capture workflow, coverage audit
│       ├── schema.md        # per-screen record shape (tree / tags / structure / links)
│       ├── taxonomy.md      # controlled vocab: page types, flows, UI-pattern tags, edge types
│       └── build_atlas.py   # the enhanced viewer generator (tree + tags + structure + flow graph)
├── assets/                  # listing screenshots + demo GIF
└── docs/                    # GitHub Pages landing page

License

MIT © sleepyw33kday

↑↓ navigate openesc close