CLAUDE CODE · SKILL & PLUGIN

Map every screen
of any app.

AppAtlas walks a whole application end to end, covering every screen, sub-tab, setting, and flow. You get screenshots, flow recordings, behavior specs, and a browsable atlas you can filter and share. Web and mobile.

appatlas · browsable indexOpen live ↗
The AppAtlas browsable gallery: a filterable grid of 350 captured app screens beside a collapsible app-structure tree
350screens in one run
13sections mapped
web + appiOS, Android
1 fileself-contained atlas

Not a pile of screenshots. A structured atlas.

Every section gets a behavior spec covering each control, empty state, and gated screen, cross-linked to the exact captures. Read it like a map key.

A · ATLAS

One browsable index, four ways in

A single self-contained page: an auto-built app tree, a faceted Mobbin-style tag rail (page types, flows, patterns) that filters across the whole atlas, a structure panel on every screen, and a Flows graph wiring screens together by what opens what. Light and dark, zero dependencies.

B · COVER

Complete coverage

Every nav item, dropdown, and little setting, audited at the feature level so nothing hides in a menu.

C · SPEC

Behavior specs

What each button does, what options exist, how the UI responds. Not just what it looks like.

D · REC

Flow recordings

Short GIFs of the key product moments, captured as the crawl exercises them.

E · CREW

Subagent crew

A fleet of Claude subagents crawl in parallel where safe, sequential where they would collide.

F · SAFE

Safe on live accounts

No payment data, no destructive actions. Paid checkouts hand back to you; trials it starts, it can cancel.

See the whole app as a graph.

Every screen is tagged and linked. Filter by an auto-built tree or by Mobbin-style tags, open any screen for its structure and features, then switch to Flows to see the app as a node graph, screens wired together by what opens what.

app treefaceted tagsstructure readsemantic linksflow graph

Explore the live atlas ↗

The Flows view: screens of one product journey laid out left to right as a node graph, connected by curved edges for what opens or navigates to what

Key moments, captured on the way through.

A record drill-down, a generation completing: the moments worth watching become short GIFs during the crawl, not a manual re-run afterward.

A recorded flow: opening a response, viewing its detail, toggling brand-name highlighting, and switching to citations

How a walk runs

An orchestration pattern, not just a prompt, with a blocker playbook for the things that derail automated crawls.

ReconClaude signs in, walks onboarding capturing every step, maps the full navigation, and writes an agent briefing plus a folder skeleton.
CrawlOne subagent per section, sequential per UI surface (a browser tab or a simulator is shared, so parallel agents would fight over it). Each captures and specs its section.
Review, fix, coverage auditParallel QA agents verify every screenshot against its spec, then a feature-level audit fills the gaps that hide in menus, dialogs, and scoped route twins.
AtlasA generator assembles the self-contained browsable index, the thing you actually share.

Install

This repo is its own Claude Code plugin marketplace.

# as a plugin

# add this repo as a marketplace
/plugin marketplace add sleepyw33kday/appatlas
# install it
/plugin install appatlas@appatlas

# as a skill (manual)

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

# then just ask

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

Requires Claude Code with the Claude-in-Chrome tools for web apps; Xcode and a simulator (idb) or an Android emulator (adb) for mobile.