Every app is open source now.

Not the original's source. Its behaviour, rewritten from scratch by agents that never see the original, behind a wall that hooks enforce and log.

/plugin marketplace add remakecraft/remakecraft
/plugin install remakecraft@remakecraft

A Claude Code plugin, MIT, early alpha.Quickstart Source on GitHub

Brushling demo recording

Brushling, a Windows 98-era Paint clone built by Remakecraft in {{BRUSHLING_AGENT_HOURS}} agent-hours for ${{BRUSHLING_COST_USD}} in tokens. Try it in your browser.

How it works

Name an app. Remakecraft inventories it, writes a functional spec, writes the tests before any code, seals about 30% of them away from the builders, builds a native rewrite with parallel agents, has a different model family review it, and gives you a parity score that counts the sealed tests too.

/remakecraft:remake <app> [--stack rust-egui|tauri|web] [--builders 1-5] [--cheap] [--out <dir>]

1InventoryEvery menu, tool, shortcut, dialog and file format, each with an idroom
2Specspec.json: behaviour only, no expressionroomYou approve the spec here
3TestsConformance tests from the spec only: ~70% visible, ~30% sealedroom
4Pilot10 features, then a cost extrapolation for the full runclean sideYou approve the cost here
5Build1 to 5 builders, one worktree each, from spec and visible testsclean side
6VerifyHash check, stub scan, parity report with sealed countsclean side
7ReviewCodex (or Claude, if Codex is absent) ranks findings that go back to Buildclean side

Runs resume from progress files and git checkpoints after a crash or a usage-limit stop. Every phase, in detail

The wall

A run has two trees. Everything that came from the original lives in the room. Builders work in the app tree, and a PreToolUse hook stops them reaching across. If the hook crashes, the call is blocked.

The room

<app>.remake-room/

Your screenshots, recordings and docs
Inventory drafts
The sealed tests

The app tree

<app>/

The approved spec
The visible tests
The code

A builder trying the room twice in the live wall test on 2026-10-13. Every denial is appended to .remake/wall.log.

role   builder   tool  Read
input  …\wall-live\brushling.remake-room\evidence\secret.txt
reason Clean-room wall: builder may not touch the room. Work from spec/ and tests/ in the app tree.
role   builder   tool  Bash
input  pwd; cat ../brushling.remake-room/evidence/secret.txt; echo "exit=$?"
reason Clean-room wall: builder may not touch the room. Work from spec/ and tests/ in the app tree.

The rules are pure logic with 14 unit tests. Read them; the wall is only worth what you can check. What the wall blocks, and why

An honest parity score

One number per clone, then the layers. The builders never see about 30% of the tests, and those sealed tests count. Nothing is rounded up.

Brushling parity

{{BRUSHLING_PARITY}}%

UI: {{UI}}%
Behaviour: {{BEH}}%
File formats: {{FMT}}%
Sealed pass: {{SEALED_PASS}}/{{SEALED_TOTAL}}

Parity is the pass rate of the conformance suite, visible and sealed, by layer. It measures what the spec says the original does, so the spec is published next to each clone. How parity is scored

Other agents

The pipeline is agent-neutral. The wall is a Claude Code feature: everywhere else the same rules ship as instructions, and nothing enforces them.

Agent	Install	Wall
Claude Code (Recommended)	`/plugin marketplace add remakecraft/remakecraft` `/plugin install remakecraft@remakecraft`	Enforced by hooks, logged
Codex	`codex plugin marketplace add remakecraft/remakecraft` `codex plugin add remakecraft@remakecraft`	Instructions only
Cursor, Gemini CLI, Copilot, OpenCode and other Agent Skills hosts	`npx skills add remakecraft/remakecraft`	Instructions only