earakely-scale/InkscapeFromRequestComputerEnv-REA

Inkscape: a computer-use environment built from a one-line request

The real Inkscape 1.2 vector graphics editor running on an Ubuntu 24.04 desktop, as an agentenv computer environment. An agent completes the task only through screenshots, mouse and keyboard, and a verifier grades the saved files with no model.

Claude Code built this environment from a one-line request in about 19 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. In the environment, Inkscape starts on its first-run Welcome screen with no document open. The agent has to open the poster in Documents/poster.svg, change its headline from Spring Plant Swap to Autumn Seed Exchange, and save. It sees only the screen and acts only with the mouse and keyboard.

Watch

• Agent runs (Claude Code with Claude Sonnet 5.5): run 1, run 2, run 3
• Witness, which proves the task can be done through the computer: witness.mp4

Each video is the computer's own screen recording at 10 frames per second.

The task

Change the poster's headline in Documents/poster.svg to Autumn Seed Exchange and save it.

bundle/artifacts/poster-headline-check/verify.py reads the app's files through the computer's data/get view and passes only if every check holds:

Check	Kind	File
headline-text	value_equals	Documents/poster.svg
date-unchanged	value_equals	Documents/poster.svg

The checks read the saved SVG as XML and find each line of text by its layer label and id, never by position: the headline must read "Autumn Seed Exchange" exactly, and the date line must be unchanged. The headline's pointer is identical, character for character, to the one in the hand-built Inkscape environment, though this poster's wording was invented independently.

Results

3 of 3 runs passed, about $0.92 per run.

Run	Outcome	Turns	Cost	Video
1	passed	23	$1.55	video
2	passed	21	$0.51	video
3	passed	18	$0.70	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (inkscape 1.2.2).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/InkscapeFromRequestComputerEnv-REA --repo-type dataset --local-dir inkscape-request
agent-env run inkscape-request/bundle --task poster-headline --agent claude-code --model <model>

The agent needs a model provider configured for agentenv. --task smoke only boots the computer, with no agent. bundle/README.md describes the bundle in detail.

How it was made

Claude Code built this environment from a one-line request, given only REA's build skill and REA's MCP tools:

Build me an Inkscape environment where the agent changes a poster's headline and saves it.

It asked only the three questions the skill asks up front, wrote no app-specific scripts, and finished in about 19 minutes. The steps were REA's usual ones:

1. Capture. REA launched Inkscape on a Linux desktop and recorded its package, windows and opened files while the edit was made. It also recorded the app's state folders before and after.
2. Export. REA turned the capture into an agentenv computer_app. It installs the app, seeds the starting files, starts the app, and waits for its main window. It also serves typed views of the app's files for verifiers.
3. Task. The task's checks were validated against the capture: they fail on the starting state and pass on the captured end state.
4. Witness. In a fresh computer, the edit was made through computer actions alone, and the checks were read at four stages. They held on the first try:

Stage	Checks
seed	fail, as expected
after_edit	pass
near_miss	fail, as expected: the headline was retyped as "Autumn Seed Exchnage"
after_reset	fail, as expected: data/reset restored the seed

The agent runs came after the build: Claude Code with Claude Sonnet 5.5 (--model anthropic/claude-sonnet-5-5 --no-emulate, on an amd64 host) solved the task 3 times, each run recorded.

What's inside

bundle/      the agentenv bundle: envs/ (app module, seed files), tasks/, artifacts/ (verifier), evals/
media/       witness and agent-run recordings (MP4) with short GIF previews
results.json the runs above, with each check's result

Limits

• Inkscape starts on its first-run Welcome screen with no document open, so the agent must open Documents/poster.svg itself. That makes this task harder than the hand-built Inkscape environment's, where the poster is already open: these runs took 18 to 23 turns, against 10 there.
• The new headline is longer than the old one and, at 48 pt, runs past the page's right edge. The checks read only the text, so this is cosmetic; all three agents pointed it out and left the size alone.
• The checks expect the headline to stay one tspan inside the text with id headline, in the layer labelled Layer 1, and to match exactly. A correct-looking poster whose headline was retyped as a new text object, moved to another layer or split over two lines fails.
• The computer installs whatever inkscape Ubuntu 24.04's archive holds at build time (1.2.2 for these runs); it is not pinned.
• The seed also holds an identical copy of the poster under ~/.local/share/Trash/files, which the capture picked up; the checks read only Documents/poster.svg.
• Three runs is a small sample: the solve rate shows the task is solvable, not how hard it is.

Licences

The bundle code, the task and the synthetic seed files are Apache-2.0. Inkscape itself is not included: the computer installs it from Ubuntu's archive (inkscape 1.2.2) at build time, under its own licence (GNU GPL).