README.md

metadata

license: apache-2.0
pretty_name: Sigil computer-use environment from a one-line request (REA)
tags:
  - computer-use
  - cua
  - gui-agent
  - rl-environment
  - agentenv
  - desktop
  - linux
  - rea
  - ebook
  - epub
  - publishing
  - from-request

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

The real Sigil 2.0 EPUB 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 29 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. Sigil opens a short three-chapter ebook, Documents/lantern-keeper.epub, at launch: The Lantern Keeper's Daughter, whose chapter two title reads "A Letter from the Mainlnad". The agent has to fix the misspelled word in that title and save the book. 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

In Sigil, the title of chapter two in my ebook Documents/lantern-keeper.epub misspells "Mainland" as "Mainlnad". Fix that word in the chapter title and save the book.

bundle/artifacts/fix-chapter-title-typo-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
check-1	text_equals	Documents/lantern-keeper.epub
chapter-two-text-kept	text_contains	Documents/lantern-keeper.epub
chapter-one-title-kept	text_equals	Documents/lantern-keeper.epub
chapter-three-title-kept	text_equals	Documents/lantern-keeper.epub

An .epub file is a zip of XHTML files, and REA reads the chapter files inside it. The checks read each chapter's h1: chapter two's must read "Chapter Two: A Letter from the Mainland" exactly, chapters one and three must keep their titles, and chapter two's text must still contain its closing sentence. REA's draft suggested the chapter two title check, kept exactly as written; Claude Code added the other three.

Results

3 of 3 runs passed, about $0.90 per run.

Run	Outcome	Turns	Cost	Video
1	passed	12	$1.04	video
2	passed	24	$0.88	video
3	passed	22	$0.77	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (sigil 2.0.1).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/SigilFromRequestComputerEnv-REA --repo-type dataset --local-dir sigil-request
agent-env run sigil-request/bundle --task fix-chapter-title-typo --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 a Sigil environment where the agent fixes a word in a chapter title of a short ebook and saves it.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 29 minutes. The steps were REA's usual ones:

1. Capture. REA launched Sigil 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: the title reads Mainlnad; the kept checks pass
after_edit	pass
near_miss	fail, as expected: the word was changed to "Mainlands"
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

• Only chapter two's heading is checked. The book's table of contents (OEBPS/toc.ncx) also spells the title "Mainlnad", by design: fixing it there too, or leaving it, makes no difference to the checks.
• The seed holds a stale working copy of the book under ~/.local/share/sigil-ebook/sigil/workspace, captured while Sigil had the book open; Sigil normally deletes it on exit. It also holds the storage and GPU cache files of Sigil's built-in preview. Sigil started cleanly with them in the witness and in every run.
• The computer installs whatever sigil Ubuntu 24.04's archive holds at build time (2.0.1 for these runs); it is not pinned.
• 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. Sigil itself is not included: the computer installs it from Ubuntu's archive (sigil 2.0.1) at build time, under its own licence (GPL-3.0-or-later).