license: apache-2.0
pretty_name: ghostwriter computer-use environment from a one-line request (REA)
tags:
- computer-use
- cua
- gui-agent
- rl-environment
- agentenv
- desktop
- linux
- rea
- markdown
- text-editor
- writing
- from-request
ghostwriter: a computer-use environment built from a one-line request
The real ghostwriter 23.08 Markdown 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 28 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. ghostwriter opens a short Markdown note, Documents/camping-trip.md, at launch: a camping-trip plan with a Gear checklist of four items, one of them ticked, and a Notes section. The agent has to add Water filter to the Gear checklist 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
- 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 ghostwriter, add Water filter to the Gear checklist in Documents/camping-trip.md and save it.
bundle/artifacts/add-water-filter-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 |
|---|---|---|
water-filter |
lines_present |
Documents/camping-trip.md |
added-once |
line_count |
Documents/camping-trip.md |
kept-items-near |
lines_present |
Documents/camping-trip.md |
kept-items-top |
lines_present |
Documents/camping-trip.md |
REA reads the Markdown file as text lines. The checks need the line "- [ ] Water filter", an unticked task item, only one line mentioning Water filter, and the four gear items unchanged. REA's draft suggested the new-line check and a check on the two items nearest it, both kept exactly as written; Claude Code added the count and a check on the first two items.
Results
2 of 2 runs passed, about $0.71 per run.
Run it
You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (ghostwriter 23.08.5).
pip install agentenv-framework==0.9.1431
hf download earakely-scale/GhostwriterFromRequestComputerEnv-REA --repo-type dataset --local-dir ghostwriter-request
agent-env run ghostwriter-request/bundle --task add-water-filter --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 ghostwriter environment where the agent adds an item to a Markdown checklist and saves it.
It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 28 minutes.
It took a second capture. The first bundle's prompt, "Add Water filter to the Gear checklist in Documents/camping-trip.md and save it.", did not name the app. Its witness held, but the one solve run on it never reached the computer: the agent worked in its own shell, where the file does not exist, and failed. Claude Code rewrote the prompt to name ghostwriter and captured again. Both runs in Results are on this second bundle; the failed run on the first is not counted in them. The steps were REA's usual ones:
- Capture. REA launched ghostwriter 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.
- 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. - Task. The task's checks were validated against the capture: they fail on the starting state and pass on the captured end state.
- Witness. In a fresh computer, the item was added through computer actions alone, and the checks were read at four stages. They held on the first try:
| Stage | Checks |
|---|---|
seed |
fail, as expected: there is no Water filter line; the kept-item checks pass |
after_edit |
pass |
near_miss |
fail, as expected: the new item was ticked, "- [x] Water filter" |
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 twice, 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 two runs, both on this bundle: How it was made explains the failed run on the first one. Two runs show the task is solvable, not how hard it is.
- The checks read the file as lines and don't prove the item is under the Gear heading: "- [ ] Water filter" anywhere in the file passes, under Notes for example, as long as it appears once and the four gear items are still there.
- The item must be written exactly as an unticked task, "- [ ] Water filter": a ticked box, a plain bullet or other wording fails.
- The seed also holds a backup copy of the note and the storage and GPU cache files of ghostwriter's built-in preview (
~/.local/share/ghostwriter/QtWebEngine), which the capture picked up. ghostwriter started cleanly from them in the witness and in both runs. - The computer installs whatever
ghostwriterUbuntu 24.04's archive holds at build time (23.08.5 for these runs); it is not pinned.
Licences
The bundle code, the task and the synthetic seed files are Apache-2.0. ghostwriter itself is not included: the computer installs it from Ubuntu's archive (ghostwriter 23.08.5) at build time, under its own licence (GPL-3.0-or-later).