earakely-scale/RedNotebookFromRequestComputerEnv-REA

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

The real RedNotebook 2.32 journal 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 20 minutes, with only REA's build skill and MCP tools; How it was made has the request and the steps. RedNotebook opens its journal at launch, on today's date. The journal holds four synthetic October 2026 entries, kept in ~/.rednotebook/data/2026-10.txt; the one for 20 October ends "The rosemary roots had circled the whole pot." The agent has to go to Tuesday, 20 October 2026, add the sentence "Moved the rosemary to the sunny corner by the railing." to the end of that entry, and save the journal. 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 RedNotebook, add the sentence "Moved the rosemary to the sunny corner by the railing." to the end of my journal entry for Tuesday, 20 October 2026, then save the journal.

bundle/artifacts/rosemary-journal-sentence-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
sentence-added	line_count	.rednotebook/data/2026-10.txt
entry-kept	line_count	.rednotebook/data/2026-10.txt
other-day-kept	line_count	.rednotebook/data/2026-10.txt

RedNotebook keeps each month as a YAML file of day entries, folded at about column 80, and REA reads it as text lines. The checks count lines: exactly one must contain "pot. Moved the rosemary to the sunny corner by the railing.", the new sentence right after the entry's last words, and the entry's first sentence and the 24 October entry must each still appear once. REA's draft suggested a check on the whole folded line, which Claude Code narrowed to the sentence; Claude Code discarded the draft's checks on the line the edit re-folds and on the app's log, which the export leaves out of the seed, and wrote the two kept-text checks.

Results

3 of 3 runs passed, about $0.75 per run.

Run	Outcome	Turns	Cost	Video
1	passed	10	$0.96	video
2	passed	11	$0.47	video
3	passed	12	$0.81	video

Run it

You need Docker. agentenv-framework builds the computer: Ubuntu 24.04 with the app installed from Ubuntu's archive (rednotebook 2.32).

pip install agentenv-framework==0.9.1431
hf download earakely-scale/RedNotebookFromRequestComputerEnv-REA --repo-type dataset --local-dir rednotebook-request
agent-env run rednotebook-request/bundle --task rosemary-journal-sentence --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 RedNotebook environment where the agent adds a sentence to an existing day's journal entry and saves the journal.

It asked only the skill's upfront questions, wrote no app-specific scripts, and finished in about 20 minutes. The steps were REA's usual ones:

1. Capture. REA launched RedNotebook 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 sentence 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: the 20 October entry has no rosemary sentence; the kept checks pass
after_edit	pass
near_miss	fail, as expected: the sentence was changed to end "by the railings."
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

• The check depends on the new sentence staying on one line of the YAML file, after "pot.". It does for this entry as asked, but a correct sentence that YAML folds across two lines, for example because other words were added to the entry, fails.
• The seed records a last backup date of 2026-11-06 in RedNotebook's configuration, so when the computer's clock is well past it, RedNotebook may open with a backup reminder. This is untested.
• The seed also holds the storage files of RedNotebook's built-in web view under ~/.local/share/rednotebook, which the capture picked up.
• The computer installs whatever rednotebook Ubuntu 24.04's archive holds at build time (2.32 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. RedNotebook itself is not included: the computer installs it from Ubuntu's archive (rednotebook 2.32) at build time, under its own licence (GPL-2.0-or-later).