File size: 3,414 Bytes
fa29f75 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 | # RedNotebook computer app
Exported by REA's `export-desktop-application` from a capture of the `rednotebook` package
(capture Evidence `ev_19faf9f2f91a4f5b5480001053401e2393e7439ad948abe2b8e077a527ca6ce3`). It was checked against agentenv-framework
0.9.1431.
- `envs/rednotebook/` is a computer app on the `ubuntu-24.04` base with the `reference` driver, for
`amd64` and `arm64`. It installs `rednotebook` from the distribution archive, launches `sh -c cd "$HOME" 2>/dev/null; exec 'rednotebook'` at boot and is ready
when a window of class `Rednotebook` appears.
- Owned paths: `~/.local/share/rednotebook`, `~/.rednotebook`.
`data/get` returns their files and a typed JSON view of each; `data/reset` restores the seed.
- Seed: 18 file(s) made for the task, as saved at checkpoint `seed`.
Text files and dconf keys spell the capture host's name as `localhost`, `user@host` as `user@localhost`
and the home as `/home/user`, which the computer respells to its own at boot.
- Seeded as captured, though the app keeps them for its own running, so a stale one can change how it starts:
WebKit or Chromium storage `~/.local/share/rednotebook/hsts-storage.sqlite`, `~/.local/share/rednotebook/mediakeys`, `~/.local/share/rednotebook/storage`.
- Left out of the seed: the app's logs, which record facts about the capture host, such as its kernel, CPU and memory,
and which it does not need to start: `~/.rednotebook/rednotebook.log`.
- `tasks/smoke.json` deploys the app.
Run the task with `agent-env run <this folder>`, or only the deploy with `--task smoke`.
The seed was made for this task, not taken from a user's own data; read `envs/rednotebook/seed/` before publishing the bundle.
## Task `rosemary-journal-sentence`
`tasks/rosemary-journal-sentence.json` deploys the app, has the agent work through the computer from the prompt below, and grades
the outcome with `artifacts/rosemary-journal-sentence-check/verify.py`, which reads the app's `data/get` views and needs
no model. It grades once the app's files have held for 3 s, so a save the app makes on its own within that
time of the agent stopping counts; each result's `settled` is false when they were still changing 15 s after the
first read. `evals/rednotebook.toml` names it, so a run of the bundle runs it.
> 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.
Checks, each of which must pass:
- `sentence-added`: `{"kind":"line_count","file":".rednotebook/data/2026-10.txt","contains":"pot. Moved the rosemary to the sunny corner by the railing.","equals":1}`
- `entry-kept`: `{"kind":"line_count","file":".rednotebook/data/2026-10.txt","contains":"Repotted the basil and the rosemary on the balcony.","equals":1}`
- `other-day-kept`: `{"kind":"line_count","file":".rednotebook/data/2026-10.txt","contains":"Cycled to the farmers market with Priya.","equals":1}`
REA kept the task because its checks fail on the seed (`sentence-added`) and all
pass on the captured end state.
To measure a solve rate, run the task k times (3 by default) with the agent and model to measure, and count the runs
whose summary line says `passed`; a run that says `failed` did not finish and is run again rather than counted:
agent-env run <this folder> --task rosemary-journal-sentence --agent claude-code --model MODEL
|