How to set default configuration?

[HoaSoup]

Hello,

I would like to make some changes to the default configuration:

• A default path for saving output files. For example, I want the output literature review md file to be saved in my Ubuntu home folder.
• Default citation style.
• Don't search unpublished papers on Biorxiv, Medrxiv...
• Prefer to search papers of Q1, Q2.
• Prefer to search open-acess papers from 2020 onwards and papers, both open-access and paywalled, before 2020.

Thanks,
Hoa

[Imbad0202]

Thanks for the question. Short answer: ARS deliberately has no configuration file of its own, so these five preferences route through Claude Code rather than ARS.

Standing preferences: put them in your project's CLAUDE.md. Claude Code loads it at session start and ARS agents inherit it, which gives you "set once, applies every run":

# ARS preferences
- Save all ARS output files to /home/<you>/reviews/
- Citation style: APA 7
- Literature search: exclude preprint servers (bioRxiv, medRxiv, SSRN)
- Prefer Q1/Q2 journals
- Prefer open-access papers from 2020 onward; before 2020, both OA and paywalled are fine

The search-related lines land in the right place automatically: they become inclusion/exclusion criteria in the search strategy that deep-research records, so each run's Search Strategy report will show them applied.

Two things ARS does natively that are relevant:

• Preprints are already flagged even if you don't exclude them. The bibliography layer computes contamination signals on recent preprints from the major servers, and citation markers carry advisory CONTAMINATED-PREPRINT suffixes. Full exclusion is a per-project judgment, which is why it's an instruction rather than a config key.
• Citation style is a first-class option in academic-paper (APA 7 / Chicago / MLA / IEEE / Vancouver at the formatting stage), so the CLAUDE.md line just saves you restating it.

Two honest limitations:

• Journal quartiles: none of the four indexes ARS queries (Semantic Scholar, OpenAlex, Crossref, arXiv) carries JCR/SJR quartile data, so "prefer Q1/Q2" is applied as model judgment, not a deterministic filter. Expect it to be approximately right, and check the Search Strategy report rather than assuming.
• Output path: there is intentionally no output-directory setting in ARS (an earlier design round rejected one to keep the session's working directory and Material Passport path as the single anchor). The CLAUDE.md instruction above is the supported route.

The env vars that do exist (cache paths, cross-model verification, etc.) are documented in docs/SETUP.md; none of your five items lives there, which is why CLAUDE.md is the answer.

[HoaSoup]

Thank you!