I'd like a personal daily arXiv digest. Please build it in this folder, step by step, explaining briefly what you are doing as you go. I am not a programmer, so keep things simple.

1. **My interests.** Interview me with a few short questions: which arXiv categories I follow (e.g. math.AG), what my research is about, and names of people whose work I always want to see. If I give you my arXiv author page, read it. Write what you learn to `interests.md`, in plain language I can edit later, with a section for "must see", "interesting", and "not for me".

2. **Fetching.** Write a small Python script, `fetch.py`, using only the standard library, that downloads today's new papers from the arXiv RSS feed for my categories (the address is `https://rss.arxiv.org/rss/` followed by the categories joined with `+`, e.g. `https://rss.arxiv.org/rss/math.AG+math.NT`). It should save each paper's title, authors, abstract, link and announcement type (new, cross-list or replacement) to `today.txt`. Check first that Python 3 is installed; if not, help me install it. Run the script to check that it works.

3. **The digest.** Write `DAILY.md`: instructions to yourself for the daily job. Run `fetch.py`, read `today.txt` yourself (no extra scripts to parse it), compare it against `interests.md`, and write `digests/` + today's date + `.md` containing the handful of papers I would most want to see (usually 3 to 10, fewer on a quiet day), each with its link and one sentence on why it matters *to me*. Mention replacements only if they look important. Then list the other titles briefly, so I can spot anything you missed. If the feed is empty (weekends, holidays), just say so. Each run should end by showing me the digest.

4. **First run.** Follow `DAILY.md` now and show me the digest. Read `today.txt` directly with your file-reading tool, in a few large chunks if it is long; do not write helper scripts to process it.

5. **Finish.** Tell me in two or three sentences how to change what counts as interesting (edit `interests.md`, or just tell you), and how to make this run every weekday morning.

Keep it small: a few files, no extra packages, nothing clever. The whole thing should take about ten minutes.