---
name: reading-digest
description: Use when the user asks for a reading digest, paper digest, "what's new in {field}", a weekly roundup of research papers, newsletters, or saved articles, or when a scheduled reading routine runs. Finds new papers and long-form pieces on the user's topics, picks the few worth their time, and explains each in plain language with links.
metadata:
  id: reading-digest
  version: "1.0.0"
  author: BotPlaza
---

# Reading Digest

> This week's papers worth reading, each explained in three lines.

## What it does

The handful of papers and essays from this week worth your time, each explained in three plain sentences, so you can decide what to actually read.

## When to use

- "What's new in {field} this week", "summarize this week's papers on {topic}", "go through my saved articles".
- A weekly routine (default: Friday 5:00 PM).

## Requires

Web search and web access (arXiv, journals, preprint servers). Optional: a read-later app (Readwise, Instapaper, Notion) to save picks.

## Schedule

Default: Weekly, Friday 5:00 PM. Run it from a routine or scheduled task at that time; the user can change the time in the routine, not in this skill.

## Inputs

| Setting | Default | Notes |
|---|---|---|
| `topics` | ask | E.g. "LLM agents", "CRISPR delivery", "urban planning". Up to 5. |
| `sources` | preprints, journals, quality long-form | Toggle: arXiv, bioRxiv/medRxiv, SSRN, journals, newsletters, the user's read-later list, specific blogs. |
| `window` | last 7 days | |
| `count` | 5 | Items in the digest. |
| `level` | informed non-specialist | Or `expert` (keep jargon, add methods detail) or `beginner`. |
| `include_skipped` | yes | A short "also out this week" list. |

## Steps

1. **Search** each topic across the enabled sources for items published in the window. Collect title, authors, venue, date, and link.
2. **Screen** by relevance to the topic, novelty (new result vs incremental), credibility (venue, methods, data), and attention from credible experts. Note when something is a preprint and not peer reviewed.
3. **Read** the abstract, introduction, figures, and conclusion of each pick at minimum. For articles, read the whole piece.
4. **Explain** each pick in three sentences: what they did, what they found, why it matters (or why to be skeptical).
5. **Add one honest caveat** per item (small sample, preprint, single lab, industry-funded, benchmark only).
6. **Write** in the format below, ordered by how much the user should care.

## Output format

```markdown
# Reading digest · {topics} · week of {date}
*{N} items reviewed, {count} picked*

## 1. {Title}
{Authors (first author et al.)} · {Venue or "Preprint"} · {date} · ~{X} min read
**What:** {one sentence} **Found:** {one sentence} **Why it matters:** {one sentence}
*Caveat: {one line}*
[Read]({url})

## Also out this week
- [{Title}]({url}), {one line}
```

## Edge cases

- **Nothing notable:** say so and list the 2 or 3 closest items instead of padding.
- **Paywalled:** use the abstract and say so; link a free version if the authors posted one.
- **Retractions or major corrections** to anything in past digests: lead with them.

## Boundaries

Summarize and link; don't reproduce full text. Never state a finding more strongly than the authors do.