catalan-adobe/news-digest

News Digest

Fetch Google News RSS feeds by topic, extract full article content, and produce a rich markdown digest with article summaries, images, and video links.

Script Location

if [[ -n "${CLAUDE_SKILL_DIR:-}" ]]; then
  NEWS_SCRIPTS="${CLAUDE_SKILL_DIR}/scripts"
else
  NEWS_SCRIPTS="$(find ~/.claude -path "*/news-digest/scripts" \
    -type d 2>/dev/null | head -1)"
fi
if [[ -z "$NEWS_SCRIPTS" || ! -d "$NEWS_SCRIPTS" ]]; then
  echo "Error: news-digest scripts not found." >&2
  exit 1
fi

First-Run Setup

Before the first invocation, install dependencies:

cd "$NEWS_SCRIPTS" && [ -d node_modules ] || npm install

Run this at the start of every invocation. The [ -d node_modules ] check makes subsequent runs instant.

Prerequisites

Google News URLs require a headless browser to resolve. The script uses playwright-cli (must be on PATH). Verify with:

which playwright-cli || echo "playwright-cli not found — install it first"

Config

The user maintains a config.yaml with topics to track. If no config exists yet, copy the default:

DEFAULT_CONFIG="${NEWS_SCRIPTS}/../references/default-config.yaml"

Ask the user where they want their config (e.g., ai/news/config.yaml) and copy the default there. Then they can edit topics and settings.

Invocation Modes

Full digest (all configured topics)

cd "$NEWS_SCRIPTS" && node news-fetch.mjs --config /path/to/config.yaml

Ad-hoc query (no config needed)

cd "$NEWS_SCRIPTS" && node news-fetch.mjs --query "search terms"

Single topic from config

cd "$NEWS_SCRIPTS" && node news-fetch.mjs --config /path/to/config.yaml --topic "AI"

Output Processing

The script outputs a JSON array to stdout. Each item has:

• title — headline
• source — publisher name
• url — Google News redirect URL
• resolvedUrl — actual article URL after redirect. null if decoding failed
• pubDate — publication date string
• timestamp — epoch ms
• topic — which configured topic matched
• content — full article text (plain text). null if extraction failed
• paywall — true if paywall detected, false otherwise
• error — error message if extraction failed, null on success
• images — array of { src, alt } for article images (ads filtered out)
• videos — array of { src, domain } for embedded videos

Your job as the LLM:

1. Read the JSON output
2. Group stories by topic
3. For each story with content: write a 2-3 sentence summary from the full article text
4. For each story without content (paywall or error): fall back to headline-only format
5. Include the lead image (first item in images) inline when available
6. Include video links when available
7. Use resolvedUrl for article links (not the Google News redirect url)
8. Present in conversation AND save to a digest file

Digest File

Save to ai/news/YYYY-MM-DD-digest.md in the user's project. If a digest for today already exists, append new sections rather than overwriting.

Format:

# News Digest — YYYY-MM-DD

## Topic Name

### Headline here
**Source:** Source Name (2h ago) | [article](resolvedUrl)

2-3 sentence summary of the full article content generated from the
`content` field.

![Alt text](image-src)

---

### Paywalled headline
**Source:** Source Name (3h ago) | [article](resolvedUrl) | 🔒 Paywall

*(Full article not available)*

• Stories with content: full summary + lead image + video links
• Stories with paywall: true: headline only, marked with 🔒
• Stories with error and no paywall: headline only, no special marker

User Guidance

• If the user has no config yet, help them create one by copying the default and editing topics
• If the user asks to add/remove a topic, edit their config.yaml
• If the user asks for an ad-hoc search, use --query mode
• Keep digests tight: max_stories from config (default 15)