reading-routine

Turn your 12,000-item reading backlog into a daily triage digest — for Claude Code Routines, local cron, or manual review.

The problem

You have too many saved articles. Every morning you open the reading app, get overwhelmed, close it. Six months in, 90% of your backlog you'll never touch.

The problem isn't storage — you already have Raindrop, Readwise, Burn, or a Pocket CSV rotting in ~/Downloads. The problem is triage: which 5 of the 12,000 are worth your next 30 minutes?

The fix

reading-routine is a backend-agnostic CLI that:

1. Eats your backlog (JSON in — from any source)
2. Scores each item (freshness × depth × tag signal)
3. Spits out a ranked markdown digest (or the Claude Code Routine prompt that wraps all of it)

It doesn't know or care where your reading list lives. It does one thing: rank.

Install

npm install -g reading-routine
# or run without installing:
npx reading-routine help

Quick start

# 1. Try with the sample backlog
curl -s https://raw.githubusercontent.com/Fisher521/reading-routine/main/examples/backlog.sample.json \
  | reading-routine digest --top 5

# 2. Print the Claude Code Routine prompt (paste into claude.ai/code/routines)
reading-routine routine-prompt

# 3. Or run locally on your own backlog JSON
reading-routine digest --input ~/my-backlog.json --top 10

Three ways to run it

1. As a Claude Code Routine (cloud — hands off)

Claude Code Routines (announced April 15, 2026) let Claude run your workflows on a schedule, even with your laptop closed. reading-routine ships a ready prompt:

reading-routine routine-prompt | pbcopy

Then open claude.ai/code/routines, paste, set schedule (e.g. daily 9 AM), pick the MCP connectors (Burn / Raindrop / Readwise) and the delivery channel (Slack / Email).

Morning arrives → digest in your inbox. You didn't open your laptop.

2. As a local cron (laptop open — private)

# crontab -e
0 9 * * * /usr/local/bin/reading-routine digest \
  --input ~/Library/reading-routine/backlog.json --top 10 \
  | tee ~/Library/reading-routine/today.md \
  | /usr/local/bin/terminal-notifier -title "Today's reading"

You handle the data pull (any script that outputs JSON); reading-routine handles the ranking.

3. As a manual one-off

# Paste your backlog URL list into a file → parse to JSON → digest.
reading-routine digest --input ~/Downloads/backlog.json

Backlog format

JSON array of objects:

[
  {
    "url": "https://example.com/article",
    "title": "Why X matters",
    "saved_at": "2026-04-10T14:00:00Z",
    "tags": ["rust", "performance"],
    "word_count": 2800,
    "snippet": "optional preview text"
  }
]

Required: url, title. Everything else improves ranking but is optional.

Adapters (bring your own)

Because reading-routine just wants JSON, any source works. Examples:

From Burn (MCP or CLI):

burn-mcp export-queue --status unread --limit 200 | reading-routine digest

From Raindrop.io:

curl -sH "Authorization: Bearer $RAINDROP_TOKEN" \
  "https://api.raindrop.io/rest/v1/raindrops/0?perpage=100" \
  | jq '[.items[] | {url: .link, title, saved_at: .created, tags, word_count: .word_count}]' \
  | reading-routine digest

From Readwise Reader:

curl -sH "Authorization: Token $READWISE_TOKEN" \
  "https://readwise.io/api/v3/list/?location=new" \
  | jq '[.results[] | {url: .source_url, title, saved_at: .created_at, tags: (.tags // [] | keys), word_count}]' \
  | reading-routine digest

From a Pocket CSV (for folks who exported before the Nov 2025 deadline):

# One-liner csv → json conversion — tags are pipe-separated in Pocket, hence the split
node -e "const csv=require('fs').readFileSync('pocket.csv','utf8');const lines=csv.split('\n').slice(1);console.log(JSON.stringify(lines.filter(Boolean).map(l=>{const[title,url,ts,cursor,tags]=l.split(',');return{title:title.replace(/\"/g,''),url,saved_at:new Date(Number(ts)*1000).toISOString(),tags:(tags||'').split('|').filter(Boolean)}})))" \
  | reading-routine digest

How the score works

freshness × 0.6 + depth × 0.3 + tag_bonus

• Freshness: peaks around day 3 (out of the hot-take zone, still relevant), decays over ~60 days
• Depth: 2k–6k-word pieces score highest (long-form payoff zone)
• Tag bonus: +0.1 if curated with tags (signal of intent, not drive-by save)

Simple on purpose. The score is a first pass — if you plug it into the Routine prompt, Claude re-ranks against your current priorities in Step 3.

Part of the Burn ecosystem

| Tool | What it does | |---|---| | burn-mcp-server | MCP server (26 tools) for Burn reading triage | | burn451 | Terminal CLI for your reading queue | | morning-brief | Daily briefing CLI | | reading-digest | Weekly digest from your bookmarks | | reading-routine (this) | Backend-agnostic daily triage CLI |

Roadmap

• [x] v0.1 — Core CLI, JSON in, markdown out, Claude Routine prompt generator
• [ ] v0.2 — Bundled source adapters (reading-routine pull burn, reading-routine pull raindrop)
• [ ] v0.3 — Config file (~/.reading-routine.yaml) with priorities so the score learns your focus
• [ ] v0.4 — Built-in delivery: Telegram / email / Slack webhooks

Why this exists

Routines just landed (HN #5, 700 pts on April 15, 2026). The reading-triage use case is obvious — everyone has a backlog, nobody wants another web UI. What's missing is a boring CLI that does ranking well, so the Routine prompt stays short and composable.

Building publicly at @hawking520. Feedback, PRs, adapters for other sources — all welcome.

Contributing

PRs welcome. Most useful contributions right now: adapter examples for sources I don't have (Instapaper, Wallabag, Matter, Obsidian Web Clipper).

About

Built by @hawking520 — exploring AI-era content & attention workflows in public.

Pairs naturally with Burn — the 24h-countdown reading queue that forces you to absorb, not hoard.

License

MIT — see LICENSE