The Problem

You have 200 saved articles. You'll read maybe 5 today.

Which 5?

Most reading tools sort by "date saved" — which is just FIFO. That's not a strategy, it's a queue.

reading-filter scores every item in your reading queue by relevance, novelty, and actionability, then tells you: MUST READ, SKIM, or SKIP.

Install

npm i -g reading-filter

Requires Node.js 18+ and a Burn account (free).

Quick Start

# Set your token (get it from burn451.cloud → Settings → MCP Server)
export BURN_MCP_TOKEN=your_token

# Rank your reading queue
rf

# Top 10 must-reads only
rf -n 10

# JSON output for piping
rf -f json | jq '.items.must_read[].title'

# Only unread items (flame = 24h timer)
rf --source flame

# Filter by collection
rf -c "AI Research"

Example Output

# Reading Filter Report

> 47 items scored | 8 must-read | 21 skim | 18 skip

## MUST READ (8)

- **The Local LLM Ecosystem Doesn't Need Ollama** (82.3) [deep_read] — https://example.com/...
  > Key insight: direct llama.cpp integration outperforms Ollama for production use cases

- **Claude Code Routines Are Game-Changing** (78.1) [deep_read] — https://example.com/...
  > Automate repetitive Claude Code workflows with cron-triggered prompt routines

## SKIM (21)

- **Weekly Rust Newsletter #412** (61.2) [skim] — https://example.com/...
  > Tokio 2.0 migration guide, new async patterns

## SKIP (18 items hidden)

Use `--show-skip` to see these.

---
Filtered by reading-filter | Try Burn for full reading triage

How It Works

Your Burn library
    |
    v
[Fetch bookmarks with AI metadata]
    |
    v
[Score: relevance x novelty x actionability x freshness]
    |
    v
[Rank and tier: MUST_READ / SKIM / SKIP]
    |
    v
Markdown report or JSON

Scoring

Each bookmark is scored 0-100 using four weighted dimensions:

| Dimension | Weight | Source | What it measures | |-----------|--------|--------|-----------------| | Relevance | 35% | ai_relevance | How relevant to your interests | | Novelty | 30% | ai_novelty | How new/surprising the content is | | Actionability | 20% | ai_actionability | Can you act on this? | | Freshness | 15% | created_at | Time decay over 7 days |

Tiers

| Tier | Score | Meaning | |------|-------|---------| | MUST READ | 70+ | High value, read today | | SKIM | 45-69 | Scan for key points | | SKIP | <45 | Probably not worth your time |

Options

| Flag | Short | Description | Default | |------|-------|-------------|---------| | --limit | -n | Max items to show | all | | --format | -f | Output: md or json | md | | --tier | -t | Filter: must_read, skim, skip | all | | --show-skip | | Include skip-tier items | hidden | | --category | -c | Filter by collection name | all | | --source | | flame, spark, vault | flame+spark | | --help | -h | Show help | | | --version | -v | Show version | |

Environment

| Variable | Description | |----------|-------------| | BURN_MCP_TOKEN | Your Burn MCP token (get it here) |

Burn Ecosystem

reading-filter is part of the Burn reading tools ecosystem:

| Tool | What it does | |------|-------------| | burn-mcp-server | MCP server for AI agents (26 tools) | | reading-filter | AI prioritization for your reading queue | | digest-or-burn | Burn-pile CLI for markdown hoards | | reading-routine | Daily reading triage via Claude Code Routines | | burn-daily-triage | Automated reading triage routine | | burn451-cli | Terminal UI for Burn | | Burn | The full reading triage app |

Philosophy

Reading more doesn't make you smarter. Reading the right things does.

Most people treat their reading queue like a todo list — everything gets equal weight. But a breaking industry analysis and a "10 tips for productivity" listicle are not the same.

reading-filter applies the same triage logic that Burn uses internally: score every piece by how much it matters to you, right now, and surface the signal.

Read less. Absorb more.

License

MIT

Built by @hawking520 | Try Burn for full reading triage