pulsefeed-digest

name	pulsefeed-digest
description	Generate an information digest from Hacker News, GitHub Trends, and Product Hunt. Use this skill when the user wants to collect and summarize tech news, trending repos, or new products. Supports scheduled collection (e.g., every 2 hours) and handles cache staleness automatically.

PulseFeed Information Digest

This skill runs the PulseFeed information aggregation tool to collect and summarize content from multiple sources with LLM-powered deep analysis.

Prerequisites

1. PulseFeed installed in ~/workspace/pulsefeed
2. Config file at ~/.pulsefeed/config.yaml with LLM settings:

llm:
  provider: minimax
  api_type: anthropic
  model: MiniMax-M2.1
  api_key: your-api-key
  api_base: https://api.minimax.io/anthropic
  max_tokens: 4096

cache:
  ttl_hours: 24
  db_path: ~/.pulsefeed/cache.db

Instructions

Step 1: Acquire Lock (Prevent Multiple Instances)

Before running, check and acquire a lock to prevent multiple instances:

LOCKFILE=~/.pulsefeed/pulsefeed.lock

# Check if another instance is running
if [ -f "$LOCKFILE" ]; then
  PID=$(cat "$LOCKFILE")
  if ps -p "$PID" > /dev/null 2>&1; then
    echo "Another instance is running (PID: $PID). Killing it..."
    kill "$PID" 2>/dev/null
    sleep 2
    # Force kill if still running
    kill -9 "$PID" 2>/dev/null || true
  fi
  rm -f "$LOCKFILE"
fi

# Create lock with current PID
echo $$ > "$LOCKFILE"

Step 2: Run Collection

cd ~/workspace/pulsefeed && \
python3 -m pulsefeed collect \
  --sources hackernews,github,producthunt \
  --max-items 10 \
  --format detailed \
  -o ~/.pulsefeed/output/digest_$(date +%Y%m%d_%H%M).md

Options:

• --sources: Comma-separated list (hackernews, github, producthunt, rss, twitter)
• --max-items: Total items to process (default: 50)
• --demo: Quick test mode (1 item per source)
• --format: Output format (markdown, html, json, detailed)
• -o: Output file path

Step 3: Release Lock and Verify

# Release lock
rm -f "$LOCKFILE"

# Verify output
ls -la ~/.pulsefeed/output/digest_*.md | tail -1

Scheduled Collection (Every 2 Hours)

For automated collection, use this complete workflow:

#!/bin/bash
set -e

PULSEFEED_DIR=~/workspace/pulsefeed
OUTPUT_DIR=~/.pulsefeed/output
LOCKFILE=~/.pulsefeed/pulsefeed.lock
TIMESTAMP=$(date +%Y%m%d_%H%M)

# Ensure directories exist
mkdir -p "$OUTPUT_DIR"
mkdir -p "$(dirname $LOCKFILE)"

# === LOCK MANAGEMENT ===
acquire_lock() {
  if [ -f "$LOCKFILE" ]; then
    PID=$(cat "$LOCKFILE")
    if ps -p "$PID" > /dev/null 2>&1; then
      echo "⚠️  Previous instance still running (PID: $PID). Terminating..."
      kill "$PID" 2>/dev/null || true
      sleep 3
      kill -9 "$PID" 2>/dev/null || true
    fi
    rm -f "$LOCKFILE"
  fi
  echo $$ > "$LOCKFILE"
  echo "🔒 Lock acquired (PID: $$)"
}

release_lock() {
  rm -f "$LOCKFILE"
  echo "🔓 Lock released"
}

# Ensure lock is released on exit
trap release_lock EXIT

# === MAIN ===
acquire_lock

cd "$PULSEFEED_DIR"
python3 -m pulsefeed collect \
  --sources hackernews,github,producthunt \
  --max-items 10 \
  --format detailed \
  -o "$OUTPUT_DIR/digest_$TIMESTAMP.md"

# Report result
OUTPUT_FILE="$OUTPUT_DIR/digest_$TIMESTAMP.md"
if [ -f "$OUTPUT_FILE" ]; then
  LINES=$(wc -l < "$OUTPUT_FILE")
  echo "✅ Digest generated: $OUTPUT_FILE ($LINES lines)"
else
  echo "❌ Collection failed"
  exit 1
fi

Cache Behavior

The cache is used to save LLM costs by not re-processing the same URLs:

• Location: ~/.pulsefeed/cache.db
• TTL: 24 hours (configurable in config.yaml)
• Behavior: URLs processed within TTL are skipped, saving LLM API calls
• Do NOT clear cache for scheduled runs - this is by design

If you see fewer items than expected, it means those URLs were already processed recently. The cache ensures you don't pay for duplicate LLM processing.

Output Format

The detailed format includes:

Hacker News

• Article Summary: Core views, key information, technical details, why it matters
• Comment Analysis: Selected quotes from valuable discussions + overall summary

GitHub Trends

• README Summary: One-line description, core features, technical highlights, use cases, quick start

Product Hunt

• Product Analysis: Positioning, problem solved, core features, target users, unique value

Troubleshooting

"No items collected" or very few items

• This is normal if cache has recent entries
• Check cache TTL in config.yaml
• Wait for TTL to expire, or manually clear cache if needed: rm -f ~/.pulsefeed/cache.db

LLM processing fails

• Verify config.yaml has correct LLM settings
• Check api_type matches your provider (openai vs anthropic format)
• Ensure api_base is set for third-party providers

Lock file stale

• If process crashed, manually remove: rm -f ~/.pulsefeed/pulsefeed.lock

Slow processing

• Reduce --max-items for faster runs
• Use --demo mode for quick tests

Success Criteria

• Lock acquired successfully (no conflicts)
• Collection command completes without errors
• Output file generated with expected sections
• Lock released on completion
• LLM summaries are in Chinese with proper formatting