Hn Filter

Hacker News Filter

Filter top 100 Hacker News posts based on user interests with AI-powered semantic matching.

When to Use This Skill

Use this skill when the user:

• Wants to see filtered/personalized Hacker News content
• Asks about interesting tech news or articles
• Wants a curated HN digest
• Mentions "hacker news", "HN", or tech news filtering

Workflow

Step 1: Get User Interests (Interactive)

Ask the user what topics they're interested in and what they want to exclude:

What topics interest you today? (Press Enter to use saved defaults)

Examples:
- Interested in: AI, systems programming, developer tools, startups
- Exclude: crypto, blockchain, politics

If user presses Enter or says "use defaults", load from config file.

Step 2: Load/Save Config

Config file location: ~/.claude/hacknews-interest.json

Expected format:

{
  "interests": ["AI", "machine learning", "systems programming", "developer tools", "open source"],
  "blacklist": ["crypto", "blockchain", "NFT", "web3", "politics"]
}

If config doesn't exist, create it with user's input. If user provides interests interactively, optionally ask if they want to save as new defaults.

Step 3: Fetch Top 100 HN Stories

Use the Hacker News API:

# Fetch top story IDs
curl -s "https://hacker-news.firebaseio.com/v0/topstories.json" | jq '.[0:100]'

# Fetch individual story details
curl -s "https://hacker-news.firebaseio.com/v0/item/{id}.json"

For each story, extract:

• id: Story ID
• title: Story title
• url: Link to article (may be missing for Ask HN, Show HN)
• score: Upvotes
• descendants: Comment count
• by: Author username

Step 4: Fetch Article Content (with Caching)

For each story with a URL:

1. Check cache first: Look in ~/.cache/hn-filter/articles/{story_id}.txt
2. If not cached, fetch article:
   • Use curl or web fetch to get the article content
   • Extract main text content (strip HTML, ads, navigation)
   • Cache the result for future runs
3. Fallback: If fetching fails (paywall, blocked, etc.), use title + URL domain for analysis

Cache structure:

~/.cache/hn-filter/
├── articles/
│   ├── 12345.txt
│   ├── 12346.txt
│   └── ...
└── comments/
    ├── 12345.json
    └── ...

Step 5: Fetch and Summarize Comments

For each story:

1. Fetch top-level comments (limit to first 10-20 for performance)
2. For each comment, fetch its content
3. Look for substantive discussion points, not meta-commentary

# Story item includes 'kids' array of comment IDs
curl -s "https://hacker-news.firebaseio.com/v0/item/{comment_id}.json"

Step 6: Semantic Filtering

For each story, analyze:

1. Title - Quick keyword + semantic match
2. Article content - Deep semantic analysis of full text
3. URL domain - Source credibility/relevance (github.com, arxiv.org, etc.)

Filtering Logic:

1. First, check blacklist - if ANY blacklist term matches semantically, EXCLUDE the story
2. Then, check interests - if ANY interest matches semantically, INCLUDE the story
3. Blacklist takes priority over interests

Semantic matching means:

• Not just exact keyword match
• Understanding context (e.g., "LLM" matches "AI" interest)
• Understanding negation (article criticizing crypto still matches crypto blacklist)
• Understanding related concepts (e.g., "Rust compiler" matches "systems programming")

Step 7: Generate Output

For each matching story, output:

## [Story Title](URL)
Score: X | Comments: Y

**Why this matched:** [1-line AI explanation of why this matches user interests]

**Key discussion points:**
- [Balanced point 1 - present both perspectives if debated]
- [Balanced point 2]
- [Balanced point 3]

---

Discussion point guidelines:

• Extract substantive insights, not meta-commentary ("great article!")
• When there's debate, present BOTH sides neutrally
• Focus on technical insights, lessons learned, contrarian views
• Ignore flame wars, personal attacks, off-topic tangents
• Limit to 3-5 key points per story

Output Example

# Hacker News Digest - Filtered for: AI, systems programming
Excluding: crypto, blockchain

Found 12 matching stories from top 100

---

## [Anthropic releases Claude 3.5 with improved reasoning](https://example.com/article)
Score: 542 | Comments: 231

**Why this matched:** Directly relevant to AI/ML interests - covers new LLM capabilities and benchmarks.

**Key discussion points:**
- Performance comparison shows 2x improvement on coding tasks vs previous version
- Debate on benchmark validity: some argue real-world performance differs from synthetic tests
- Several commenters report success using it for complex refactoring tasks
- Discussion of API pricing changes and impact on indie developers

---

## [Building a Modern Text Editor in Rust](https://example.com/rust-editor)
Score: 312 | Comments: 89

**Why this matched:** Combines systems programming (Rust) with developer tools (text editor).

**Key discussion points:**
- Author shares performance metrics: <5ms latency for 100k line files
- Debate on using GPU rendering: proponents cite smoothness, critics cite battery drain
- Several suggestions for async architecture patterns for plugin systems

---

Error Handling

• No config file: Create default with empty interests/blacklist, ask user to configure
• API rate limiting: Add delays between requests, inform user of progress
• Article fetch failures: Log which articles couldn't be fetched, continue with title-only analysis
• No matches found: Report this to user, suggest broadening interests or narrowing blacklist

Performance Tips

• Fetch stories in parallel (batch of 10-20 at a time)
• Cache aggressively - articles don't change
• For repeat runs on same day, skip re-fetching if cache is <24 hours old
• Show progress to user: "Fetching stories... 50/100"