last30days-Skill Getting Started Guide: The All-in-One Intelligence Collector for Reddit/HN/Polymarket

Getting Started | 15 Minutes | Master Cross-Platform Intelligence Gathering, Double Your Research Efficiency

GitHub: mvanhorn/last30days-skill

---

Target Audience Profile

• Developers with 1-5 years of experience.
• Users who perform daily AI tool selection, competitive research, or technical trend tracking.
• Users with command-line basics, familiar with Claude Code or Codex CLI.
• Those wishing to escape the fragmented process of "scrolling Reddit/HN for info."

---

Core Dependencies & Environment

---

Project Structure Tree

last30days-skill/
├── SKILL.md                    # Skill definition (deployed to ~/.claude/skills/)
├── SPEC.md                     # Full technical specification
├── CLAUDE.md                   # Claude Code development guidelines
├── scripts/
│   ├── last30days.py           # Python entry point (research engine)
│   ├── sync.sh                 # Deployment sync script
│   └── lib/
│       ├── __init__.py         # Package entry (eager imports prohibited)
│       ├── env.py              # Environment variable loader
│       ├── dates.py            # Date range and confidence calculations
│       ├── cache.py            # 24h TTL cache
│       ├── http.py             # Standard library HTTP client
│       ├── models.py           # OpenAI/xAI model auto-selection
│       ├── openai_reddit.py    # Reddit search (OpenAI Responses API)
│       ├── xai_x.py            # X search (xAI Responses API)
│       ├── reddit_enrich.py    # Deep metric pulling for Reddit posts
│       ├── hackernews.py       # Hacker News (Algolia Free API)
│       ├── polymarket.py       # Polymarket prediction market (Gamma API)
│       ├── bluesky.py          # Bluesky/AT Protocol search
│       ├── truthsocial.py      # Truth Social search
│       ├── normalize.py        # Raw response → Normalized Schema
│       ├── score.py            # Multi-signal scoring model
│       ├── dedupe.py           # Near-duplicate detection
│       ├── render.py           # Markdown / JSON report rendering
│       └── schema.py           # Type definitions and validation
├── skills/last30days/
│   └── last30days.sh            # Shell wrapper (Claude Code Skill entry)
└── fixtures/                   # Test fixture data

---

Step 1 Installation & Auth Configuration

1.1 Install via Claude Code Plugin (Recommended)

If you are already using Claude Code, simply use the plugin command:

/plugin marketplace add mvanhorn/last30days-skill
/plugin install last30days@last30days-skill

Alternatively, use the official ClawHub tool:

clawhub install last30days-official

1.2 Manual Git Clone Installation

Don't want to use the plugin? Clone directly to your local machine:

# Clone to Claude Code skills directory
git clone https://github.com/mvanhorn/last30days-skill.git \
  ~/.claude/skills/last30days

# Enter directory and view files
cd ~/.claude/skills/last30days
ls -la scripts/

1.3 Configure ScrapeCreators API Key (Required)

This is the unified entry point for Reddit, TikTok, and Instagram. One key covers everything:

1. Visit scrapecreators.com to register and get an API Key.
2. Create the configuration file:

mkdir -p ~/.config/last30days

cat > ~/.config/last30days/.env << 'EOF'
# Required: 3-in-1 Key for Reddit + TikTok + Instagram
SCRAPECREATORS_API_KEY=sc_xxxxxxxxxxxxxxxxxxxx

# Optional: OpenAI API (can be omitted if logged into Codex)
OPENAI_API_KEY=sk-xxxxxxxxxxxxxxxxxxxx
EOF

chmod 600 ~/.config/last30days/.env  # Protect sensitive files

1.4 Optional Auth for X / Bluesky

X Search (Recommended Method):

# 1. Log in to x.com, open browser DevTools (F12)
# 2. Application → Cookies → Copy the Values for auth_token and ct0
# 3. Write to .env

cat >> ~/.config/last30days/.env << 'EOF'

# X Search Auth (Cookie method)
AUTH_TOKEN=xxxxxxxxxxxxxxxxxxxx
CT0=xxxxxxxxxxxxxxxxxxxx

# X Alternative: xAI API (No cookies required)
XAI_API_KEY=xai-xxxxxxxxxxxxxxxxxxxx
EOF

Bluesky Search:

# 1. Go to bsky.app/settings/app-passwords and create an App Password
# 2. Write to .env

cat >> ~/.config/last30days/.env << 'EOF'

# Bluesky/AT Protocol
BSKY_HANDLE=yourhandle.bsky.social
BSKY_APP_PASSWORD=xxxx-xxxx-xxxx-xxxx
EOF

1.5 Verify Successful Installation

# Run a simple test (--mock uses local test data, saves API quota)
python3 ~/.claude/skills/last30days/scripts/last30days.py "Claude Code tips" --mock --emit=compact

If you see output like this, everything is working correctly:

=== last30days Report: Claude Code tips ===
Sources: reddit, hackernews | Time: 2026-03-25 | Mode: mock
[results...]

---

Step 2 Basic Usage

2.1 /last30days Command Line

In Claude Code, type the command directly:

/last30days best Claude Code prompts

Using Codex CLI:

python3 ~/.claude/skills/last30days/scripts/last30days.py "best Claude Code prompts" --emit=compact

2.2 Understanding the Report Structure

After execution, you will receive a structured report looking like this:

# last30days Report: best Claude Code prompts

## Sources Searched (6 active)
reddit | x | hackernews | polymarket | youtube | reddit_threads

## Top Findings
...

## Best Practices (Community-verified methods)
...

## Prompt Pack (Ready to copy)
...

## Recent Developments
...

## References (With source links)
...

The report automatically performs:

• Convergence Detection: Content mentioned on multiple platforms receives higher weight.
• Deduplication: Only the best version of semantically similar posts is kept.
• Recency Decay: Newer content is ranked higher.
• Citations: Every conclusion is accompanied by original links.

2.3 --quick Mode

In a hurry? Add --quick to skip some deep search steps:

python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "Cursor AI vs Windsurf" \
  --quick \
  --emit=compact

2.4 --days=N Custom Time Window

You aren't limited to 30 days; you can customize the range:

python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "OpenClaw latest news" \
  --days=7 \
  --emit=compact

2.5 --refresh Force Cache Refresh

Results are cached for 24 hours by default. To search again:

python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "AI code editors comparison" \
  --refresh \
  --emit=compact

---

Step 3 Multi-Source Parallel Search Mechanism

3.1 How Reddit Search Works

Reddit is the most important signal source. The skill uses a two-step process:

# scripts/lib/openai_reddit.py (Simplified)
def search_reddit(query: str, days: int = 30) -> list[dict]:
    # Step 1: Preliminary discovery via OpenAI Responses API + web_search
    response = openai.responses.create(
        model="gpt-4o",
        input=f"Find active Reddit discussions about: {query}",
        tools=[{"type": "web_search"}],
        max_tokens=2000
    )
    # Step 2: Pull raw post JSON to get actual upvotes/comments count
    enriched = reddit_enrich.fetch_threads(response.urls, days)
    return enriched

3.2 Hacker News Integration

Free access, no API Key required:

# scripts/lib/hackernews.py
def search_hackernews(query: str, days: int = 30) -> list[dict]:
    # HN Algolia API is open and free
    url = "https://hn.algolia.com/api/v1/search"
    params = {
        "query": query,
        "tags": "story",
        "numericFilters": f"created_at_i>{cutoff_timestamp(days)}"
    }
    resp = requests.get(url, params=params)
    items = resp.json()["hits"]
    # Ranked by combined points + num_comments
    return sorted(items, key=lambda x: x["points"] + x["num_comments"] * 2, reverse=True)

3.3 Polymarket Prediction Market Data

This is a unique highlight of last30days—it doesn't just look at "what people say," but also at "how much money people are betting":

# scripts/lib/polymarket.py
def search_polymarket(query: str) -> list[dict]:
    # Gamma API free access
    url = "https://gamma-api.polymarket.com/markets"
    resp = requests.get(url, params={"topic": query, "limit": 20})
    markets = resp.json()

    # Five-factor scoring model
    for m in markets:
        m["composite_score"] = (
            m["text_relevance"] * 0.30 +
            m["volume_24h"] * 0.30 +
            m["liquidity"] * 0.15 +
            m["price_velocity"] * 0.15 +
            m["outcome_competitiveness"] * 0.10
        )
    return sorted(markets, key=lambda x: x["composite_score"], reverse=True)

Actual effect: If you search for "OpenClaw," you'll see not only Reddit threads but also Polymarket data regarding "OpenClaw Monthly Active Users." The odds backed by real money are often more convincing than any forum post.

3.4 Multi-Signal Scoring Model Analysis

Results from all sources flow into a unified scoring pipeline:

# scripts/lib/score.py
def compute_composite_score(item: dict, query: str, days: int) -> float:
    # 1. Text Relevance (Bidirectional similarity + synonym expansion)
    text_score = bidirectional_similarity(item["text"], query)

    # 2. Engagement Normalization
    engagement_score = normalize_velocity(item["engagement"])

    # 3. Source Authority Weighting (HN > Reddit > X > TikTok)
    authority_score = SOURCE_WEIGHTS[item["source"]]

    # 4. Cross-platform Convergence Bonus (Multiple platforms = Weight bonus)
    convergence_score = detect_convergence(item, all_results)

    # 5. Recency Decay (0.98 exponential decay per day)
    recency_score = 0.98 ** days_since_post(item["timestamp"])

    return (
        text_score * 0.35 +
        engagement_score * 0.25 +
        authority_score * 0.20 +
        convergence_score * 0.10 +
        recency_score * 0.10
    )

In v2.5 blind tests, scoring accuracy improved from 3.73/5.0 to 4.38/5.0 (~17% improvement).

---

Step 4 Comparison Mode (X vs Y)

4.1 "Cursor vs Windsurf" Case Study

One of the coolest features—comparing two tools directly:

/last30days cursor vs windsurf

Or via command line:

python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "cursor vs windsurf" \
  --emit=compact

4.2 Parallel Three-Way Research Mechanism

Comparison mode triggers three independent research threads:

# scripts/lib/score.py (Comparison mode branch)
def run_comparative_mode(query_a: str, query_b: str, base_query: str):
    # Three-way parallel search
    results_a = run_research(query_a)       # Path 1: Search Cursor only
    results_b = run_research(query_b)       # Path 2: Search Windsurf only
    results_base = run_research(base_query) # Path 3: Search comparative discussions

    # Generate side-by-side report
    return render_comparison(results_a, results_b, results_base)

4.3 Understanding the Comparison Report

The output format looks roughly like this:

## Comparative Analysis: Cursor vs Windsurf

### Strengths

| Dimension | Cursor | Windsurf |
|---|---|---|
| Completion Speed | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ |
| Context Understanding| ⭐⭐⭐⭐ | ⭐⭐⭐⭐⭐ |
| Multi-file Editing | ⭐⭐⭐⭐ | ⭐⭐⭐ |
| Ecosystem | ⭐⭐⭐⭐⭐ | ⭐⭐⭐ |

### Weaknesses

| Dimension | Cursor | Windsurf |
|---|---|---|
| Memory Usage | High | Medium |
| Offline Support | Poor | Good |

### Community Sentiment (30-day)
- Cursor: 78% positive (1,240 discussions)
- Windsurf: 65% positive (890 discussions)

### Data-Driven Verdict
Cursor leads in community engagement and code completion quality.
Windsurf excels in contextual understanding for complex refactors.
**Recommendation**: Use Cursor for daily coding; Windsurf for architecture planning.

---

Step 5 Embedding into Other Skills / CI Pipelines

5.1 Injection as Context

Other skills can directly reference research results from last30days:

## Recent Research Context
!python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "your research topic" \
  --emit=context

5.2 Reading Context from Files

## Research Context
!cat ~/.local/share/last30days/out/last30days.context.md

5.3 JSON Output for CI/CD

Inject research results into automated pipelines:

# Output in JSON format for program consumption
python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "best LLM for code review 2026" \
  --emit=json > research.json

# Check recommended Top 3 models
python3 -c "
import json
data = json.load(open('research.json'))
for r in data['top_results'][:3]:
    print(f\"- {r['title']} (score: {r['score']:.2f})\")
"

5.4 Project-Level .env Overrides

Don't want global config? Placing a .claude/last30days.env in the project root will override global settings:

mkdir -p .claude

cat > .claude/last30days.env << 'EOF'
# API Key valid only for this project
SCRAPECREATORS_API_KEY=sc_project_specific_key
OPENAI_API_KEY=sk-project-specific-key
EOF

5.5 Session Startup Auto-Validation

v2.9.5 added config checks at session start. Claude Code automatically verifies if your .env config is complete every time it starts:

# Trigger validation manually
python3 ~/.claude/skills/last30days/scripts/last30days.py \
  --validate-config

Clear errors appear if keys are missing:

❌ SCRAPECREATORS_API_KEY is missing (required)
✅ OPENAI_API_KEY found
⚠️  BSKY_HANDLE is missing (optional)

---

Troubleshooting

Q1: Error reporting SCRAPECREATORS_API_KEY missing

Cause: Environment variables were not loaded correctly.

Solution:

# Verify key exists
cat ~/.config/last30days/.env | grep SCRAPECREATORS_API_KEY

# If installed via plugin, try manual sync
bash ~/.claude/skills/last30days/scripts/sync.sh

Q2: X search always returns empty results

Cause: AUTH_TOKEN / CT0 cookies have expired or are invalid.

Solution:

# 1. Log in to x.com again
# 2. Copy the latest auth_token and ct0
# 3. Update .env

# Or use xAI API alternative (no cookies needed)
echo "XAI_API_KEY=xai-your-key" >> ~/.config/last30days/.env

Q3: Search is slow (over 10 minutes)

Cause: Defaults to up to 10 parallel sources; niche topics require more API retries.

Solution:

# 1. Use --quick to skip some deep search logic
python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "niche topic" \
  --quick

# 2. Check if specific sources are timing out
python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "topic" \
  --sources=reddit  # Search Reddit only for faster debugging

Q4: Polymarket returns irrelevant markets

Cause: Polymarket matching is keyword-based; some topics may not have active markets.

Solution:

# Manually specify keyword expansion
python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "AI coding assistant" \
  --polymarket-tags="artificial intelligence,llm,gpt" \
  --emit=compact

Q5: Claude Code shows Permission denied on scripts

Cause: Shell scripts lack execution permissions.

Solution:

chmod +x ~/.claude/skills/last30days/skills/last30days.sh
chmod +x ~/.claude/skills/last30days/scripts/last30days.py

Q6: Few Reddit posts in report, but lots of discussion exists

Cause: Post-v2.9 defaults to ScrapeCreators; Key is configured but format may be wrong.

Solution:

# Verify if ScrapeCreators Key is valid
curl -H "x-api-key: sc_your_key" \
  https://api.scrapecreators.com/v1/reddit/search?q=test

# If Key is valid but no data, check if it reverted to OpenAI
# Force ScrapeCreators in .env
echo "FORCE_SCRAPECREATORS=1" >> ~/.config/last30days/.env

---

Further Reading / Advanced Directions

1. Fine-tuning the Scoring Model

Weights in scripts/lib/score.py are hardcoded. If you have specific source preferences, you can fork and modify them:

# Increase Hacker News authority weight from 0.20 to 0.35
SOURCE_WEIGHTS = {
    "hackernews": 0.35,   # ↑ Changed from 0.20
    "reddit": 0.25,
    "x": 0.15,
    "polymarket": 0.15,
    "youtube": 0.10,
}

2. Adding Custom Sources

Want to integrate GitHub Issues or LinkedIn? Follow the interface spec in hackernews.py, create a new module under scripts/lib/, and register it in last30days.py. SPEC.md contains full interface definitions.

3. Automated Scheduled Research

Use cron to run periodic research and update your knowledge base:

# Run "AI tools weekly" research every morning at 8 AM
0 8 * * * python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "AI developer tools weekly" \
  --days=7 \
  --emit=md \
  --output ~/Documents/Last30Days/ai-tools-weekly-$(date +\%Y-\%m-\%d).md

4. Knowledge Base Integration

Inject research results directly into your personal knowledge base (RAG pipeline):

# Generate context with citations to feed another skill
python3 ~/.claude/skills/last30days/scripts/last30days.py \
  "Claude Code advanced techniques" \
  --emit=context > ~/.knowledge/last30_context.md

---

Resources

• GitHub: mvanhorn/last30days-skill
• ClawHub: last30days-official
• ScrapeCreators API: scrapecreators.com
• OpenClaw Official: openclaw.ai
• Hacker News Algolia API: hn.algolia.com/api
• Polymarket Gamma API: gamma-api.polymarket.com