ai-frontier-monitor

by lynxpurr

Install this Skill

curl -fsSL -o ai-frontier-monitor.zip https://skill.seembox.com/api/v1/skills/ai-frontier-monitor/download && unzip ai-frontier-monitor.zip

Download ai-frontier-monitor.zip

SKILL.md About

---
name: ai-frontier-monitor
description: "AI frontier intelligence briefing — aggregate, score, and deliver structured daily briefings from 5 tracks (RSS enterprise, 36kr hotlist, arXiv papers, GitHub Trending, Anthropic web search). Use when user mentions: AI前沿, 情报汇总, 每日情报, 行业动态, AI动态, 技术趋势, 行业信号, 今天有什么信号, AI动态汇总, frontier monitor, daily briefing AI, signal check."
---

# AI Frontier Intelligence Briefing

> Information aggregation ≠ information overload. Daily briefings are curated, scored, and tiered — not a dump of 50 headlines.

## When to Use

Trigger on (in any language):
- "AI 前沿" / "frontier monitor" / "daily briefing" / "signal check" → full briefing
- "今天有什么信号" / "what signals today" / "quick scan" → quick signal detection
- "arXiv 最新" / "latest papers" / "paper tracker" → arXiv track only
- "GitHub Trending" / "AI 热榜" / "trending AI" → GitHub track only

## Architecture: 5-Track Parallel

| Track | Source | Script | Priority |
|-------|--------|--------|----------|
| 🏢 Enterprise | 11 RSS feeds (OpenAI/AWS/Techmeme/...) | `{baseDir}/scripts/generate-briefing.py --candidates <path>` | ⭐⭐⭐⭐⭐ |
| 🇨🇳 China | 36kr Hotlist API | `curl https://openclaw.36krcdn.com/media/hotlist/{date}/24h_hot_list.json` | ⭐⭐⭐⭐ |
| 📚 Papers | arXiv cs.AI/cs.LG/cs.CL | `{baseDir}/scripts/arxiv-fetch.sh --category cs.AI --days 7 --max 10` | ⭐⭐⭐ |
| 🔥 GitHub | GitHub Trending (AI/ML) | `{baseDir}/scripts/github-trending-fetch.sh --period daily` | ⭐⭐⭐ |
| 🔍 Anthropic | Web search supplement | `web_search` tool | ⭐⭐⭐⭐⭐ |

> For full data source details, read `{baseDir}/references/data-sources.md`

## Workflow

### Step 1: Fetch All Tracks

```bash
# Track 1: RSS (run crawler first)
python3 ~/.openclaw/workspace/research/enterprise/rss-crawler.py

# Track 2-4: Generate briefing (all tracks auto-fetched)
python3 {baseDir}/scripts/generate-briefing.py --mode full
```

Modes: `full` | `quick` | `arxiv` | `github`

### Step 2: Auto-Score & Tier

Each candidate without a score is auto-scored (0-5) by keyword matching across 4 dimensions:

| Dimension | Weight | What to look for |
|-----------|--------|-----------------|
| Enterprise landing | 40% | Real company name, deployment scale |
| Data support | 20% | Quantified metrics (% improvement, $ saved) |
| Learnability | 20% | Methodology, architecture, lessons learned |
| Novelty | 20% | New scene, new product, not old news |

Source bonus: OpenAI/AWS +1.0, Techmeme +0.5, PH/HN +0.3

Tiers are **dynamic** (based on actual score distribution, not hardcoded thresholds):
- 🔴 Core: top ~15% or ≥3.5 (max 3)
- 🟡 Worth watching: top ~30% or ≥2.5 (max 5)
- 🟢 Quick scan: ≥1.0 (max 8, 36kr first)

> For scoring keywords and signal detection rules, read `{baseDir}/references/scoring.md`

### Step 3: Detect Signals

Extract cross-track signals into 3 dimensions:
- 🛠 **Tech trends** — new models, architectures, frameworks, benchmarks
- 🏢 **Product launches** — new releases, open-source, GA announcements
- 💰 **Funding/M&A** — investments, acquisitions, IPOs

### Step 4: Render Briefing

Strict format — emoji headers, tiered sections, signal summary. Total ≤ 16 items across all tiers.

```
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
🤖 AI Frontier Intelligence · {Day} {Date}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📡 Sources: 11 RSS + 36kr + arXiv + GitHub + Anthropic
   Candidates: {N} | High quality: {M} | Thresholds: core≥{X} / watch≥{Y}

## 🔴 Core Intelligence ({N} items)
### 1. {Title}
🔗 {Link}
💡 Insight: {One-line takeaway for the user}

## 🟡 Worth Watching ({N} items)
1. **{Title}**
   🔗 {Link}

## 🟢 Quick Scan ({N} items)
• [{Title}]({Link})

## 📚 arXiv · Paper Tracker (≤3)
**{Title}** — {Authors} | {Date}
Abstract: {Abstract[:150]}... → {Link}

## 🔥 GitHub Trending · AI (≤3)
**{Repo}** ({Lang}) +{TodayStars}⭐ → {Link}

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📊 Today's Signals
🛠 Tech trends: {signal}
🏢 Product launches: {signal}
💰 Funding/M&A: {signal}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⏰ {HH:MM} | ai-frontier-monitor v2.1
```

### Step 5: Deliver

Reply directly in conversation. To push to a specific channel (Feishu, Discord, Telegram, etc.), configure the delivery target in `{baseDir}/references/BRIEFING_CONFIG.md` and use OpenClaw's `message` tool or a cron job with `sessionTarget: "current"` / `"isolated"` to deliver at your preferred schedule.

## Edge Cases

| Situation | Action |
|-----------|--------|
| No candidates (RSS empty) | Run with 36kr + arXiv + GitHub only, skip RSS section |
| arXiv API timeout (>30s) | Skip paper section, log warning |
| GitHub fetch fails | Skip trending section, log warning |
| 36kr API 404 (no data yet) | Skip 36kr items in quick scan |
| Zero high quality items (<2 at ≥2.5) | Return `NO_REPLY` instead of empty briefing |
| Same company appears in multiple sources | Deduplicate, keep highest-scored entry |

## Skill Integration

| Skill | Role |
|-------|------|
| **wechat-curator** | WeChat articles → 🟢 Quick scan supplement |
| **zsxq-helper** | Zsxq content → independent push (not in main briefing) |
| **rss-crawler.py** | RSS fetching engine (11 sources) |

## Configuration

Edit `{baseDir}/references/BRIEFING_CONFIG.md` to customize:
- Quantity limits per tier
- Data source on/off switches
- Signal detection thresholds
- Delivery target (Feishu user ID / Discord channel / etc.)

## Quality Gates

- Max 16 items per day (3+5+5+3 papers)
- `NO_REPLY` when <2 quality candidates
- Deduplicate same company/product, keep highest score
- 3 consecutive days below 3 core items → trigger keyword review

---

_Last updated: 2026-05-02 | v2.1_

_meta.json Metadata

{
  "ownerId": "kn7d5x409z88swt4n3fh3v77yn833x7m",
  "publishedAt": 1777705146439,
  "slug": "ai-frontier-monitor",
  "version": "2.2.1"
}