SharpAgent Intelligence Monitor v1.0.0#

Let your agent scan the frontier for you every day. Multi-track parallel collecting → 3D dynamic scoring → Five-factor trust verification → Structured briefing output. Based on AI Frontier Monitor architecture + SharpAgent five-factor verification + frontier scouting experience.

Contract#

yaml

contract:
  name: sharpagent-intelligence-monitor
  version: "1.0.0"
  category: monitor
  trust_level: verified
  reads:
    - InformationSource
    - FiveFactorResult
  writes:
    - InformationSource
    - CrossValidation
  preconditions:
    - "Access to web_search tool"
    - "Access to curl/jq for API fetching"
  postconditions:
    - "Each info item has a score (0-5)"
    - "Output tiered: core/watching/quick-scan"
    - "Cross-track signals extracted"
  calibration:
    default_mode: professional
    modes_supported: [warm, professional, deep]
  compliance:
    jurisdiction: global
    safety_level: standard
  lifecycle:
    status: active
    publish_as: SharpAgent

Architecture: 5-Track Parallel + Five-Factor Verification#

text

Sources (5 tracks parallel)
    ↓
3D Automatic Scoring (relevance/quality pre-filter)
    ↓
Dynamic Tiers (core / watching / quick-scan)
    ↓
Cross-Track Signal Detection
    ↓
Five-Factor Trust Verification ← SharpAgent differentiator
    ↓
Structured Briefing Output
    ↓
Archive to Ontology

Track 1: 🏢 Enterprise — 11 RSS Feeds#

FeedURLPriorityOpenAI Blogopenai.com/blog⭐⭐⭐⭐⭐Anthropic Bloganthropic.com/blog⭐⭐⭐⭐⭐AWS ML Blogaws.amazon.com/blogs/machine-learning⭐⭐⭐⭐⭐Google AI Blogai.googleblog.com⭐⭐⭐⭐Meta AI Blogai.meta.com/blog⭐⭐⭐⭐Techmemetechmeme.com/feed⭐⭐⭐⭐The Verge AItheverge.com/ai-artificial-intelligence⭐⭐⭐Hacker Newsnews.ycombinator.com⭐⭐⭐Product Huntproducthunt.com⭐⭐Ars Technica AIarstechnica.com/ai⭐⭐Wired AIwired.com/tag/artificial-intelligence⭐⭐

Track 2: 🇨🇳 China — 36kr Hotlist#

bash

curl -s "https://openclaw.36krcdn.com/media/hotlist/{date}/24h_hot_list.json"

Covering: China tech hotspots, AI dynamics, funding, industry trends

Track 3: 📚 Papers — arXiv#

Fetch latest from:

• cs.AI (Artificial Intelligence)
• cs.LG (Machine Learning)
• cs.CL (Computation and Language)

Track 4: 🔥 GitHub Trending (AI/ML)#

Fetch daily trending repos in:

• AI agents
• LLM tools
• ML frameworks

Track 5: 🔍 Web Search Supplement#

Use web_search tool for topics with insufficient coverage.

---

Scoring: 3-Dimensional Dynamic#

Each candidate is scored on 3 dimensions:

DimensionWeightWhat to Look For🏢 Enterprise Landing40%Real deployment, company name, scale, customer evidence📊 Data Support30%Quantified results (%, improvements, benchmarks)💡 Learnability30%Methodology, architecture, lessons learned, patterns

Source Bonuses#

SourceBonusOpenAI / Anthropic / AWS official+1.0Techmeme / peer-reviewed papers+0.5Product Hunt / HN+0.336kr (China relevance)+1.0 for Chinese audience

Dynamic Tiers (based on actual score distribution)#

text

Score Distribution → Dynamic Thresholds
    ↓
🔴 Core: top ~15% or ≥3.5 (max 3)
🟡 Watching: top ~30% or ≥2.5 (max 5)
🟢 Quick Scan: ≥1.0 (max 8)

---

Signal Detection#

Extract cross-track signals into 3 categories:

Signal TypeKeywordsOutput🛠 Tech Trendsnew model, architecture, framework, benchmark, SOTATech radar update🏢 Product Releaseslaunch, GA, open-source, preview, betaRelease tracker💰 Funding/M&Aseries, raised, acquire, investment, valuationMoney map

---

SharpAgent Integration: Five-Factor Secondary Verification#

After the 3D scoring pass, add the SharpAgent five-factor as a secondary trust gate:

text

Article → 3D Score → Five-Factor Verification → Final Tier

Five-factor weights (in intel context):

• 🔗 Source Anchor: 0.30 — Is the source reliable?
• 🧠 Logic Anchor: 0.20 — Is the analysis self-consistent?
• 🌍 Compliance Anchor: 0.15 — Is it compliant?
• 🏳️ Interest Anchor: 0.15 — Marketing bias?
• 🔄 Cross Anchor: 0.20 — Multiple sources confirm?

Final Confidence = score_3d * 0.6 + five_factor_confidence * 0.4

Quality Gates:

• Five-factor < 5 → Excluded from briefing
• Source Anchor < 3 → Discarded
• Interest = confirmed → Manual review required

---

Output Format#

text

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📡 SharpAgent Intelligence Briefing · {Day} {Date}
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📊 Overview
   Sources: {N} tracks
   Candidates: {total} | High quality: {quality}
   🔗 Trust check: passed {pass}/{total}

🔴 Core Intelligence ({N} items)
### 1. {Title}
🔗 {Link}
💡 Takeaway: {One-line insight}
🔗 Trust score: {score}/10

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

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

📚 arXiv Papers (≤3)
**{Title}** — {Authors}
Abstract: {Abstract[:150]} → {Link}

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

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
📊 Today's Signals
🛠 Tech Trends: {signal}
🏢 Product Launches: {signal}
💰 Capital Movements: {signal}

🔍 Five-Factor Trust Analysis
   🔗 Source Anchor: {avg}/10
   🧠 Logic Anchor: {avg}/10
   🌍 Compliance: {pass_rate}%
   🏳️ Interest Conflicts: {conflict_rate}%
   🔄 Cross Anchor: {avg}/10

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
⏰ {HH:MM} | sharpagent-intelligence-monitor v1.0 | SharpAgent

Workflow#

Step 1: Fetch All Tracks#

bash

# Enterprise RSS
python3 scripts/rss-crawler.py

# 36kr
curl -s "https://openclaw.36krcdn.com/media/hotlist/$(date +%Y-%m-%d)/24h_hot_list.json"

# arXiv
bash scripts/arxiv-fetch.sh --category cs.AI --days 7 --max 10

# GitHub Trending
bash scripts/github-trending-fetch.sh --period daily

Step 2: Score Candidates#

Run each candidate through the 3D scoring engine. Source bonuses applied per track.

Step 3: Apply Five-Factor Verification#

Each core-tier candidate gets full five-factor review:

1. 🔗 Is the source reliable?
2. 🧠 Is the analysis internally consistent?
3. 🌍 Is it compliant?
4. 🏳️ Any marketing bias?
5. 🔄 Can we verify it independently?

Watch-tier candidates get a lightweight check (source + logic). Scan-tier candidates skip verification.

Step 4: Compute Final Confidence#

text

final_confidence = score_3d * 0.6 + five_factor_confidence * 0.4

Step 5: Detect Cross-Track Signals#

Compare candidates across all 5 tracks. Same topic in multiple tracks = signal, not just a single item. High signal = high priority.

Step 6: Render & Deliver#

Render in calibration-appropriate mode:

• Warm: Tier labels + confidence indicators only
• Professional: Full briefing with per-item analysis
• Deep: Full briefing + five-factor breakdown per core item

Step 7: Archive#

Save to data/briefings/{YYYY-MM-DD}-briefing.md

---

Edge Cases#

SituationActionRSS emptyRun with remaining tracks, skip RSS sectionarXiv API timeoutSkip papers, log warningGitHub fetch failsSkip trending, log warning36kr 404 (no data)Skip 36kr itemsZero quality items (<2 at ≥2.5)Return NO_REPLYSame company multiple sourcesDeduplicate, keep highest score3 consecutive days <3 core itemsTrigger source reviewFive-factor fails all core itemsReturn "No reliable intel today"

Quality Gates#

CheckWhatFail actionMax 16 items/day3+5+5+3(papers)+3(GitHub)Trim tiersNO_REPLY when <2 quality<2 items at score ≥2.5Return NO_REPLYDedup same entityCross-source same-companyKeep highest scoreFive-factor filterCore items must pass verificationDrop or flag3-day threshold failTrigger reviewReview alert

Integration Points#

Five-Factor Review Skill#

• sharpagent-five-factor-review called per core candidate
• Verification results appended to briefing

Calibration Framework#

• Output mode controlled by calibration settings
• Deep mode includes full five-factor breakdown

Ontology#

• Each briefed item archived as InformationSource
• FiveFactorResult attached as validation

Version History#

• v1.0.0 — Initial release. 5-track intel monitor with five-factor verification.

---

SharpAgent · MIT-0 · 2026-05-11