| name | env-scanner |
| description | Quintuple Workflow Environmental Scanning System — WF1-1 (source exploration) + WF1 (general multi-source) + WF2 (arXiv academic deep scan) + WF3 (Naver News with FSSF/Tipping Point) + WF4 (Multi&Global-News, 43 direct news sites, multilingual) with integrated report. Use when the user wants to (1) run daily environmental scanning, (2) check scanning workflow status, (3) review or approve scanning results, or (4) manage the signals database. |
| agent | master-orchestrator |
Environmental Scanner Skill
Invocation Policy (Option A‴, 2026-04-16)
Prefer slash commands over Skill tool invocation for WF execution:
- ✅ Use
/env-scan:run, /env-scan:run-arxiv, /env-scan:run-naver, /env-scan:run-multiglobal-news, /env-scan:run-weekly in the main session
- ❌ Avoid
Skill({skill: "env-scan:run"}) from sub-agents — forks into a context without the Agent tool, breaking v3.8.0 Context Isolation Protocol
The agent: master-orchestrator frontmatter below is intentionally retained for Skill-based discovery, but the executable entry for WF delegation flow is the slash command path. See master-orchestrator.md §Dual Execution Path (v3.10.0) for the authoritative invocation protocol.
Description
Quintuple Workflow Environmental Scanning System for detecting weak signals of future changes across STEEPs domains (Social, Technological, Economic, Environmental, Political, spiritual). Operates four completely independent workflows — WF1 (general multi-source), WF2 (arXiv academic deep scan), WF3 (Naver News with FSSF classification, Three Horizons tagging, and Tipping Point detection), and WF4 (Multi&Global-News: direct crawling of 43 global news sites across 11 languages with FSSF/Three Horizons/Tipping Point) — each producing independently complete reports, which are then merged into a unified integrated report.
Purpose
Enable systematic daily scanning of global information sources to catch early signals of trends, paradigm shifts, and emerging futures "AS FAST AS POSSIBLE" across Korea, Asia, Europe, Africa, and Americas — from mainstream sources, academic research, and Korean media.
Architecture
Quintuple Workflow System
┌─────────────────────────────────────────────────────┐
│ Master Orchestrator │
│ SOT: workflow-registry.yaml (81 checks) │
├─────────────────────────────────────────────────────┤
│ WF1-1: Source Exploration (v4.0.0) │
│ Purpose: Discover, evaluate, promote new sources │
│ Python-enforced determinism: 8 modules (gap, score, simulate) │
│ Runs FIRST (execution_order: 0, gate M0) │
│ Checkpoints: 2 (Step 2.7, Step 3.6) │
├─────────────────────────────────────────────────────┤
│ WF1: General Environmental Scanning │
│ Sources: 20+ (patents, policy, blogs, etc.) │
│ arXiv: EXCLUDED │
│ Checkpoints: 2 (Step 2.5, Step 3.4) │
├─────────────────────────────────────────────────────┤
│ WF2: arXiv Academic Deep Scanning │
│ Sources: arXiv ONLY (30+ categories) │
│ Parameters: 14 days, 50/category │
│ Checkpoints: 2 (Step 2.5, Step 3.4) │
├─────────────────────────────────────────────────────┤
│ WF3: Naver News Environmental Scanning │
│ Sources: NaverNews ONLY (6 sections) │
│ FSSF 8-type + Three Horizons + Tipping Point │
│ Checkpoints: 2 (Step 2.5, Step 3.4) │
├─────────────────────────────────────────────────────┤
│ WF4: Multi&Global-News Environmental Scanning │
│ Sources: 43 direct news sites (11 languages) │
│ FSSF + Three Horizons + Tipping Point + Multilingual│
│ Checkpoints: 2 (Step 2.5, Step 3.4) │
├─────────────────────────────────────────────────────┤
│ Integration: Report Merger (5 inputs) │
│ Merge: pSST unified ranking (top 20 from WF1-4) │
│ WF1-1: separate Source Exploration Summary section │
│ Cross-workflow analysis (WF1↔WF2↔WF3↔WF4) │
│ Checkpoint: 1 (final approval) │
└─────────────────────────────────────────────────────┘
Independence: WF1-1, WF1, WF2, WF3, and WF4 are completely independent — separate data directories, separate signal databases, no runtime data sharing. Any can be disabled or run standalone without affecting the others.
Source of Truth: env-scanning/config/workflow-registry.yaml — single authoritative definition validated at startup.
Commands
This skill provides the following slash commands:
/run-daily-scan - Execute full quintuple scan (WF1-1 + WF1 + WF2 + WF3 + WF4 + Integration)/run-daily-scan --base-only - WF1 base sources only + WF2 + Integration/run-daily-scan --arxiv-only - WF2 only (standalone arXiv scan)/run-arxiv-scan - WF2 standalone (alias for --arxiv-only)/run-naver-scan - WF3 standalone (Naver News with FSSF/Tipping Point)/run-multiglobal-news-scan - WF4 standalone (Multi&Global-News, 43 sites, 11 languages)/run-weekly-scan - Execute weekly meta-analysis (no new scanning, analyzes last 7 days)/status - Check current workflow progress/review-filter - Review duplicate filtering results (Step 1.4)/review-analysis - Review analysis results and adjust priorities (Step 2.5)/approve - Approve report (Step 3.4 / Integration)/revision "feedback" - Request report revision with specific feedback/trigger-delphi - Manually activate expert panel validation (Phase 1.5)/generate-scenarios - Manually activate scenario builder (Step 7.5)
Workflow Overview
Execution Flow: WF1-1 → WF1 → WF2 → WF3 → WF4 → Integration
WF1-1 (Source Exploration) runs first to discover and promote new sources before WF1 scans them. Each signal-scanning workflow (WF1, WF2, WF3, WF4) follows the same 3-phase structure:
Phase 1: Research (Information Collection)
- Load historical signals database (each workflow has its own)
- Scan sources (WF1: multi-source marathon / WF2: arXiv deep scan / WF3: Naver News crawl / WF4: 43 global news sites)
- Filter duplicates using 4-stage cascade (URL → String → Semantic → Entity)
- (Optional) Expert panel validation if >50 new signals
Phase 2: Planning (Analysis & Structuring)
- Classify signals into STEEPs categories
- Analyze impacts using Probabilistic Cross-Impact Matrix
- Rank by priority (Impact 40%, Probability 30%, Urgency 20%, Novelty 10%)
- (Optional) Generate plausible scenarios
Phase 3: Implementation (Report Generation)
- Update signals database with backups
- Generate comprehensive Korean report (skeleton-fill method)
- Archive report and send notifications
- Self-Improvement Analysis (independent per workflow)
Usage
Basic Usage
/run-daily-scan
/run-daily-scan --base-only
/run-arxiv-scan
/status
/review-filter
/review-analysis
/approve
Execution Modes
Full Quintuple Scan (default): WF1-1 exploration → WF1 marathon → WF2 arXiv deep → WF3 Naver News → WF4 Multi&Global-News → Integrated report
- 11 human checkpoints, top 20 signals in integrated report + Source Exploration Summary
- WF1-1 discovers & evaluates new sources; WF1 scans 20+ sources; WF2 scans arXiv 30+ categories; WF3 crawls Naver 6 sections; WF4 crawls 43 global news sites across 11 languages
- Autopilot mode (v3.8.0): Each WF is invoked as a context-isolated sub-agent to prevent context exhaustion. Checkpoints are self-approved.
context_isolation_manager.py (Python-enforced determinism) generates deterministic sub-agent prompts and verifies results. Manual mode uses inline execution (pre-v3.8.0 behavior).
arXiv Standalone: WF2 only, produces independent arXiv academic report
- 2 human checkpoints, no integration
- Use when you need academic-focused signals only
Naver Standalone: WF3 only, produces independent Naver report with FSSF/Tipping Point
- 2 human checkpoints, no integration
- Use when you need Korean mainstream media signals only
Multi&Global-News Standalone: WF4 only, produces independent Multi&Global-News report with FSSF/Three Horizons/Tipping Point
- 2 human checkpoints, no integration
- Use when you need multilingual global mainstream media signals only
Advanced Usage
/trigger-delphi
/generate-scenarios
/revision "Add more detailed analysis for the top 5 signals"
Configuration
Source of Truth
env-scanning/config/workflow-registry.yaml - SOT — single authoritative definition of the quadruple workflow system
Shared Configs (read-only reference)
env-scanning/config/domains.yaml - STEEPs domain definitions and keywordsenv-scanning/config/thresholds.yaml - Filtering and scoring thresholdsenv-scanning/config/core-invariants.yaml - Immutable workflow boundaries (SIE safety)env-scanning/config/self-improvement-config.yaml - Self-Improvement Engine settings
Workflow-Specific Configs
env-scanning/config/sources.yaml - WF1 data sources (arXiv disabled)env-scanning/config/sources-arxiv.yaml - WF2 data sources (arXiv only, extended)env-scanning/config/sources-naver.yaml - WF3 data sources (NaverNews only, crawl)env-scanning/config/sources-multiglobal-news.yaml - WF4 data sources (43 global news sites, 11 languages, direct crawl)
Output Files
WF1 (General) — env-scanning/wf1-general/
- Report:
reports/daily/environmental-scan-{date}.md
- Database:
signals/database.json
- Archive:
reports/archive/{year}/{month}/
WF2 (arXiv) — env-scanning/wf2-arxiv/
- Report:
reports/daily/environmental-scan-{date}.md
- Database:
signals/database.json
- Archive:
reports/archive/{year}/{month}/
WF3 (Naver News) — env-scanning/wf3-naver/
- Report:
reports/daily/environmental-scan-{date}.md
- Database:
signals/database.json
- Archive:
reports/archive/{year}/{month}/
- Alerts:
logs/alerts-{date}.json
- Tipping Points:
analysis/tipping-point-indicators-{date}.json
WF4 (Multi&Global-News) — env-scanning/wf4-multiglobal-news/
- Report:
reports/daily/environmental-scan-{date}.md
- Database:
signals/database.json
- Archive:
reports/archive/{year}/{month}/
- Alerts:
logs/alerts-{date}.json
- Tipping Points:
analysis/tipping-point-indicators-{date}.json
Integrated — env-scanning/integrated/
- Report:
reports/daily/integrated-scan-{date}.md
- Timeline Map:
reports/daily/timeline-map-{date}.md (enhanced signal evolution analysis)
- Timeline Summary:
reports/daily/timeline-summary-{date}.txt (embedded in integrated report §7.6)
- Archive:
reports/archive/{year}/{month}/
Self-Improvement (per workflow)
- WF1:
env-scanning/wf1-general/self-improvement/
- WF2: managed within WF2 data root
- WF3: managed within WF3 data root
- WF4: managed within WF4 data root
Self-Improvement Engine (Step 3.6)
After quality metrics are generated, the system automatically analyzes performance and tunes parameters:
- MINOR changes (threshold ±10%, timeout adjustments): Auto-applied safely
- MAJOR changes (behavioral changes): Proposed for user approval
- CRITICAL changes (core invariant violations): Always blocked
Safety boundaries are defined in config/core-invariants.yaml. Workflow phases, human checkpoints, STEEPs categories, and the VEV protocol can never be modified by the engine.
Analysis covers 5 areas: threshold tuning, agent performance, classification quality, workflow efficiency, and hallucination tracking. SIE failure never halts the main workflow.
Quality Standards
Performance Targets
- Duplicate detection accuracy: > 95%
- Processing time reduction: 30% vs baseline
- Signal detection speed: 2x vs manual
- Expert feedback time: < 3 days (if Phase 1.5 activated)
Quality Metrics
- Precision and Recall for deduplication
- Classification accuracy for STEEPs
- Human-AI agreement scores
- Report completeness checklist
Task Verification Protocol (v2.2.0)
Every workflow step is governed by the VEV (Verify-Execute-Verify) protocol:
- Pre-Verification: Precondition checks before each step execution
- Post-Verification (3-Layer):
- Layer 1 (Structural): File existence, JSON validity, schema compliance
- Layer 2 (Functional): Data integrity, count accuracy, range validation
- Layer 3 (Quality): Accuracy targets, completeness, consistency
- Retry Protocol: Automatic re-execution on Layer 1/2 failure (max 2 retries)
- Pipeline Gates: Data continuity checks between Phase 1→2, Phase 2→3, Phase 3 completion
Verification results are recorded in logs/verification-report-{date}.json and summarized in quality metrics.
See orchestrator protocol (.claude/agents/protocols/orchestrator-protocol.md) for full specification.
STEEPs Framework (Core Classification)
6 Categories (IMMUTABLE):
- S - Social (demographics, education, labor - NO spiritual)
- T - Technological (innovation, digital transformation)
- E - Economic (markets, finance, trade)
- E - Environmental (climate, sustainability, resources)
- P - Political (policy, law, regulation, institutions, geopolitics)
- s - spiritual (ethics, psychology, values, meaning, AI ethics)
Example Workflow Execution
/run-daily-scan
/review-analysis
/approve
Troubleshooting
Common Issues
Issue: Low filter rate (< 30%)
Solution: Check if sources are returning old content. Verify date filters.
Issue: Too many signals (> 100)
Solution: Adjust keyword specificity in config/domains.yaml
Issue: Classification errors
Solution: Review STEEPs definitions in config/domains.yaml, adjust LLM prompt
Issue: Database corruption
Solution: Restore from signals/snapshots/database-{recent-date}.json
Dependencies
Required
- Claude Code CLI (version >= 1.0)
- Python 3.9+ (for utility scripts)
- API keys for data sources (Google Scholar, etc.)
Optional
- SBERT model for semantic similarity
- Expert panel contact list for Phase 1.5
- Slack/Email notification setup
References
See references/ directory for:
steep-framework.md - Detailed STEEPs definitionssignal-template.md - Standard signal structurereport-format.md - Report generation guidelinesreport-skeleton.md - Individual workflow report skeleton (10 signals)integrated-report-skeleton.md - Integrated report skeleton (20 signals)
Version
- Skill Version: 4.0.0 (Quintuple Workflow System)
- Compatible with: Claude Code v1.0+
- System Version: 4.0.0 (Quintuple Workflow + SOT + Integration)
- Sources Version: 6.0.0 (arXiv → WF2, NaverNews → WF3, Multi&Global-News → WF4)
- Protocol Version: 2.2.0 (VEV + Pipeline Gates)
- Last Updated: 2026-02-24
Support
For issues or questions:
- Check logs in
env-scanning/logs/
- Review quality metrics in
logs/quality-metrics/
- Consult orchestrator state in
logs/workflow-status.json
