aegis（aegis 自动化）

AEGIS — Automated Emergency Geopolitical Intelligence System

Civilian-first threat intelligence for conflict zones. Monitors 28+ sources (RSS, web, APIs, OSINT aggregators) with two delivery modes:

1. CRITICAL-only scan (every 15 min) — Posts ONLY for verified, active emergencies (3-phase validation)
2. Morning/evening briefings (8am + 8pm local) — Agent-powered situation reports with actionable guidance

Powered by World Monitor (real-time geopolitical intelligence API) and LiveUAMap (verified OSINT event feed), supplemented by government sources, major news agencies, and aviation data.

Not a panic tool. Calm, factual, action-oriented. Follows official government guidance.

Requirements

• curl — used for all HTTP fetching (RSS, web, APIs). Pre-installed on most systems.
• python3 (3.8+) — runs all scripts. No extra pip packages needed (stdlib only).
• No API keys required for baseline 23 sources. Optional: NewsAPI free tier for expanded coverage.
• LLM (optional) — Improves CRITICAL alert accuracy via semantic verification. See LLM Verification below.

Quick Start

First-time setup (interactive)

python3 scripts/aegis_onboard.py

Creates ~/.openclaw/aegis-config.json with location, language, and alert preferences.

Manual config

Create ~/.openclaw/aegis-config.json:

{
  "location": {
    "country": "AE",
    "country_code": "AE",
    "city": "Dubai",
    "timezone": "Asia/Dubai"
  },
  "language": "en",
  "alerts": { "critical_instant": true, "high_batch_minutes": 30, "medium_digest_hours": 6 },
  "briefings": { "morning": "07:00", "evening": "22:00" },
  "scan_interval_minutes": 15,
  "api_keys": {}
}

Run a scan

python3 scripts/aegis_scanner.py

Set up cron monitoring

# CRITICAL-only scan (every 15 min) — posts ONLY for imminent threats
openclaw cron add --every 15m --message "Run AEGIS scan: python3 <skill-dir>/scripts/aegis_cron.py"

# Morning briefing (adjust time for user timezone — example: 4:00 UTC = 8:00 Dubai)
openclaw cron add --cron "0 4 * * *" --tz UTC --message "Run AEGIS morning briefing: python3 <skill-dir>/scripts/aegis_briefing.py morning"

# Evening briefing (example: 16:00 UTC = 20:00 Dubai)
openclaw cron add --cron "0 16 * * *" --tz UTC --message "Run AEGIS evening briefing: python3 <skill-dir>/scripts/aegis_briefing.py evening"

# Optional: Live feed (every 5 min) — for high-tempo situations only, disable during calm periods
openclaw cron add --cron "*/5 * * * *" --message "Run AEGIS live feed: python3 <skill-dir>/scripts/aegis_feed.py" --disabled

Optional: Telegram channel delivery

Set environment variables for direct channel posting:

• AEGIS_BOT_TOKEN — Telegram bot token (from BotFather)
• AEGIS_CHANNEL_ID — Telegram channel ID

Or add to config:

{
  "telegram": {
    "bot_token": "your-token",
    "channel_id": "-100xxxxxxxxxx"
  }
}

Data Storage

AEGIS stores scan state in ~/.openclaw/aegis-data/ (or $AEGIS_DATA_DIR if set):

• seen_hashes.json — content dedup hashes (48h rolling window for scans, 6h for feed)
• pending_alerts.json — alerts awaiting batch delivery
• scan_log.json — recent scan results
• last_scan.json — most recent scan output (used by briefings)
• feed_state.json — live feed dedup state and last post timestamp
• last_alert_time.json — channel post cooldown tracker
• scan_history.log — rolling log of scan results (last 500 entries)

All files are local JSON. No data is sent to external servers beyond the listed sources in source-registry.json.

Sources (30+)

All sources are defined in references/source-registry.json. The scanner only contacts URLs listed there.

Outbound connections

The scanner contacts only:

1. URLs listed in references/source-registry.json (RSS feeds, news sites, government pages)
2. https://world-monitor.com/api/signal-markers (World Monitor public API)
3. LiveUAMap regional pages (e.g., https://iran.liveuamap.com)
4. Telegram Bot API (https://api.telegram.org/bot.../sendMessage) — only if channel delivery is configured

No telemetry. No analytics. No phone-home.

Alert Classification

CRITICAL is reserved for "act now, your life may be in danger." Regional developments and background context are HIGH.

CRITICAL Validation Pipeline (v3.2)

CRITICAL alerts go through 3-phase validation to eliminate false positives:

1. Regex pre-filter — Tightened patterns match only active attacks, sirens, shelter orders
2. Negative pattern filter — Disqualifies: past tense ("has been hit"), sports, analysis, speculation, historical context
3. LLM verification (optional) — Confirms the item describes an ACTIVE emergency

Additionally:

• Government trust — Tier 0 sources (NCEMA, government) bypass LLM verification
• Source corroboration — Single source: 4h cooldown; 2+ sources: 1h cooldown
• Fail-open safety — If LLM unavailable or disabled, CRITICAL alerts pass through (safety first)

LLM Verification

LLM verification is optional but recommended. It adds a semantic check that catches false positives the regex filters miss (e.g., "cricket match cancelled due to war" triggering CRITICAL).

Three modes (configured in aegis-config.json under "llm"):

Without LLM: AEGIS still works well — the regex pre-filter and negative pattern filter catch most false positives. You may see occasional false CRITICAL alerts that would have been caught by LLM verification.

Config examples:

// Local Ollama (recommended if you have a GPU)
"llm": { "enabled": true, "provider": "ollama", "endpoint": "http://localhost:11434", "model": "qwen3:8b" }

// OpenRouter (cheap cloud API)
"llm": { "enabled": true, "provider": "openai", "endpoint": "https://openrouter.ai/api", "model": "meta-llama/llama-3-8b-instruct", "api_key": "sk-or-..." }

// No LLM (regex-only, no external calls)
"llm": { "enabled": false, "provider": "none" }

The onboarding wizard (aegis_onboard.py) will help you choose during setup.

Anti-Hoax Protocol

• Tier 0 sources: trusted directly (government/emergency)
• Tier 1+ CRITICAL candidates: must pass LLM semantic verification
• Tier 2+: require corroboration from ≥1 Tier 0-1 source for HIGH
• Social media: excluded entirely
• Extraordinary claims: require ≥3 independent sources

Briefings

Morning and evening briefings use agent-powered synthesis to create human-readable situation updates:

Situation Update Format

Each briefing includes:

• Summary — What happened in plain English with real numbers
• Status — Current safety status (threat level with description)
• Actions — 4-6 concrete things a civilian should do right now
• Daily Impact — Flights, schools, work, supplies, roads, hospitals
• Outlook — What to expect in the next 12-24 hours
• Sources — Where to verify and get more info

Delivery Modes

Anti-Spam

• CRITICAL scan: 4-hour cooldown (single source) / 1-hour cooldown (multi-source corroborated)
• Live feed: 5-minute batch interval, max 8 events per post
• Briefings: Agent-powered, pin to channel, once per cycle

Country Profiles

Each supported country has a profile in references/country-profiles/. Profiles contain:

• Emergency contacts and hotlines
• Neighboring countries and regions of interest (for source filtering)
• Local threat keyword patterns
• Shelter and evacuation info

Currently available: UAE (uae.json). To add a country: copy _template.json, fill in details, submit PR.

Preparedness Resources

See references/preparedness/ for:

• go-bag-checklist.md — What to pack for emergency evacuation
• communication-plan.md — Family communication plan
• shelter-guidance.md — Shelter-in-place protocol
• evacuation-guidance.md — Routes, transport, embassy registration

Configuration Reference

See references/config-reference.md for all configuration options.

Cost

• Sources (30+): Free (RSS + web scraping + public APIs)
• LLM verification (optional): Free with Ollama; ~$0.001/check with cloud APIs; or disabled (regex-only)
• Briefings: Free with Copilot or local models; ~$0.03-0.05/day with commercial LLMs
• Optional NewsAPI: Free tier (100 req/day) is sufficient