Geo Event Router

Analyzes, classifies, scores, and routes geostrategic news events using multi-factor scoring and optional LLM semantic analysis for push decision-making.

Security Scan

The skill's code and instructions mostly match its stated purpose, but there are unexplained gaps (missing config, unclear push implementation) and undeclared external dependencies that could require credentials or perform network I/O.

Purpose & Capability

The name/description match the provided code: functions for detecting event type, fingerprinting, multi-factor scoring, and optional LLM analysis are present. However SKILL.md references a local config.py and a push/dispatch behavior (push decisions) but no config.py is included and there is no visible push transport or delivery implementation in the provided file fragment. The SKILL.md lists dependencies (llm_news_analyzer, geo_market_impact_mapper) that are plausible for the purpose, but their provenance is unknown.

Instruction Scope

SKILL.md provides a narrow Python-API usage guide that stays within the stated domain (detect, score, fingerprint, decide). It does not instruct the agent to read unrelated system files or environment variables. Concern: SKILL.md and the code indicate optional LLM analysis and external market mapper usage but do not explain what data flows to those modules or whether they perform network calls; the skill's instructions do not declare that external services or credentials may be required.

Install Mechanism

No install spec (instruction-only with a single Python module). That minimizes install-time risk because nothing is downloaded or installed automatically by the skill bundle itself.

Credentials

The skill declares no required environment variables or credentials, yet depends (optionally) on external packages that commonly require API keys (LLM analyzers or market-data mappers). Those credentials are not declared or explained. This is a proportionality concern because the code could end up calling external services without the SKILL.md listing required env/config or warning about data sent to third parties.

Persistence & Privilege

The skill is not always-enabled and uses default autonomous invocation settings. It does not request persistent system privileges or declare changes to other skills' configurations in the provided material.

What to consider before installing

This skill appears to implement the event-detection and scoring it claims to do, but there are gaps you should clear up before installing or running it with real data: - Missing files: SKILL.md references a config.py and mentions push/dispatch behavior, but config.py is not included and no delivery/push implementation is visible. Ask the author for the missing config and the code path that actually sends/dispatches pushes. - External dependencies: The skill optionally imports llm_news_analyzer and geo_market_impact_mapper. Those modules may perform network calls or require API keys. The skill does not declare required environment variables or where data is sent. Confirm the provenance of those packages and whether they transmit news text to remote services (and whether they require credentials). - Data exfiltration risk: Even though this bundle itself has no declared network calls, the optional LLM analyzer could send news content to a third-party LLM service. Do not run it on sensitive data until you inspect or vendor-audit those dependencies. - Test safely: Run the code in a sandboxed environment with representative but non-sensitive input. Enable logging/inspection (or monkeypatch analyze_semantic) to observe outbound network activity before allowing it access to production data. If the publisher can provide the missing config.py, a clear list of required environment variables (if any), and the source/license/provenance of the optional dependencies, the remaining concerns would be largely addressable.

宏观地缘事件路由与评分技能。负责新闻事件的分析、分类、评分和推送决策。

功能

• 事件检测: 识别军事、地缘政治、央行政策、商品价格、市场异动等事件类型
• 多因子评分: 基于事件类型、来源权威度、市场联动、噪声检测进行综合评分
• 事件指纹: 生成多维度事件指纹（区域 | 主题 | 资产 | 行为），支持事件演化追踪
• LLM 语义分析: 对高价值候选新闻进行深度语义分析（可选）
• 推送决策: 根据评分阈值和冷却期策略决定是否推送

事件类型

类型	基础分	关键词示例
military	35	导弹、空袭、军事行动、战争、袭击
central_bank	30	美联储、央行、加息、降息、利率决议
geopolitics	28	制裁、外交、霍尔木兹、中东、俄乌
commodity	25	原油、黄金、大宗商品、油价、金价
market	25	熔断、暴涨、暴跌、跳水、崩盘

评分机制

总分 = 基础分 + 置信度加分 + 市场映射加分 + LLM 加分 - 降噪扣分

因子	说明	分值
基础分	事件类型决定	10-35
置信度	权威来源（财联社）	+10
市场映射	原油/黄金异动联动	+8~15
LLM 分析	高紧急度判定	+5~15
降噪	不确定词汇（传闻、或、可能）	-20

推送阈值

• ≥70 分: 高优先级，立即推送
• 50-69 分: 观察池，等待确认
• <50 分: 忽略

使用方法

Python API

from geo_event_router import (
    detect_event_type,
    score_news_item,
    generate_event_fingerprint,
    should_push_event,
    Event,
    EventScore,
    NewsItem,
)

# 创建新闻对象
news = NewsItem(
    title="伊朗总统：伊朗将继续进行正当防御",
    content="伊朗总统发表讲话...",
    source="财联社",
    time_str="15:00",
    time_ts=1711555200,
)

# 检测事件类型
event_type = detect_event_type(news)  # → "military"

# 评分
market = MarketImpact(oil_pct=2.5, gold_pct=1.2)
score = score_news_item(news, market, state)  # → EventScore(total=68.0)

# 生成指纹
fingerprint = generate_event_fingerprint(news, event_type)
# → "middle-east|military|oil|attack"

# 推送决策
should_push, reason = should_push_event(event, state)

配置

# 推送阈值
PUSH_THRESHOLD_HIGH = 70
PUSH_THRESHOLD_WATCH = 50

# 冷却时间（分钟）
COOLDOWN_HIGH = 90
COOLDOWN_MEDIUM = 180

依赖

• llm_news_analyzer (可选): LLM 语义分析模块
• geo_market_impact_mapper: 市场影响数据

相关文件

• 主脚本：geo_event_router.py
• 配置：同目录 config.py

版本

• v1.0.0: 初始版本，从 smart-geo-push.py v2.0 拆分