Security News Subscription Skill v1.0
金融行业安全资讯自动采集、评分、交叉验证、MD 报告生成系统。
1. Skill Scope
Core: Financial industry (bank/insurance/securities/asset/payment) news collection pipeline.
Focus: National security standards | Security tech | Enforcement info | AI finance | AI security | Industry dynamics
Abilities: Info collection | Quality scoring (100pt + 6-cat + source rating + cross-verify) | MD report generation
Boundaries: NOT responsible for content creation, deep analysis, policy interpretation
2. Trigger Scenarios
| Scenario | Trigger | Note |
|---|---|---|
| Daily report | "daily news" | Generate daily financial security news |
| Weekly report | "weekly report" | Summarize last week news |
| Monthly report | "monthly report" | Summarize this month news |
| Source update | "subscribe XX" | Manage news sources |
3. Classification System
Priority: Enforcement(5) > AI Finance(4) >= AI Security(4) > Security Tech(3) > National Standards(2) > Industry(1)
Important: Security tech判断优先于National standards
| Category | Weight | Content | Keywords | Authority Source |
|---|---|---|---|---|
| Enforcement | 5 | Bank/securities penalty, AML violation | penalty/fine/rectification/warning letter/administrative penalty | nfra.gov.cn, csrc.gov.cn |
| AI Finance | 4 | LLM, intelligent investment, smart risk control | LLM, intelligent investment, smart risk control, digital employee | 36kr, Xinhua, THF |
| AI Security | 4 | Prompt injection, deepfake, adversarial attack, AI supply chain | Prompt injection, deepfake, adversarial, model theft, AIGC security, AI governance | FreeBuf, SecBot |
| Security Tech | 3 | CVE, APT, ransomware, data breach, cloud security | CVE, 0day, APT, DDoS, supply chain attack, zero trust, cloud config error | SecK, FreeBuf, CNNVD |
| National Standards | 2 | GB/T, JR/T, multi-level protection, CII, cryptography | GB/T, JR/T, JR/T 0068, TC260, data security law | cfstc.pbc.gov.cn, nfra.gov.cn |
| Industry | 1 | Regulatory dynamics, rating methods, market trends | regulatory dynamics, annual report, Basel | nfra.gov.cn, pbc.gov.cn |
Scoring Keywords (11 categories)
Network security | AI security | Securities standards | Banking standards | PBOC regulation
Cloud security | Data security | Threat intel | Compliance | AI finance applications | Industry orgs
See scripts/score_news.py SECURITY_KEYWORDS and CATEGORY_BOOST_KEYWORDS
4. Source Maintenance Principles
P0: Never delete sources | P1: Disable > delete | P2: Replace > delete | P3: Build crawler > abandon
5. Execution Flow (3 Steps)
Step 1: Confirm sources
Step 2: Fetch and store
Step 2.5: Source rating (auto) - Excellent/Good/Normal/Poor, with score multiplier
Step 2.7: Cross-verification (auto) - Multi-source verification, VERIFIED/SINGLE/DISPUTED/UNKNOWN
Step 3: Generate MD report
Source Rating
| Rating | Score | Multiplier | Threshold adjust |
|---|---|---|---|
| Excellent | 90-100 | x1.15 | base-10 |
| Good | 70-89 | x1.05 | base-5 |
| Normal | 50-69 | x1.00 | base |
| Poor | 0-49 | x0.85 | base+15 |
Cross-Verification
| Status | Note |
|---|---|
| VERIFIED | Multi-source confirmed |
| SINGLE | Single source, no corroboration |
| DISPUTED | Title similar but key info contradicts |
| UNKNOWN | Source unreachable or info insufficient |
Rules: Score>=80 (mandatory verify) | Score 60-79 (random 30%) | Score<60 (skip)
6. Quick Commands
python3 news_digest.py --mode daily
python3 news_digest.py --mode weekly
python3 news_digest.py --mode monthly
python3 news_digest.py --mode daily --top 50 --min-score 60
python3 news_digest.py --mode daily --skip-cross-verify
7. Authority Sources
Threat Intel (14 vendors): Tencent, QAX, 360, Sangfor, Chaitin, DBAPPSecurity, Venustech, NSFOCUS, SecWiki
Financial Reg: PBC, NFRA, CSRC, CFSTC, SecRSS, Xinhua Finance
8. Version History
| Ver | Date | Changes |
|---|---|---|
| v1.0 | 2026-06-01 | 初始版本：三步流程（确认信息源 → 采集入库 → 生成 MD 报告），纯 MD 输出。支持 74 个信息源、六分类体系、100 分评分 + 来源评级加权、交叉验证、知识库归档。 |
9. File Structure
news-subscription-lite/
SKILL.md (v1.0)
scripts/
news_digest.py - main entry (3-step CLI)
pipeline.py - 3-step pipeline engine
score_news.py - scoring engine
manage_sources.py - source management
discover_sources.py - WeChat source discovery
source_rating.py - source 4-level rating
cross_verify.py - cross-verification module
knowledge_manager.py - raw archive only (no wiki/IMA sync)
operation_logger.py - pipeline logging
fetch_rss.py, fetch_website.py, fetch_wechat.py,
fetch_wechat_advanced.py, fetch_wechat_mp.py,
fetch_secrss.py, fetch_finance_gov.py,
fetch_cnnvd.py, fetch_cnvd.py
references/
source_domains.json, scoring_keywords.md, security_news_format.md,
template_spec.md, setup_guide.md, version_history.md
References 按需加载指引
以下 references 文件在特定场景下需读取，避免每次加载全部：
| 文件 | 何时读取 |
|---|---|
references/security_news_format.md |
需要了解资讯分类规则、评分字段定义时 |
references/scoring_keywords.md |
调试评分异常、需要调整关键词加权时 |
references/source_domains.json |
新增/修改订阅源域名、排查来源评级时 |
references/setup_guide.md |
首次安装、排查环境依赖时 |
references/version_history.md |
追溯版本变更、确认功能迭代时间线时 |
references/template_spec.md |
修改 MD 报告模板排版时 |
