policy-news-analysis
Analyze policy documents and current affairs for decision signals using a seven-step framework with pre-analysis knowledge diagnosis and active knowledge-gap filling. Use for policy/event interpretati
SKILL.mdskills/policy-news-analysis/SKILL.md
GitHubPolicy News Analysis Skill
- Confirm audience first: investor / industry practitioner / business operator / researcher-media.
- Run Step 0 before analysis: identify domain tags and explicit knowledge-gap hypotheses.
- Query SQLite directionally (no full-table scans):
domain_knowledgehistorical_referencecurrent_state
- Execute workflow in strict order:
- Step 1 anomaly scan
- Step 2 self-explanation from source material
- Step 3 external explanation for anomalies
- Step 4 priority from ordering/structure
- Step 5 logic-closure subtraction
- Step 6 changed vs unchanged (must reach wording-level, not only policy-tone)
- Step 7 global external calibration
- Step 8 synthesize investor decision summary (delivery layer)
- If a critical-node knowledge gap blocks the core logic chain, run active gap-filling flow (mandatory, cannot skip):
- Level 1: SQLite retry
- Level 2A: built-in web search/fetch (single factual query)
- Level 2B:
agent-browser-clawdbotonly for deep page navigation - Evaluate source quality/time-validity/logic consistency before use
- Mark sources and uncertainty explicitly:
[数据库]database-backed[推断]analyst inference (not directly stated in source)[待验证]uncertain external data[知识缺口]missing critical knowledge[外部补充]external search supplement[外部补充:browser]browser-based supplement
- Never mix
[推断]into source-grounded conclusions. - Output full knowledge sedimentation content (records), then write to SQLite via script. Do not report counts only.
- Follow reference markdown format strictly. Do not reorder/merge sections.
- In knowledge sedimentation JSON, include source annotation fields for each record:
source_type: A(政策原文) / B(官方解读) / C(二手报道)source_url: article URLsource_title: article titlesource_quote: one directly supporting sentence excerpt from sourcesource_date: publication date if available
- If
source_typeis B, append this marker to quote context:[解读稿引用,非原文条款]. - C-type sources are not allowed for sediment write-back.
Output structure (required)
A. 分析日志(Step 0~7)
- 核心信号摘要(<=3句)
- 信号来源(第1-3步)
- 核心逻辑链
- 值得深挖的2-3个关键点(含行动意义)
- 延续未变部分(含措辞层检查)
- 知识缺口与待验证事项
- 知识沉淀包(完整记录内容,不仅计数)
B. 交付摘要(Step 8,必须单独输出) 8. 投资决策摘要(按框架“受众专项输出”模板)
Resources
- Main framework:
references/openclaw_policy_analysis_framework.md - DB init schema:
scripts/init_policy_kb.py - Minimal writer:
scripts/write_policy_knowledge.py
Commands
python3 skills/policy-news-analysis/scripts/init_policy_kb.py --db memory/policy_knowledge.db
python3 skills/policy-news-analysis/scripts/write_policy_knowledge.py --db memory/policy_knowledge.db --input /tmp/sediment.json