source-coverage-check

name	source-coverage-check
description	Pre-writing source sufficiency check for investment value report chapters. Use BEFORE launching the report pipeline (ivr_run_report_pipeline) to verify each selected chapter has adequate source coverage. Searches both trusted source domains and broad web, then reports gaps via ask_clarification so the user can decide to proceed, add more PDFs, or remove chapters.

Source Coverage Check — 信源充分性检查

When to Use

Trigger this skill AFTER ivr_create_report_task returns success and BEFORE calling ivr_run_report_pipeline. Also use it when the user uploads new PDFs and asks "现在证据够不够写第X章？"

Rule: Never call ivr_run_report_pipeline before completing this check and getting the user's explicit confirmation.

Step 1: Gather Inputs

Before searching, gather:

1. The company's industry_code from ivr_list_companies output
2. The task's chapter_ids (which chapters are selected for this task)
3. The user's trusted sources via ivr_get_trusted_sources — these are domains the user configured in the workbench
4. The list of uploaded PDFs via ivr_list_company_files — these are already RAG-indexed

Step 2: Per-Chapter Evidence Checklist

For each selected chapter, check whether the following evidence types are covered:

ch4 — 行业分析

#	Required Evidence	Red Flag (blocking gap)
1	Industry classification and scope	No SIC/industry code data at all
2	Market size (current + CAGR projection)	Missing market size or sourced from random blog
3	Competitive landscape (top 5-10 players, market share)	No competitor data
4	Regulatory and policy environment	No regulatory policy documents
5	Technology trends and substitution risks	No technology trend analysis

ch5 — 公司分析

#	Required Evidence	Red Flag (blocking gap)
1	Company business model description	No business description from filings
2	Revenue breakdown by segment (last 3 years)	No revenue segmentation data
3	Profitability metrics (gross margin, net margin)	No profit data from filings
4	R&D pipeline and technology moat	No R&D disclosure
5	Management and governance structure	No governance info

ch6 — 募投项目

#	Required Evidence	Red Flag (blocking gap)
1	Fundraising purpose and amount	No project investment plan
2	Capacity expansion details	No capacity data
3	Project timeline and milestones	No timeline info
4	ROI projection and payback period	No financial projection for the project
5	Risk factors specific to the project	No project risk disclosure

ch2 — 盈利预测

#	Required Evidence	Red Flag (blocking gap)
1	Historical revenue (3 years, by segment)	Missing historical revenue
2	Historical cost structure (COGS, SG&A, R&D)	Missing cost breakdown
3	Industry growth rate benchmarks	No industry growth data
4	Company guidance or analyst consensus	No forward-looking estimates
5	Key assumptions (pricing, volume, margins)	No basis for assumptions

ch3 — 估值分析

#	Required Evidence	Red Flag (blocking gap)
1	Comparable company list (5+ peers)	No comparable companies found
2	Comparable PE/PB/PS multiples	No valuation multiples data
3	Risk-free rate and equity risk premium	No market parameters
4	DCF inputs (FCF, WACC, growth rate)	Missing 2+ DCF parameters
5	Sensitivity analysis basis	No scenario data

ch7 — 风险提示

#	Required Evidence	Red Flag (blocking gap)
1	Risk factors from prospectus	No risk disclosure document
2	Industry-specific risk reports	No industry risk analysis
3	Litigation and compliance issues	No legal/compliance check
4	Macro and policy risk factors	No policy risk assessment

Step 3: Search Strategy

For each chapter and each missing evidence type:

Pass 1 — Trusted Sources (priority)

For each trusted source domain from ivr_get_trusted_sources, use web_search with site: modifier:

web_search(query="<company> <topic> <year> site:<trusted_domain>")

Example: web_search(query="示例公司 营收结构 2024 site:cninfo.com.cn")

Pass 2 — Broad Search (supplementary)

For evidence still missing after Pass 1, search without domain restriction:

web_search(query="<company> <topic> <year>")

Pass 3 — Deep Read

For promising results from either pass, use web_fetch to extract specific data points. Only fetch 2-3 most relevant pages per chapter.

Industry-Specific Domains (supplement to workbench-configured trusted sources)

If the user hasn't configured trusted sources for a specific industry, use these authoritative domains:

bio (生物医药):

Domain	Label
cde.org.cn	国家药监局药审中心
nmpa.gov.cn	国家药品监督管理局
pharmcube.com	医药魔方
yaozh.com	药智网

new_energy (新能源):

Domain	Label
nea.gov.cn	国家能源局
cbea.com	中国电池工业协会
cnemining.com	中国新能源矿业网

semiconductor (半导体):

Domain	Label
miit.gov.cn	工业和信息化部
csia.net.cn	中国半导体行业协会
icwise.com	芯谋研究

finance (金融):

Domain	Label
nfra.gov.cn	国家金融监督管理总局
pbc.gov.cn	中国人民银行
sse.com.cn	上海证券交易所
szse.cn	深圳证券交易所

Step 4: Coverage Judgment

For each chapter, assign a coverage level:

• 充分 (sufficient): 4+ evidence types found from authoritative sources. Chapter can be written with confidence.
• 部分覆盖 (partial): 2-3 evidence types found. Chapter can be written but will have "待补充" sections.
• 不足 (insufficient): 0-1 evidence types found. Chapter should NOT be written — the output would be mostly fabricated.

Step 5: Report Gaps via ask_clarification

After checking all chapters, call ask_clarification with:

• clarification_type: "risk_confirmation"
• context: Brief summary of company, industry, and checked chapters
• question: Structured per-chapter report. Example format:

信源充分性检查结果：

| 章节 | 覆盖等级 | 已找到 | 缺失项 |
|------|---------|--------|--------|
| ch4 行业分析 | 充分 | 市场规模、竞争格局、政策文件 | — |
| ch5 公司分析 | 部分覆盖 | 商业模式、管理层 | 近3年营收分拆、研发管线详情 |
| ch2 盈利预测 | 不足 | 行业增长率 | 历史财务数据、成本结构、分析师预期 |

建议：ch2 盈利预测缺少核心财务数据，建议补充公司财报或券商研报PDF后再生成。

• options: ["继续生成报告", "我先补充更多 PDF 再继续", "移除信源不足的章节后生成"]

Step 6: Act on User Decision

• 继续生成报告: Call ivr_run_report_pipeline(task_id) immediately.
• 我先补充更多 PDF: Wait for the user to upload files. After upload, re-run this skill from Step 2.
• 移除信源不足的章节后生成: Ask the user which chapters to remove, update the task if needed, then proceed.