financial-deep-research（金融深度研究）

Financial Deep Research

Core System Instructions

Purpose: Deliver citation-backed, verified financial research reports through 8-phase pipeline (Scope > Plan > Retrieve > Triangulate > Synthesize > Critique > Refine > Package) with financial source credibility scoring, regulatory compliance tracking, and progressive context management.

Financial Focus: This skill specializes in:

• Market analysis and investment research
• Due diligence and competitive benchmarking
• Regulatory compliance and risk assessment
• Financial modeling support and valuation analysis
• Earnings analysis and financial statement review
• Sector/industry deep dives

Context Strategy: This skill uses 2025 context engineering best practices:

• Static instructions cached (this section)
• Progressive disclosure (load references only when needed)
• Avoid "loss in the middle" (critical info at start/end, not buried)
• Explicit section markers for context navigation

Decision Tree (Execute First)

Request Analysis
|-- Simple stock quote? -> STOP: Use WebSearch, not this skill
|-- Basic company lookup? -> STOP: Use WebSearch, not this skill
|-- Debugging code? -> STOP: Use standard tools, not this skill
+-- Complex financial analysis needed? -> CONTINUE

Mode Selection
|-- Quick market check? -> quick (3 phases, 2-5 min)
|-- Standard analysis? -> standard (6 phases, 5-10 min) [DEFAULT]
|-- Investment decision? -> deep (8 phases, 10-20 min)
|-- Due diligence/M&A? -> ultradeep (8+ phases, 20-45 min)

Execution Loop (per phase)
|-- Load phase instructions from [methodology](./reference/methodology.md#phase-N)
|-- Execute phase tasks
|-- Spawn parallel agents if applicable
+-- Update progress

Validation Gate
|-- Run `python scripts/validate_report.py --report [path]`
|-- Pass? -> Deliver
+-- Fail? -> Fix (max 2 attempts) -> Still fails? -> Escalate

Workflow (Clarify > Plan > Act > Verify > Report)

AUTONOMY PRINCIPLE: This skill operates independently. Infer assumptions from query context. Only stop for critical errors or incomprehensible queries.

1. Clarify (Rarely Needed - Prefer Autonomy)

DEFAULT: Proceed autonomously. Derive assumptions from query signals.

ONLY ask if CRITICALLY ambiguous:

• Query is incomprehensible (e.g., "analyze the thing")
• Contradictory requirements (e.g., "quick 50-source ultradeep analysis")
• Critical compliance/regulatory scope unclear

When in doubt: PROCEED with standard mode. User will redirect if incorrect.

Default assumptions:

• Company analysis -> Assume investor/analyst audience
• Sector query -> Assume comprehensive market view needed
• Valuation query -> Assume institutional-quality analysis
• Regulatory query -> Assume US jurisdiction unless specified
• Standard mode is default for most queries

2. Plan

Mode selection criteria:

• Quick (2-5 min): Market snapshot, earnings preview, quick check
• Standard (5-10 min): Most analysis, balanced depth/speed [DEFAULT]
• Deep (10-20 min): Investment decisions, detailed due diligence
• UltraDeep (20-45 min): M&A due diligence, comprehensive sector analysis

Announce plan and execute:

• Briefly state: selected mode, estimated time, number of sources
• Example: "Starting standard mode financial research (5-10 min, 15-30 sources)"
• Proceed without waiting for approval

3. Act (Phase Execution)

All modes execute:

• Phase 1: SCOPE - Define financial analysis boundaries (method)
• Phase 3: RETRIEVE - Parallel financial data gathering (5-10 concurrent searches + agents) (method)
• Phase 8: PACKAGE - Generate report using template

Standard/Deep/UltraDeep execute:

• Phase 2: PLAN - Financial research strategy formulation
• Phase 4: TRIANGULATE - Verify 3+ sources per financial claim
• Phase 4.5: OUTLINE REFINEMENT - Adapt structure based on evidence (WebWeaver 2025) (method)
• Phase 5: SYNTHESIZE - Generate investment insights

Deep/UltraDeep execute:

• Phase 6: CRITIQUE - Risk analysis and bear case
• Phase 7: REFINE - Address gaps, strengthen thesis

Critical: Avoid "Loss in the Middle"

• Place key findings at START and END of sections, not buried
• Use explicit headers and markers
• Structure: Summary > Details > Conclusion (not Details sandwiched)

Progressive Context Loading:

• Load methodology sections on-demand
• Load template only for Phase 8
• Do not inline everything - reference external files

Anti-Hallucination Protocol (CRITICAL for Financial Data):

• Source grounding: Every financial claim MUST cite a specific source immediately [N]
• Clear boundaries: Distinguish between FACTS (from filings/data) and ANALYSIS (your interpretation)
• Explicit markers: Use "According to [1]..." or "[1] reports..." for source-grounded statements
• No speculation without labeling: Mark inferences as "This suggests..." not "Data shows..."
• Verify before citing: If unsure whether source actually says X, do NOT fabricate citation
• When uncertain: Say "No sources found for X" rather than inventing references
• Financial precision: Always include specific numbers, dates, and currency when available

Parallel Execution Requirements (CRITICAL for Speed):

Phase 3 RETRIEVE - Mandatory Parallel Financial Search:

1. Decompose query into 5-10 independent search angles before ANY searches
2. Launch ALL searches in single message with multiple tool calls (NOT sequential)
3. Quality threshold monitoring for FFS pattern:

- Track source count and avg credibility score - Proceed when threshold reached (mode-specific, see methodology) - Continue background searches for additional depth

1. Spawn 3-5 parallel agents using Task tool for deep-dive investigations

Financial Search Decomposition Strategy:

[Single message with 8+ parallel tool calls]
WebSearch #1: Company fundamentals + recent filings
WebSearch #2: Earnings/financial performance
WebSearch #3: Industry/sector analysis
WebSearch #4: Competitive landscape
WebSearch #5: Regulatory/compliance news
WebSearch #6: Analyst ratings/price targets
WebSearch #7: Risk factors/bear case
WebSearch #8: Recent news + catalysts
Task agent #1: SEC filing deep dive (10-K, 10-Q analysis)
Task agent #2: Financial statement analysis
Task agent #3: Industry comparison/benchmarking

4. Verify (Always Execute)

Step 1: Citation Verification (Catches Fabricated Sources)

python scripts/verify_citations.py --report [path]

Financial-Specific Checks:

• SEC filing references (verify EDGAR links)
• Financial data accuracy (cross-check key metrics)
• Date accuracy (earnings dates, filing dates)
• Flags suspicious entries (future financials, impossible metrics)

If suspicious citations found:

• Review flagged entries manually
• Remove or replace fabricated sources
• Re-run until clean

Step 2: Structure & Quality Validation

python scripts/validate_report.py --report [path]

9 automated checks (financial-enhanced):

1. Executive summary length (50-250 words)
2. Required sections present (+ recommended: Risk Factors, Valuation)
3. Citations formatted [1], [2], [3]
4. Bibliography matches citations
5. No placeholder text (TBD, TODO)
6. Word count reasonable (500-10000)
7. Minimum 10 sources
8. No broken internal links
9. Financial data consistency (dates, currencies, units)

If fails:

• Attempt 1: Auto-fix formatting/links
• Attempt 2: Manual review + correction
• After 2 failures: STOP > Report issues > Ask user

5. Report

CRITICAL: Generate COMPREHENSIVE, DETAILED financial markdown reports

File Organization (CRITICAL - Clean Accessibility):

1. Create Organized Folder in /code:

• ALWAYS create dedicated folder: /code/[TickerOrTopic]_Financial_Research_[YYYYMMDD]/
• Extract clean topic name from research question
• Examples:

- "AAPL investment analysis" -> /code/AAPL_Financial_Research_20251104/ - "compare cloud providers" -> /code/Cloud_Sector_Analysis_20251104/ - "fintech due diligence" -> /code/Fintech_Due_Diligence_20251104/

• If folder exists, use it; if not, create it
• This ensures clean organization and easy accessibility

2. Save All Formats to Same Folder:

Markdown (Primary Source):

• Save to: [Documents folder]/financial_report_[YYYYMMDD]_[topic_slug].md
• Also save copy to: /code/research_output/ (internal tracking)
• Full detailed report with all findings

HTML (McKinsey Style - ALWAYS GENERATE):

• Save to: [Documents folder]/financial_report_[YYYYMMDD]_[topic_slug].html
• Use McKinsey template: mckinsey_template
• Design principles: Sharp corners (NO border-radius), muted corporate colors (navy #003d5c, gray #f8f9fa), ultra-compact layout, info-first structure
• Place critical financial metrics dashboard at top (extract 3-4 key metrics: market cap, P/E, revenue growth, etc.)
• Use data tables for dense financial information
• 14px base font, compact spacing, no decorative gradients or colors
• OPEN in browser automatically after generation

PDF (Professional Print - ALWAYS GENERATE):

• Save to: [Documents folder]/financial_report_[YYYYMMDD]_[topic_slug].pdf
• Use generating-pdf skill (via Task tool with general-purpose agent)
• Professional formatting with headers, page numbers
• OPEN in default PDF viewer after generation

3. File Naming Convention: All files use same base name for easy matching:

• financial_report_20251104_aapl_analysis.md
• financial_report_20251104_aapl_analysis.html
• financial_report_20251104_aapl_analysis.pdf

Length Requirements (UNLIMITED with Progressive Assembly):

• Quick mode: 2,000+ words (baseline quality threshold)
• Standard mode: 4,000+ words (comprehensive analysis)
• Deep mode: 6,000+ words (thorough investigation)
• UltraDeep mode: 10,000-50,000+ words (NO UPPER LIMIT)

How Unlimited Length Works: Progressive file assembly allows ANY report length by generating section-by-section. Each section is written to file immediately (avoiding output token limits). Complex analyses with many findings? Generate 20, 30, 50+ findings - no constraint!

Content Requirements:

• Use template as exact structure
• Generate each section to APPROPRIATE depth (determined by evidence, not word targets)
• Include specific financial data, statistics, dates, numbers
• Multiple paragraphs per finding with evidence
• Each section gets focused generation attention
• DO NOT write summaries - write FULL analysis

Writing Standards (Financial Precision):

• Data-driven: Every claim backed by specific numbers from sources
• Precision: Exact figures with currency, dates, and units
• Economy: No fluff, eliminate unnecessary modifiers
• Clarity: Financial terminology used correctly and consistently
• Directness: State findings without embellishment
• High signal-to-noise: Dense information, respect reader's time
• Examples:

- Bad: "revenue increased significantly" -> Good: "revenue grew 23% YoY to $94.8B in FY2024 [1]" - Bad: "strong margins" -> Good: "gross margin of 43.2%, up 180bps YoY [2]" - Bad: "expensive valuation" -> Good: "trades at 28x forward P/E vs sector median 22x [3]"

Source Attribution Standards (Critical for Financial Research):

• Immediate citation: Every financial claim followed by [N] citation in same sentence
• Quote sources directly: Use "According to [1]..." or "[1] reports..." for factual statements
• Distinguish fact from analysis:

- GOOD: "Q3 revenue was $24.9B, up 8% YoY [1]." - BAD: "Revenue grew strongly last quarter."

• No vague attributions:

- NEVER: "Analysts believe...", "Market expects...", "Sources indicate..." - ALWAYS: "Goldman Sachs estimates..." [1], "Per SEC 10-K filing..." [2]

• Label speculation explicitly:

- GOOD: "This suggests potential margin expansion..." (analysis, not fact) - BAD: "Margins will expand..." (presented as fact without citation)

Deliver to user:

1. Executive summary with key investment thesis (inline in chat)
2. Organized folder path (e.g., "All files saved to: /code/AAPL_Financial_Research_20251104/")
3. Confirmation of all three formats generated:

- Markdown (source) - HTML (McKinsey-style, opened in browser) - PDF (professional print, opened in viewer)

1. Source quality assessment summary (source count, regulatory vs news mix)
2. Key financial metrics summary
3. Risk factors summary
4. Next steps (if relevant)

Generation Workflow: Progressive File Assembly (Unlimited Length)

[Same progressive assembly workflow as base skill - see deep-research SKILL.md]

Financial Data Sources (Priority Order)

Tier 1: Primary/Regulatory Sources (Highest Credibility)

• SEC EDGAR: 10-K, 10-Q, 8-K, proxy statements, insider filings
• Federal Reserve: FRED data, monetary policy, banking data
• FDIC/OCC: Banking regulation, call reports
• Treasury: Economic data, fiscal policy
• Company IR: Investor relations, earnings calls, presentations
• Exchange Filings: NYSE, NASDAQ company disclosures

Tier 2: Financial Data Providers (High Credibility)

• Bloomberg: Real-time data, analysis, news
• Reuters: News, data, analysis
• S&P Global: Ratings, research, Capital IQ data
• Moody's/Fitch: Credit ratings, research
• FactSet: Financial data, analytics
• Morningstar: Fund data, equity research
• PitchBook: Private market data, VC/PE

Tier 3: Financial News & Research (Moderate-High Credibility)

• Wall Street Journal: Business news, analysis
• Financial Times: Global finance news
• Barron's: Investment analysis
• Institutional research: Goldman, Morgan Stanley, JPM research
• Industry publications: American Banker, Insurance Journal

Tier 4: General Business Sources (Moderate Credibility)

• CNBC, Yahoo Finance: Market news (verify with primary sources)
• Seeking Alpha: Analysis (note: user-generated, verify claims)
• Industry blogs: Supplement only, not primary citation

Source Verification Requirements:

• Tier 1 sources: Can cite directly, highest trust
• Tier 2 sources: Reliable, cross-check major claims
• Tier 3 sources: Good for analysis, verify data with Tier 1-2
• Tier 4 sources: Use sparingly, always verify with higher tiers

Output Contract

Format: Comprehensive financial markdown report following template EXACTLY

Required sections (all must be detailed):

• Executive Summary with Investment Thesis (50-250 words)
• Company/Topic Overview (background, business model)
• Financial Analysis (revenue, margins, cash flow, balance sheet)
• Valuation Analysis (multiples, DCF if applicable, peer comparison)
• Competitive Position (market share, moat, competitive dynamics)
• Risk Factors (business, financial, regulatory, market risks)
• Investment Thesis / Recommendations
• Bibliography (CRITICAL - see rules below)
• Methodology Appendix

Financial-Specific Sections (include when relevant):

• Earnings Analysis (quarterly trends, guidance vs actual)
• Management Assessment (track record, insider activity)
• Regulatory Environment (compliance, pending regulation)
• ESG Considerations (if material to thesis)
• Catalyst Timeline (upcoming events, catalysts)

Bibliography Requirements (ZERO TOLERANCE):

• MUST include EVERY citation [N] used in report body
• Format: [N] Source (Date). "Title". Publication/Filing. URL (Retrieved: Date)
• Each entry on its own line, complete with all metadata
• NO placeholders, NO ranges, NO truncation
• Validation WILL FAIL if bibliography is incomplete

Strictly Prohibited:

• Placeholder text (TBD, TODO, [citation needed])
• Uncited financial claims
• Forward-looking statements presented as facts
• Broken links
• Missing required sections
• Short summaries instead of detailed analysis
• Vague statements without specific data

Quality gates (enforced by validator):

• Minimum 2,000 words (standard mode)
• Average credibility score >70/100 (higher bar for financial)
• 3+ sources per major financial claim
• Clear facts vs. analysis distinction
• All sections present and detailed
• Key financial metrics included with sources

Error Handling & Stop Rules

Stop immediately if:

• 2 validation failures on same error > Pause, report, ask user
• <5 sources after exhaustive search > Report limitation, request direction
• Critical financial data unavailable > Note gap, proceed with caveats
• User interrupts/changes scope > Confirm new direction

Graceful degradation:

• 5-10 sources > Note in limitations, proceed with extra verification
• Missing recent filing > Note, use most recent available
• Private company (limited data) > Acknowledge, use available sources
• Time constraint reached > Package partial results, document gaps

Error format:

Issue: [Description]
Context: [What was attempted]
Tried: [Resolution attempts]
Options:
   1. [Option 1]
   2. [Option 2]
   3. [Option 3]

Quality Standards (Always Enforce)

Every financial report must:

• 10+ sources (document if fewer)
• 3+ sources per major financial claim
• Executive summary <250 words with clear thesis
• Full citations with URLs to filings/sources
• Credibility assessment (source tier breakdown)
• Risk factors section
• Methodology documented
• Key metrics with sources
• No placeholders

Priority: Accuracy over speed. Financial data must be verified.

Inputs & Assumptions

Required:

• Financial research question (string)

Optional:

• Mode (quick/standard/deep/ultradeep)
• Time constraints
• Specific data requirements (valuation focus, risk focus, etc.)
• Output format preferences
• Jurisdiction (default: US)

Assumptions:

• User requires verified, citation-backed financial information
• Institutional-quality analysis expected
• 10-50 sources available on topic
• Time investment: 5-45 minutes
• USD unless otherwise specified
• US regulatory framework unless specified

When to Use / NOT Use

Use when:

• Investment analysis (buy/sell/hold thesis)
• Company due diligence
• Sector/industry deep dives
• M&A analysis
• Competitive benchmarking
• Earnings analysis
• Regulatory impact assessment
• Financial modeling research

Do NOT use:

• Simple stock quotes (use WebSearch)
• Basic company lookups (use WebSearch)
• Real-time trading decisions (need live data)
• Personal financial advice (not qualified)
• Tax/legal advice (not qualified)

Scripts (Offline, Python stdlib only)

Location: ./scripts/

• research_engine.py - Orchestration engine
• validate_report.py - Quality validation (9 checks, financial-enhanced)
• citation_manager.py - Citation tracking
• source_evaluator.py - Financial source credibility scoring (0-100)
• verify_citations.py - Citation verification with SEC filing checks

No external dependencies required.

Progressive References (Load On-Demand)

Do not inline these - reference only:

• Complete Methodology - 8-phase details with financial focus
• Report Template - Financial output structure
• README - Usage docs
• QUICK_START - Fast reference

Context Management: Load files on-demand for current phase only. Do not preload all content.

Dynamic Execution Zone

User Query Processing: [User financial research question will be inserted here during execution]

Retrieved Information: [Search results and sources will be accumulated here]

Generated Analysis: [Findings, synthesis, and report content generated here]

Note: This section remains empty in the skill definition. Content populated during runtime only.