analyze

name	analyze
description	Run structured analysis (financial, property, data) using the hybrid ralph pattern. Decomposes research questions into parallel investigation tracks with verification.

Analysis Orchestrator

Run structured, multi-angle analysis using Agent Teams for hypothesis generation and Ralph loops for data crunching.

Process

Phase 1: Define the Analysis Brief

Interview the user about:

• What they want to analyze (specific stock, portfolio, property, dataset)
• Why — what decision does this inform? (buy/sell, invest, allocate, compare)
• Data sources — what data is available? (APIs, CSVs, databases, web)
• Output format — report, spreadsheet, dashboard, or just a summary?
• Time constraints — historical period, deadline for the analysis

Phase 2: Generate analysis.json

Decompose the analysis into tasks, just like prd.json but for research:

{
  "analysis": "Portfolio Risk Assessment Q1 2026",
  "branchName": "analysis/portfolio-risk-q1",
  "description": "Comprehensive risk analysis of current portfolio positions",
  "objective": "Determine if portfolio is overexposed to any single sector or risk factor",
  "tasks": [
    {
      "id": "task-001",
      "title": "Gather position data",
      "description": "Pull current holdings, cost basis, and market values",
      "priority": 1,
      "passes": false,
      "acceptance_criteria": [
        "All positions exported to data/positions.csv",
        "Data includes: ticker, shares, cost_basis, current_price, sector",
        "Totals reconcile with known portfolio value within 1%"
      ],
      "track": "mechanical",
      "dependencies": []
    },
    {
      "id": "task-002",
      "title": "Calculate sector concentration",
      "description": "Compute allocation percentages by sector and identify overweight positions",
      "priority": 2,
      "passes": false,
      "acceptance_criteria": [
        "Sector allocations sum to 100%",
        "Any sector > 30% flagged as overweight",
        "Results saved to output/sector-analysis.md"
      ],
      "track": "mechanical",
      "dependencies": ["task-001"]
    },
    {
      "id": "task-003",
      "title": "Interpret results and form recommendations",
      "description": "Analyze the findings, compare to benchmarks, and draft actionable recommendations",
      "priority": 3,
      "passes": false,
      "acceptance_criteria": [
        "Recommendations are specific and actionable",
        "Each recommendation cites supporting data",
        "Risk/reward tradeoffs discussed"
      ],
      "track": "creative",
      "dependencies": ["task-002"]
    }
  ]
}

Phase 3: Track Assignment for Analysis

Mechanical tasks (Ralph loop):

• Data gathering and cleaning
• Calculations (ratios, returns, correlations, projections)
• Backtesting with defined parameters
• Generating charts and tables
• Validation (numbers reconcile, formulas correct)

Creative tasks (Agent Teams):

• Interpreting results — what do the numbers mean?
• Forming hypotheses — why did X happen?
• Comparing scenarios — which approach is better and why?
• Writing narrative analysis — executive summary, recommendations
• Challenging assumptions — devil's advocate on the findings

Phase 4: Execution

For mechanical tasks, modify ralph.sh to use analysis.json instead of prd.json:

PRD_FILE="analysis.json" ./scripts/ralph.sh 10

For creative tasks, spawn an Agent Team:

Create an agent team for analysis interpretation:
- Analyst: Interprets data, identifies trends, forms hypotheses
- Skeptic: Challenges assumptions, finds flaws in reasoning, checks for biases
- Strategist: Translates findings into actionable recommendations

All agents must cite specific data points from the mechanical track outputs.

Phase 5: Compile Report

After both tracks complete, compile findings into a final report:

• Executive summary (2-3 paragraphs)
• Key findings with supporting data
• Visualizations (charts, tables)
• Recommendations with confidence levels
• Methodology and data sources
• Caveats and limitations

Save as output/analysis-report.md or use the /docx skill for a formatted Word document.

Analysis Templates

Stock Analysis

/analyze AAPL

Tasks generated:
1. [mechanical] Pull 12-month price history and compute returns
2. [mechanical] Calculate key ratios (P/E, P/B, EV/EBITDA, FCF yield)
3. [mechanical] Compute technical indicators (RSI, MACD, moving averages)
4. [mechanical] Pull and parse latest earnings transcript
5. [creative] Compare valuation vs sector peers — is it cheap or expensive?
6. [creative] Assess competitive moat and growth catalysts
7. [creative] Bull/bear case with price targets
8. [creative] Final recommendation with confidence level

Property Investment Analysis

/analyze "Condo at 123 Main St, $500K asking"

Tasks generated:
1. [mechanical] Calculate rental yield (gross and net)
2. [mechanical] Compute cash-on-cash return with financing scenarios
3. [mechanical] Model 5/10/15 year appreciation scenarios (conservative, base, bull)
4. [mechanical] Calculate break-even occupancy rate
5. [mechanical] Pull comparable sales and rental comps from area
6. [creative] Assess location quality (transit, amenities, growth trajectory)
7. [creative] Identify risks (vacancy, maintenance, market downturn, regulation)
8. [creative] Compare vs alternative investments (index funds, REITs, other properties)
9. [creative] Final recommendation: buy / negotiate / pass

Portfolio Review

/analyze portfolio

Tasks generated:
1. [mechanical] Export all positions with cost basis
2. [mechanical] Calculate P&L (realized + unrealized) by position
3. [mechanical] Compute risk metrics (Sharpe, max drawdown, beta, correlation matrix)
4. [mechanical] Analyze sector/geography concentration
5. [creative] Identify portfolio weaknesses and blind spots
6. [creative] Suggest rebalancing actions with rationale
7. [creative] Stress test: what happens in recession / rate hike / sector crash?

Important Rules

1. Numbers must be verifiable. Every calculation must have a test or cross-check.
2. Cite sources. Every data point must reference where it came from.
3. Separate fact from opinion. Mechanical track = facts. Creative track = interpretation.
4. Disclose limitations. If data is incomplete or assumptions are made, say so explicitly.
5. No financial advice framing. Output analysis and data, not "you should buy X." Frame as "the data suggests..." not "I recommend..."

Environment Variable Override

To use analysis.json instead of prd.json with ralph.sh:

PRD_FILE=analysis.json ./scripts/ralph.sh 10