What Is the Thesis Tracker Skill for Post-Investment Analysis?

The Thesis Tracker skill implements a disciplined monitoring system that treats investment theses as sets of verifiable hypotheses and red-line triggers, automatically calculating a health score and flagging status changes to remove emotion from portfolio decisions.

The thesis tracker skill for post-investment analysis provides a systematic framework for monitoring whether an investment thesis remains valid after purchase. Implemented in the xbtlin/ai-berkshire repository, this skill enforces Warren Buffett's principle of writing sell-conditions before buying, ensuring that post-investment decisions are driven by evidence rather than market noise.

Core Concept and Philosophy

The skill operates on the principle that every investment thesis should be treated as a collection of verifiable hypotheses with predefined red-line triggers. Rather than reacting to daily price movements, the system conducts quarterly re-evaluations (or immediate checks when major events occur) to determine if the original investment logic still holds.

This design mirrors Buffett's "write the sell-conditions before you buy" approach, as documented in skills/thesis-tracker.md. By establishing concrete validation criteria upfront, the skill ensures that discipline, transparency, and actionable triggers drive decisions rather than emotional reactions to market volatility.

The Two-Mode Workflow Architecture

The skill splits operation into two distinct modes that trigger automatically based on the presence of existing thesis documentation.

Mode A: Establishing the Investment Thesis

When no existing reports/{company_name}-thesis.md file is found, the skill initiates thesis creation mode:

1. Data Gathering: Pulls price, valuation, and latest financial data via the WebSearch tool
2. Validation: Runs valuation verification through tools/financial_rigor.py
3. Documentation: Writes a concise 5-sentence core thesis, hypothesis table, red-line table, and valuation anchors to reports/{company_name}-thesis.md

Mode B: Quarterly Tracking and Validation

When a thesis file exists, the skill executes tracking mode:

1. Load Context: Reads the existing thesis, hypothesis tables, red-line conditions, and prior check records
2. Fresh Data Pull: Retrieves latest earnings, major events, stock price, and insider trades
3. Hypothesis Evaluation: Re-evaluates each hypothesis, flagging status changes using the system:
   • 🟢 成立 (Valid)
   • 🟡 弱化 (Weakened)
   • 🔴 受损 (Damaged)
   • ⚫ 破裂 (Broken)
4. Red-Line Scan: Checks all red-line conditions; any trigger forces an immediate recommendation
5. Health Calculation: Updates valuation anchors and computes a quantitative health score
6. Persistence: Appends results to the thesis file and generates a structured action recommendation

The Health Score Algorithm

The skill employs an explicit scoring formula to quantify thesis integrity, defined in lines 776-782 of skills/thesis-tracker.md:

健康度 = 10 − (⚫破裂假设数×3) − (🔴受损假设数×2) − (🟡弱化假设数×1) − (红线触发数×5)

This formula generates a maximum score of 10, penalizing broken hypotheses most severely (-3 points each), followed by damaged hypotheses (-2), weakened hypotheses (-1), and red-line triggers (-5 each). The resulting score provides an objective metric for comparing portfolio health across holdings during portfolio-review operations.

Implementation Details and File Structure

The skill integrates with the broader ai-berkshire architecture through specific file paths and tooling:

Entry Point: Invoked via the slash command /thesis-tracker (registered in README.md and README_EN.md)

Data Sources:

• WebSearch tool for market data
• tools/financial_rigor.py for valuation verification

Persistence Layer:

• Canonical thesis documents stored in reports/{company_name}-thesis.md
• Markdown format enables human readability while maintaining structured data

Codex Compatibility:

• Auto-generated version at codex-skills/thesis-tracker/SKILL.md ensures identical behavior for Claude Code and Codex users
• Maintains parity with the canonical source in skills/thesis-tracker.md

Integration Points:

• investment-research and news-pulse skills can reference thesis files for context
• portfolio-review skill aggregates health scores across holdings

Practical Usage Examples

The skill accepts specific command syntax to trigger different operational modes:

# Establish new thesis (first use)

/thesis-tracker 拼多多

# Force re-establishment (when rewriting thesis)

/thesis-tracker 拼多多 建立论文

# Quarterly check (auto-detects existing thesis)

/thesis-tracker 拼多多 季度检查

Each execution returns a structured markdown report containing:

• Health score (0-10 scale)
• Updated hypothesis table with current status flags
• Red-line status (triggered or clear)
• Valuation comparison (purchase price vs. current)
• Action recommendation (加仓/Add, 持有/Hold, 减仓/Reduce, 清仓/Exit)

Summary

• The thesis tracker skill for post-investment analysis transforms qualitative investment theses into quantitative, verifiable frameworks stored in reports/{company_name}-thesis.md
• Dual-mode operation automatically handles initial thesis creation (Mode A) and ongoing quarterly validation (Mode B) based on file existence checks
• Objective health scoring uses the formula 健康度 = 10 − (⚫×3) − (🔴×2) − (🟡×1) − (红线×5) to eliminate subjective bias in portfolio reviews
• Red-line triggers provide immediate exit signals when fundamental conditions break, enforcing the discipline of pre-defined sell criteria
• Cross-skill integration allows portfolio-review to aggregate health scores while leveraging tools/financial_rigor.py for valuation anchoring

Frequently Asked Questions

How does the health score calculation work in the Thesis Tracker?

The health score uses a subtraction-based algorithm where 10 is the maximum possible score. The system subtracts 3 points for each broken hypothesis (⚫), 2 points for each damaged hypothesis (🔴), 1 point for each weakened hypothesis (🟡), and 5 points for any triggered red-line condition. This weighted approach prioritizes structural thesis breaks over gradual deterioration, ensuring that critical failures trigger immediate action recommendations.

What determines whether Mode A or Mode B executes?

The skill automatically detects the presence of a thesis file at reports/{company_name}-thesis.md. If no file exists, Mode A (Establish Thesis) executes to create the initial hypothesis tables and red-line conditions. If the file exists, Mode B (Tracking Check) loads the existing thesis and evaluates current data against the stored hypotheses. Users can force Mode A execution by appending "建立论文" (establish thesis) to the command.

Where are the thesis documents stored and what format do they use?

All thesis documents persist as markdown files in the reports/ directory, named {company_name}-thesis.md. This format stores the core 5-sentence thesis, hypothesis tables with status flags, red-line trigger definitions, valuation anchors, and historical check records. The markdown structure enables both human readability and programmatic parsing by other skills like portfolio-review and investment-research.

How does this system prevent emotional decision-making?

By requiring investors to define specific red-line triggers and verifiable hypotheses before purchase (Mode A), the skill creates a contractual framework that governs post-investment decisions. When Mode B executes, it evaluates objective data against these pre-established criteria rather than current price action or market sentiment. The quantitative health score and forced recommendations (加仓/持有/减仓/清仓) remove discretion during periods of volatility, ensuring that exits occur when the thesis breaks rather than when fear peaks.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}