Codex adapter note

This skill is generated from

skills/thesis-drift.md

so Claude Code and Codex users share one canonical workflow.

Treat
```
$ARGUMENTS
```
as the user's request in the current Codex thread.
When the source mentions Claude-only surfaces such as Task, Agent, WebSearch, Bash, Read, or Write, use the closest Codex capability available in this session: subagents when available, web search when needed, shell commands for local tools, and normal file edits for workspace files.
Use shared project tools from
```
tools/
```
in this repository. Prefer running commands from the repository root with paths like
```
python3 tools/financial_rigor.py ...
```
; if the current thread starts outside the repo, locate the actual checkout path first instead of assuming a fixed home-directory path.
Before starting research, run the
```
date
```
command to confirm today's date; treat it as the baseline for "latest" data and state the data cutoff date in the report header. Never assume the current date from training data.
Preserve the research quality rules from
```
AGENTS.md
```
: cross-check financial data, use exact arithmetic tools for valuation/math, and clearly label uncertainty and source gaps.

Investment Thesis Drift Detection: Distinguish Between Fact Changes and Wording Changes

Perform investment thesis drift detection on $ARGUMENTS.

Supported Input Formats:

```
Company Name Old Report Path New Report Path
```
— Specify two research reports or thesis snapshots for comparison

Company Name reports/{Company Name}-thesis-old-date.md reports/{Company Name}-thesis-new-date.md

— Compare two dated thesis snapshots

```
Company Name
```
— Automatically find
```
reports/{Company Name}-thesis.md
```
and historical snapshots in the same directory; if no baseline exists, proceed to missing baseline handling

"When the facts change, I change my mind. What do you do?" — John Maynard Keynes

"Stock price fluctuations are not thesis drift; only fact changes are." — AI Berkshire

Design Philosophy

The hardest part of long-term holding is not reading the news every day, but distinguishing three things:

Fact Changes: Verifiable changes in revenue, profit margins, competitive landscape, management behavior, and capital allocation
Price Changes: Changes in market sentiment or valuation multiples, but the business itself remains unchanged
Wording Changes: Two reports express differently, but the underlying evidence and judgments remain unchanged

The goal of investment thesis drift detection is: Only acknowledge thesis changes when evidence changes. Do not create drift just because the report is rewritten, nor misjudge fundamentals because of stock price fluctuations.

This Skill relies on structured dimensions output by

/thesis-tracker

: core assumption list, red line list, valuation anchor, tracking record table. If these structures are not available, first complete the baseline before performing drift detection.

Execution Process

Step 1: Determine Operation Mode

Parse

$ARGUMENTS

If two report paths are provided → Enter Specified Report Comparison mode
If only the company name is provided → Find
```
reports/{Company Name}-thesis.md
```
and historical snapshots, enter Automatic Snapshot Comparison mode
If only one report is found or no historical baseline exists → Enter Missing Baseline Handling mode
If the two reports are not for the same company → Stop and ask the user to confirm, do not make cross-company drift judgments

Mode A: Specified Report Comparison

A1: Read and Validate Two Reports

Read the old and new reports, extract:

Report date, company name, stock symbol
Core thesis (5 sentences)
Core assumption list
Red line list
Valuation anchor
Tracking record table
Management quality judgment
Competitive moat judgment
Current recommended action (Buy / Hold / Watch / Reduce / Exit)

If the report lacks key structures, mark "Structure Missing" first, but still try to extract evidence from the text as much as possible; mark dimensions that cannot be extracted as "Cannot Judge", do not fabricate conclusions.

A2: Evidence Normalization

Organize the factual evidence from the two reports into the same table:

Dimension	Old Report Evidence	New Report Evidence	Data Source	Verifiable
Valuation Anchor
Core Assumptions
Red Lines
Management Quality
Competitive Moat

Only compare evidence, not writing style. If the old and new reports are just synonymous rewrites, order changes, or tone changes, but the factual data and judgment thresholds have not changed, judge as Unchanged.

A3: Numerical and Valuation Verification

All numerical changes must be accurately calculated using

tools/financial_rigor.py

, LLM mental calculation is prohibited:

bash

python3 tools/financial_rigor.py verify-valuation \
  --price {Current Price} \
  --eps {EPS} \
  --bvps {Book Value Per Share} \
  --fcf-per-share {Free Cash Flow Per Share}

To calculate market capitalization, percentage changes, target price differences, or scenario valuations, use:

bash

python3 tools/financial_rigor.py verify-market-cap --price {Price} --shares {Shares Outstanding} --reported {Reported Market Cap} --currency {Currency}
python3 tools/financial_rigor.py cross-validate --field {Field} --values '{JSON}' --unit {Unit}
python3 tools/financial_rigor.py three-scenario --price {Price} --eps {EPS} --shares {Shares Outstanding (100M)} --growth {Optimistic} {Neutral} {Pessimistic} --pe {Optimistic PE} {Neutral PE} {Pessimistic PE}
python3 tools/financial_rigor.py calc --expr '{Precise Formula}'

Key financial data must be cross-validated with at least two independent sources. Numbers with insufficient sources, inconsistent calibers, or unreviewable must be marked as "Low Confidence / To Be Verified".

A4: Drift Judgment by Dimension

Use the following fixed dimensions, do not add or remove temporarily:

Dimension	Judgment Focus	Improved	Unchanged	Weakened
Valuation Anchor	Intrinsic value, PE/PB/FCF Yield, margin of safety, target price range	Margin of safety expands or intrinsic value is revised upward and verified by tools	Valuation range and margin of safety have no substantial changes	Margin of safety narrows, intrinsic value is revised downward, or valuation assumptions fail
Core Assumption List	Revenue growth rate, profit margin, cash flow, verifiable assumptions such as users/orders/capacity	More assumptions are strengthened by new evidence	Assumption status is basically consistent with evidence	Assumptions are marginally weakened, damaged, or broken
Red Line List	Integrity, regulation, business decline, competitive breakthrough, abnormal management actions	Original red line risks are eliminated or significantly reduced	Not triggered and risk level remains unchanged	Red line is triggered or trigger probability increases
Management Quality	Integrity, capital allocation, share repurchases and dividends, execution, shareholder friendliness	New behaviors increase trust	Behaviors continue old judgments	Behaviors damage trust or capital allocation deteriorates
Competitive Moat	Market share, pricing power, network effects, cost advantages, substitution threats	Moat widens or competitive advantages are verified	Competitive landscape has no substantial changes	Moat is weakened or competitors break through

Each dimension can only give three types of conclusions: Improved / Unchanged / Weakened.

A5: Evidence-Driven Rules

Each non-Unchanged conclusion must cite specific new evidence that caused the change:

Financial statement items: e.g., revenue growth rate, gross profit margin, operating cash flow, repurchase amount, net cash
Regulatory disclosures: e.g., 10-K/20-F, annual report, interim report, HKEX announcement, SEC filing
News events: e.g., management changes, regulatory penalties, major customer loss, competitor breakthroughs
Price and valuation: Must explain whether this is "valuation change" or "fundamental change", cannot confuse the two

If no evidence can be found to explain the change, must judge as Unchanged or Cannot Judge, cannot infer drift from wording differences.

A6: Output Drift Report

Report Structure

I. Comparison Objects and Time Span
II. Overall Conclusion: Is the Thesis Drifting?
III. Dimension Drift Table
IV. Evidence Difference Details
V. Valuation and Numerical Verification
VI. Recommended Action Migration
VII. Uncertain Items and Required Supplementary Sources
VIII. Key Focus for Next Tracking

Dimension Drift Table

Dimension	Drift Direction	Confidence
Valuation Anchor	Improved / Unchanged / Weakened	High/Medium/Low
Core Assumption List	Improved / Unchanged / Weakened	High/Medium/Low
Red Line List	Improved / Unchanged / Weakened	High/Medium/Low
Management Quality	Improved / Unchanged / Weakened	High/Medium/Low
Competitive Moat	Improved / Unchanged / Weakened	High/Medium/Low

For Unchanged rows, write
—
in Trigger Evidence, do not fabricate evidence for filling the table.

Overall Conclusion Must Answer

Is the thesis drifting? No Drift / Positive Drift / Negative Drift / Insufficient Evidence to Judge
Where does the drift come from? Valuation / Fundamentals / Management / Competitive Landscape / Red Line Event
Is it a fact change or a price change? Clearly explain separately
How to migrate recommended actions? e.g., Watch → Buy, Buy → Hold, Hold → Reduce, Reduce → Exit
What evidence is needed next? Next financial report / Regulatory disclosure / Management explanation / Competitor data

Mode B: Automatic Snapshot Comparison

B1: Find Snapshots

Look in

reports/

for:

```
reports/{Company Name}-thesis.md
```
```
reports/{Company Name}-thesis-*.md
```
Reports containing "thesis", "论文", or "追踪" in the
```
reports/{Company Name}/
```
directory

Select the earliest file with a complete structure as the old report, and the latest file as the new report. If the user specifies a date, follow the user's specification.

B2: Prevent Incorrect Pairing

Before comparison, must confirm:

Company name or stock symbol is consistent
Report dates are different
Both reports contain extractable thesis structures or research conclusions

If it cannot be confirmed that they are for the same company, stop and ask the user to provide a clear path.

B3: Execute Mode A

After finding two valid snapshots, fully execute Mode A.

Mode C: Missing Baseline Handling

If only one report is found or no old snapshots are found:

Clearly state: Missing comparable historical baseline, cannot perform drift detection
Do not fabricate an old thesis based on memory or market impressions
Guide the user to first establish a structured baseline using
```
/thesis-tracker {Company Name} Establish Thesis
```
If the current report is complete enough, suggest saving it as
```
reports/{Company Name}-thesis.md
```
as the baseline for future drift detection

Output format:

Cannot perform thesis drift detection: Missing historical baseline.

Found:
- Current Report: {Path / Not Found}
- Historical Baseline: Not Found

Recommendations:
1. First run /thesis-tracker {Company Name} Establish Thesis
2. After the next financial report or major event, run /thesis-drift {Company Name} Old Report New Report

Key Principles

Evidence takes precedence over wording — Synonymous rewrites are not drift; only changes in factual evidence are drift
Fundamentals take precedence over stock price — Stock price fluctuations only affect valuation anchors, do not automatically change business quality
Numerical values must be verified — All percentages, valuation multiples, and target price differences must use
```
tools/financial_rigor.py
```
Mark uncertainty when uncertain — Do not force a judgment when sources are missing, calibers are inconsistent, or unreviewable
Red lines are handled separately — Red line triggering takes priority over cheap valuation, cannot be covered by low PE
Output must be reproducible — Each Improved / Weakened conclusion must be traceable to specific evidence

thesis-drift

NPX Install

Tags

SKILL.md Content (Chinese)