Earnings Review: Deep Dive into Primary Sources
Read earnings reports like Buffett and Li Lu — primary sources only, no second-hand research, with actionable investment conclusions.
Evidence: SKILL.md uses shared tools, no malicious code or permission abuse found; SECURITY.md provides vulnerability reporting, MIT license. Deductions: no explicit confirmation mechanism, tool permission scope unclear, supply chain risk from external data sources unmitigated; falls short of maximum.
Evidence: Clear process, specific commands and tools; tests exist but not verified. Deductions: Static review cannot execute, test coverage insufficient, error handling incomplete; capped at 7.
Evidence: Clearly states use cases and input formats, mentions Chinese data sources. Deductions: Inadequate discussion of mainland-China network reachability, fallback for inaccessible overseas services unclear; core function depends on overseas services.
Evidence: SKILL.md has clear structure and generation note, MIT license. Deductions: No version/changelog, known limitations not fully disclosed, maintenance responsibility unclear; deductions applied.
Evidence: Report structure and output requirements clear, data validation steps present. Deductions: Static review cannot verify output quality, relies on external data sources, marginal value evidence insufficient; capped at 6.
Evidence: Tests exist but unexecuted, report reproducibility unverified. Deductions: Static review cannot independently verify, test coverage limited, third-party evidence insufficient; capped at 4.
- Core functionality relies on overseas data sources (SEC EDGAR, Seeking Alpha, macrotrends, etc.) which may be inaccessible from mainland China; assess alternatives.
- No explicit user confirmation mechanism is provided; tool execution may produce unintended operations. Use in a controlled environment.
- Static review cannot verify tool and process efficacy; test in real environments before production use.
- The repository's track-record screenshots (2024/2025 returns) are promotional material and not evidence of reliability or effectiveness.
What does this skill do, and when should you use it?
This skill codifies the value-investing methodologies of Buffett, Munger, Duan Yongping, and Li Lu into a structured framework for deep-diving into company earnings (annual, quarterly, and earnings call transcripts). It emphasizes extracting data from primary sources to avoid the filtering and lag of secondary information, applying rigorous financial verification, management tone analysis, and footnote mining, ultimately producing a structured report with clear investment judgments (beat/miss/in-line) and action recommendations.
The skill executes an eight-step process: 1) rates data availability as A/B/C; 2) fetches primary materials in parallel (10-K/10-Q, earnings call transcripts, shareholder letters); 3) extracts and cross-validates core financial metrics (revenue, profit, cash flow, balance sheet) using tools/financial_rigor.py for precise decimal arithmetic and unit checks; 4) reads management discussion (MD&A), analyzes tone signals, tracks commitment fulfillment, and excerpts Q&A; 5) mines footnotes for hidden information (related-party transactions, stock compensation, contingencies, accounting policy changes) and flags anomalies (e.g., receivables growing faster than revenue); 6) compares with historical data and management guidance for trend analysis; 7) outputs a fixed-template report (snapshot, top 3 changes, management tone, hidden info, key questions, thesis impact, conclusion) that answers four decision questions; 8) runs data-spot-check via tools/report_audit.py to issue a pass/reject verdict. Reports are saved to reports/{company}-earnings-{period}.md.
- An individual investor who wants to deeply understand a company's latest quarterly or annual report during earnings season, bypassing brokerage summaries.
- A value investor deciding to add, hold, or trim a position, requiring systematic scrutiny of cash flow quality and management trustworthiness.
- An analyst (buy-side or sell-side) who needs standardized earnings comparisons across multiple companies for a screen or thematic review.
- A portfolio manager evaluating whether an investment thesis still holds at quarter-end, needing an answer to 'what did this report change?'
- A long-term follower tracking whether management delivered on prior commitments, to assess reliability.
What are this skill's strengths and limitations?
- Mandates primary-source reading, avoiding the filtering and lag of secondary info, staying true to value investing principles.
- Built-in anti-error mechanisms: A/B/C data-grade rating, multi-source cross-validation, exact decimal arithmetic (prevents mental-math errors), and spot-check release gate.
- Produces structured reports with explicit investment verdicts (beat/miss/in-line) and action recommendations (add/hold/trim), directly usable for decisions.
- Integrates four masters' frameworks — business model, moat, inversion thinking, long-term certainty — for multi-angle scrutiny.
- Reproducible workflow enables fair cross-company comparisons and longitudinal tracking of the same company.
- Depends on external data sources (company IR, SEC EDGAR, HKEX, CNINFO, etc.); if primary access fails, analysis quality degrades and must be flagged.
- High token consumption due to multi-round research, multi-agent parallelism, and cross-validation; not suited for low-cost scenarios.
- Requires technical setup (Python scripts, shell commands) for full rigor, which may deter non-technical users.
- No built-in test suite; tool reliability relies on community validation and may contain undiscovered bugs.
- Designed for CN/HK/US markets; non-English/Chinese financial reports (e.g., Japan, Europe) may be harder to process from primary sources.
How do you install this skill?
Install the entire AI Berkshire repo: git clone https://github.com/xbtlin/ai-berkshire.git. Then install per client: for Codex, run ./scripts/install-codex-skills.sh (macOS/Linux) or .\scripts\install-codex-skills.bat (Windows), which generates and installs this skill (codex-skills/earnings-review/SKILL.md) into ~/.codex/skills. For Claude Code, run ./scripts/install-claude-commands.sh (or .bat).
How do you use this skill?
After installation, restart Codex and describe the task in natural language, e.g., 'Use earnings-review to analyze PDD 2025 annual report' or 'Use earnings review to analyze Tencent 2025Q4'. The skill will use the current date as baseline, fetch primary sources, run financial verification tools, and generate a report into the reports/ directory.
How does this skill compare with similar options?
Compared to generic AI investment tools (like directly asking Claude or ChatGPT), the key differentiators are: 1) mandatory primary source usage rather than secondary summaries; 2) structured verification (exact arithmetic, multi-source checks) instead of relying on LLM mental math; 3) definitive conclusions and action items rather than balanced 'on one hand...' analysis. Within this repo, unlike /earnings-team (a multi-agent parallel version with public-account publishing), this skill is the single-agent core workflow.