Finance & Investment Banking earnings-analysismulti-agentvalue-investingwechat-publishingfinancial-rigoradversarial-reviewequity-research

Earnings Team: Four Masters Parallel Analysis + WeChat Publishing

Read earnings with four value-investing masters in parallel, then produce a publication-ready WeChat article.

FollowSkills review · FSRS-2.0
Not recommended
49/ 100 5-point scale 2.5 / 5
1 2 3 4 5 6
1Trust12 / 25 · 2.4/5

Evidence: The skill instructs running `date` to confirm date, uses `python3 tools/financial_rigor.py` for cross-validation, and emphasizes data accuracy. It uses Agents for parallel research but lacks detailed permission control or user confirmation steps. SECURITY.md provides vulnerability reporting, but skill-specific permissions are not detailed. Deductions: No explicit user confirmation, dependence on external services (SEC EDGAR, etc.) without isolation measures, and unclear sensitive data handling. Overall risks are visible but permissions and confirmation mechanisms are incomplete.

2Reliability8 / 20 · 2.0/5

Evidence: The skill has a clear three-phase workflow and references tools like financial_rigor.py and report_audit.py, which have test files covering key paths, indicating tool reliability. However, the skill itself has no tests, and the workflow depends on real-time agent research and data fetching, which cannot be statically verified. Deductions: Key paths (data acquisition, agent collaboration) are untested, and error handling (e.g., data unavailability) has ratings but no concrete fallback.

3Adaptability10 / 15 · 3.3/5

Evidence: The skill accepts input format like '腾讯 2025Q4' with examples. Clear audience (value investors) and scenario (earnings deep-dive). It distinguishes from other skills, defining non-fit boundaries. Chinese support is good, but core data sources (SEC EDGAR, Seeking Alpha) may be unreachable from mainland China, with no alternative provided. Deductions: External service accessibility issue for mainland users, insufficient environment adaptation.

4Convention10 / 15 · 3.3/5

Evidence: Well-structured documentation with design philosophy, workflows, output file structure. README, LICENSE (MIT), SECURITY.md present. Installation and usage instructions provided. Skill notes its generation source and uses git for versioning. Deductions: No changelog or version number, maintenance responsibility not explicitly stated in the skill itself.

5Effectiveness5 / 15 · 1.7/5

Evidence: Detailed workflow and output templates aim to produce publishable WeChat articles, with a data audit (pass/reject) step, suggesting some quality control. However, static review cannot verify actual output quality. Deductions: No representative outputs or third-party execution evidence, value proposition unverified, marginal benefit lacks evidence.

6Verifiability4 / 10 · 2.0/5

Evidence: Repository includes test files for tools (test_financial_rigor.py, test_report_audit.py) covering multiple paths with detailed descriptions, which are reproducible unit tests. However, these test tools, not the skill's overall workflow. README shows real report examples (e.g., PDD, Tencent) but these are author claims. Deductions: Key claims (returns, report quality) come only from the author, no independent verification; skill's workflow itself is not statically verifiable.

Evidence confidence:Low Reviewed Aug 07, 2026 Reviewed revision 4ddc638fd536
The upstream repository has new commits since this review. The score still applies to the reviewed revision shown and may not cover the latest changes.
Before you use it
  • Core data acquisition relies on overseas services (e.g., SEC EDGAR, Seeking Alpha, Futu securities) that may be unreachable from mainland China networks; consider alternative data sources or offline solutions.
  • The skill uses parallel agents but lacks explicit least-privilege or user confirmation mechanisms; run in a trusted environment and handle sensitive data carefully.
  • Performance claims (e.g., 2024 +69.29% returns) are self-reported and not independently verified; do not assume skill effectiveness from these figures.
Review evidence [1][2][3][4][5][6]
See the full review method →

What does this skill do, and when should you use it?

This skill is designed for deep earnings analysis of important companies. It breaks down the methodologies of four value-investing masters (Duan Yongping, Warren Buffett, Charlie Munger, Li Lu) into four parallel agents, each reading the same report from a different lens: business essence, financial quality, competitive dynamics, and risk signals. A team lead synthesizes the findings, then an editor rewrites the research into a WeChat article and a reader-reviewer critiques it before finalization. Built-in data audit gates require cross-checked financial data and exact arithmetic via tools.

Collects primary materials: earnings reports (IR pages, SEC EDGAR, HKEX, CNINFO), earnings call transcripts, and management shareholder letters. Launches four parallel research agents (Duan Yongping on business essence, Buffett on financial quality, Munger on competition, Li Lu on risk signals), each producing its own analysis. The Team Lead then synthesizes a structured research report, identifying consensus and contradictions across the four perspectives. Next, an editor Agent rewrites the report into a WeChat-style article, and a reader-reviewer Agent provides feedback for final revision. Throughout, it calls python3 tools/financial_rigor.py for market-cap verification, valuation checks, cross-validation, and three-scenario analysis, and finally python3 tools/report_audit.py to audit the final article for release. Output files are written to reports/{company}/.

  1. When a core holding releases quarterly/annual results and you need deep multi-perspective analysis rather than a quick single-agent review.
  2. For financial bloggers or WeChat publishers who need a standardized pipeline from research to polished article during earnings season.
  3. When an investment team wants a consistent methodology to analyze multiple companies' earnings for horizontal comparison (requires uniform scoring standards).
  4. When an analyst needs to cross-verify key financial data and run precise valuations to avoid LLM arithmetic mistakes.
  5. When an investor wants systematic tracking of management promises and identification of hidden risk signals in filings.

What are this skill's strengths and limitations?

Pros
  • Forces four perspectives to run in parallel and challenge each other, surfacing real conflicts and consensus instead of a single viewpoint.
  • Built-in financial rigor tools (exact decimal arithmetic, multi-source cross-validation, audit gates) reduce LLM arithmetic error risk.
  • Complete publishing pipeline: research → editing → reader review → finalization, producing a ready-to-publish WeChat article and saving drafting time.
  • Data availability grading (A/B/C) aligns analysis depth with data quality, mitigating hallucination in uncertain contexts.
  • Fixed report structure enables horizontal comparison across companies or longitudinal tracking of the same company over time.
Limitations
  • Core workflow relies on multi-agent parallel orchestration, leading to high token consumption and non-trivial cost.
  • Heavily dependent on Claude/Codex Agent capabilities; requires adaptation for bare model APIs.
  • No test suite; correctness of the tools (financial_rigor.py, report_audit.py) is not verified.
  • Does not include real-time data feeds for all markets (US, HK, A-shares); manual collection of primary documents is required.
  • Output is an article, not investment advice; may not suit users who just want a quick verdict.

How do you install this skill?

This skill is part of the ai-berkshire repository (codex-skills/earnings-team/SKILL.md). Clone it first: git clone https://github.com/xbtlin/ai-berkshire.git. The repo maintains three entry points: skills/*.md, codex-skills/*/SKILL.md, and codex-prompts/*.md. Installation varies by client: Claude Code users can run ./scripts/install-claude-commands.sh (macOS/Linux) or install-claude-commands.bat (Windows); Codex users can run ./scripts/install-codex-skills.sh to generate skills into ~/.codex/skills and optionally install slash prompts. The SKILL.md does not provide a standalone install command for this specific skill; it relies on the repo's scripts.

How do you use this skill?

In Claude Code, invoke directly with /earnings-team Company Quarter, e.g., /earnings-team Tencent 2025Q4. In Codex, after installation and restart, describe the task in natural language, e.g., "use earnings-team to analyze PDD 2025 annual report." The skill will launch four background agents in parallel to gather and analyze data. It runs date to confirm today's date as the data cutoff baseline and calls Python tools for verification. Final outputs are written to reports/{company}/, including the final WeChat article, research draft, four perspective analyses, and reader review.

How does this skill compare with similar options?

The SKILL.md explicitly compares with sibling skills /earnings-review (single-agent quick earnings review) and /investment-team (four-agent full company research). This skill sits between them, targeting deep analysis plus publication for critical earnings reports.

FAQ

How much token cost does this skill incur?
It launches four research agents in parallel, each independently searching the web and cross-validating data, plus an editor and reader reviewer. Token usage is significantly higher than the single-agent '/earnings-review'. It suits key earnings; for quick daily scans, consider lighter skills.
Can it run without Claude Code or Codex agent features?
It depends on multi-agent parallel orchestration (Task/Agent). On a bare model API or plain CLI, you would need to implement agent scheduling yourself or fall back to serial execution, which may lose parallelism efficiency. Recommended to use a client with sub-agent support.
How is data source reliability ensured?
The skill mandates using primary sources (company IR pages, SEC EDGAR, HKEX, CNINFO) and grades data availability as A/B/C, which affects analysis depth. It also requires at least two independent sources for key data and uses financial rigor tools to verify calculations, avoiding LLM arithmetic errors.
Is the output investment advice or an article?
It outputs a research process and a WeChat article including scores, tables, and actionable suggestions, but the overall purpose is content production rather than direct investment instruction. Final investment decisions are up to the reader; the repo includes a disclaimer.

More skills from this repository

All from xbtlin/ai-berkshire

Related skills