convez.ai
SkillFinanceiroIntermediário
xvary-research

XVARY Stock Analysis Skill

xvary-research/claude-code-stock-analysis-skill

Análise de ações com tese e score, SEC EDGAR + mercado, sem custo

Ver no GitHub
BomPluginScore · de 0 a 100

Skill para Claude Code que realiza análise fundamentalista de ações com dados do SEC EDGAR e mercado. Gera relatórios com teses de investimento, scores de convicção e critérios de risco em menos de 2 minutos, usando dados públicos gratuitos. Inclui comandos /analyze, /score e /compare para análise institucional de qualidade.

Instalação

$npx convez-ai xvary-research/claude-code-stock-analysis-skill

1 comando no terminal: detecta o formato do repositório (marketplace ou plugin avulso) e instala via Claude Code. Requer Node 18+ e o Claude Code instalado.

Métricas

Estrelas
21
Forks
6
Acompanhando
0
Saúde
50%
Último commit
há 8 meses
Versão
—
Issues abertas
0
Contribuidores
1

Casos de uso

  • Gerar relatórios de análise fundamentalista rápidos para ações
  • Comparar múltiplas ações com critérios quantitativos e qualitativos
  • Extrair dados financeiros de SEC EDGAR para backtesting
  • Obter scores de convicção e critérios de risco para decisão de investimento

Veredito

Ferramenta sólida para análise de ações que vai além de ratios simples, entregando teses estruturadas com scores de convicção. Recomendado para quem precisa de análise fundamentalista rápida e documentada, mas exige conhecimento de mercado para interpretar os resultados corretamente.

README

A tradução para o português deste README ainda está na fila.

XVARY Stock Research

License: MIT Python Claude Code Skill xvary.com

Type /analyze NVDA in Claude Code and get a thesis-driven equity report with conviction scoring, kill criteria, and an EDGAR-backed financial snapshot -- in under two minutes, from public data, for free.

This is the open skill layer of XVARY Research. We run a 21-stage pipeline to produce institutional-depth stock analysis. This repo gives you the methodology framework, the data tools, and the scoring models. The full 22-section deep dives live at xvary.com.

We recognize Linux Do community

From the live site (NVDA deep dive)

Captured from xvary.com/stock/nvda/deep-dive/ — same product surface the skill is designed to complement.

Regenerate these assets: npm install && npm run screenshots:nvda (see scripts/screenshot_xvary_nvda.mjs).

What you get that raw data tools don't

  • A verdict, not a spreadsheet -- "Constructive at 74/100 conviction"
  • Named kill criteria -- exactly what would break the thesis
  • Composite scores across four dimensions, not just price ratios
  • Analysis that reads like a research desk, not a terminal dump

Quick Start

Clone and verify

git clone git@github.com:xvary-research/claude-code-stock-analysis-skill.git
cd claude-code-stock-analysis-skill
python3 tools/edgar.py AAPL    # pulls SEC XBRL data
python3 tools/market.py AAPL   # pulls price + ratios

XVARY monorepo: if you already have the full workspace, this skill lives at 9. Marketing/xvary skill/.

Install as a Claude Code skill

mkdir -p ~/.claude/skills/xvary-stock-research
cp SKILL.md ~/.claude/skills/xvary-stock-research/SKILL.md
cp -R references tools examples ~/.claude/skills/xvary-stock-research/

Or skip the install entirely -- open Claude Code in this repo and say:

Read SKILL.md and run /analyze AAPL

Plugin marketplace (same folder): open this directory as the marketplace root (it contains .claude-plugin/marketplace.json), then in Claude Code run /plugin marketplace add . and /plugin install xvary-stock-research@xvary-research. Validate with claude plugin validate . before you tag a release.

Public GitHub checkout: /plugin marketplace add xvary-research/claude-code-stock-analysis-skill then /plugin install xvary-stock-research.

Commands

CommandWhat it does
/analyze {ticker}1-page thesis + scorecard + risks + EDGAR-backed financial snapshot
/score {ticker}Momentum, Stability, Financial Health, and Upside Estimate
/compare {A} vs {B}Side-by-side score, thesis, and risk differential

Example: /analyze NVDA

Full example: examples/nvda-analysis.md

Verdict: CONSTRUCTIVE (Conviction 74/100)

┌─────────────────┬───────┬──────────────────────────────────────────────┐
│ Score           │ Value │ Read                                         │
├─────────────────┼───────┼──────────────────────────────────────────────┤
│ Momentum        │  88   │ Demand + operating leverage remain strong    │
│ Stability       │  70   │ Strong execution, non-zero cyclicality risk  │
│ Financial Health│  84   │ Robust balance sheet vs obligations          │
│ Upside Estimate │  64   │ Positive setup, expectations already high    │
└─────────────────┴───────┴──────────────────────────────────────────────┘

Thesis pillars:
  1. AI infrastructure spend durability
  2. CUDA ecosystem lock-in + pricing power
  3. Operating leverage on incremental revenue
  4. Balance-sheet capacity through cycle volatility

Kill criteria: hyperscaler capex pullback + export control
escalation + gross-margin break with rising capex intensity

Financial snapshot (public, 10-K 2026-01-25):
  Revenue $215.9B · Net income $120.1B · OCF $102.7B
  Assets $206.8B / Liabilities $49.5B
  Price $172.70 · Market cap ~$4.20T · P/E 35.23 · Beta 2.34

This is the free layer. The full pipeline produces 22-section reports with DCF models, competitive matrices, risk scenarios, and adversarial challenge gates.

Open the live NVDA report: xvary.com/stock/nvda/deep-dive/ (free preview; full tabs with subscription)

How this compares

Raw data MCPsScreener APIsThis repo
FreeVariesUsually noYes
Thesis with verdictNoNoYes
Named kill criteriaNoNoYes
Composite scoring (4 dimensions)NoPartialYes
Works locally, no API keyN/ANoYes
Methodology publishedN/ANoYes

Architecture

Skill layer (this repo): public data in → methodology + scoring → structured output → link out to full deep dives on xvary.com.

Claude Code plugin bundle (ships in this folder)

PathRole
.claude-plugin/marketplace.jsonMarketplace catalog xvary-research — users run /plugin marketplace add from this directory
plugins/xvary-stock-research/Plugin wrapper; skills/xvary-stock-research/ symlinks to root SKILL.md, references/, tools/, examples/ so there is a single source tree

Monorepo checkout: open a terminal in 9. Marketing/xvary skill/ (this folder), then run /plugin marketplace add . in Claude Code — same as a standalone claude-code-stock-analysis-skill clone where this folder is the repo root.

flowchart LR
    A["/analyze ticker"] --> B["tools/edgar.py\nSEC XBRL + filings"]
    A --> C["tools/market.py\nYahoo → Finviz → Stooq"]
    B --> D["Methodology spine\n+ scoring refs"]
    C --> D
    D --> E["Structured analysis\n+ kill criteria"]
    E --> F["xvary.com deep dive"]

21-stage research spine + finalize (operational DAG)

Same DAG as references/methodology.md: 22 nodes in code (research spine + finalize). Edges show real control flow—parallel paths merge at phase_b, quality_gate, and completion_loop.

flowchart TB
  subgraph P1["① Intake & evidence integrity"]
    s1[directive_selection] --> s2[phase_a] --> s3[data_quality_gate] --> s4[evidence_gap_analysis]
  end

  subgraph P2["② Hypothesis & quant scaffolding"]
    s5[kvd_hypothesis]
    s6[pane_selection] --> s7[quant_foundation] --> s8[model_quality_gate]
  end

  subgraph P3["③ Deep enrichment & triangulation"]
    s9[phase_b] --> s10[triangulation] --> s11[pillar_discovery]
  end

  subgraph P4["④ Parallel synthesis & QA"]
    s12[phase_c]
    s13[why_tree]
    s14[quality_gate]
  end

  subgraph P5["⑤ Adversarial challenge & conviction"]
    s15[challenge] --> s16[synthesis]
  end

  subgraph P6["⑥ Audit, packaging & release control"]
    s17[audit] --> s18[report_json]
    s19[audience_calibration]
    s20[compliance_audit]
    s21[completion_loop] --> s22[finalize]
  end

  s4 --> s5
  s4 --> s6
  s5 --> s9
  s6 --> s9
  s11 --> s12
  s11 --> s13
  s12 --> s14
  s13 --> s14
  s14 --> s15
  s16 --> s17
  s18 --> s19
  s18 --> s20
  s19 --> s21
  s20 --> s21
#StageIntent
1directive_selectionChoose sector/style evidence directives
2phase_aBaseline facts, filings, market context
3data_quality_gateBlock low-integrity factual inputs
4evidence_gap_analysisFind gaps; open targeted searches
5kvd_hypothesisCandidate key value drivers
6pane_selectionChoose report panes for company profile
7quant_foundationValuation / risk scaffolding
8model_quality_gateSanity-check model outputs
9phase_bEnrichment + deeper context
10triangulationCross-check independent reasoning vectors
11pillar_discoveryWeighted thesis pillars
12phase_cModule-level synthesis (parallel)
13why_treeCausal claims + dependency chains
14quality_gateConsistency + evidence sufficiency
15challengeAdversarial test of pillars
16synthesisConviction, variant view, scenarios
17auditMulti-role verification + follow-ups
18report_jsonStructured report payload
19audience_calibrationReadability + decision speed
20compliance_auditMethodology + policy checks
21completion_loopRepair sparse / inconsistent sections
22finalizeRelease gating + artifact finalization

XVARY Scores

Definitions: references/scoring.md

ScoreWhat it measures
MomentumDirection and persistence of operating + market trajectory
StabilityEarnings durability, cyclicality resilience, variance control
Financial HealthBalance-sheet strength and cash-flow solvency
Upside EstimateAsymmetry vs. current implied expectations

Methodology (Published Framework)

Full framework: references/methodology.md

What's published:

  • 21-stage research DAG with stage purposes
  • 23 module map and what each module produces
  • Quality gate names and validation criteria
  • Conviction scoring and variant-perception philosophy
  • Kill-file risk discipline

What stays proprietary:

  • LLM prompts and chain-of-thought templates
  • Threshold tables and scoring formulas
  • Triangulation and convergence algorithms
  • Sector-specific prompt libraries

Data Sources

SourceAccessUsed for
SEC EDGARPublic, freeCompany facts (XBRL) + filing metadata
Yahoo FinanceNo API keyQuote, valuation, ratio fields
Finviz / StooqFallbackResilience when Yahoo is unavailable

EDGAR patterns: references/edgar-guide.md

Full Deep Dives

TickerLink
NVDAxvary.com/stock/nvda/deep-dive/
All coverage (3,325 names)xvary.com/discover
Methodology narrativexvary.com/methodology

Roadmap

  • MCP server for on-demand full deep dives
  • Earnings-season auto-refresh triggers
  • Additional scoring models (earnings quality, capital allocation)
  • Cursor / Windsurf / Codex skill mirrors (Claude Code marketplace ships from this folder)

Contributing

PRs welcome for:

  • EDGAR taxonomy coverage and normalization
  • Market-data fallback robustness
  • Documentation clarity and examples

License

MIT. See LICENSE.

Ver o original no GitHub

Bom para

Relacionados