XVARY Stock Research

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 [email protected]: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

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

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)

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

XVARY Scores

Definitions: references/scoring.md

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

EDGAR patterns: references/edgar-guide.md

Full Deep Dives

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.