ag-capital-analyst

ai-investment-analyst

A Claude Code plugin marketplace for AI-driven investment research.

Two skills are available:

• ag-capital-analyst — multi-agent investment analysis team producing buy/sell/hold recommendations with position sizing
• ppm-risk-review — family-office-grade due diligence on private-fund offering documents (PPM/OM), producing a CIO risk memo, sponsor diligence questionnaire, and plain-English summary

What's in this repo

.
├── .claude-plugin/
│   └── marketplace.json          # Marketplace manifest
├── ag-capital-analyst/           # ag-capital-analyst plugin
│   ├── .claude-plugin/plugin.json
│   └── skills/ag-capital-analyst/
│       ├── SKILL.md              # Orchestrator (workflow, output, registry)
│       └── references/           # Per-role definitions (8 analyst/reviewer roles)
├── ppm-risk-review/              # ppm-risk-review skill
│   ├── .claude-plugin/plugin.json
│   ├── SKILL.md
│   ├── references/               # Rubric, diligence sources, script guide
│   └── scripts/                  # Node.js Word document build scripts
├── ag-capital-analyst.skill      # Installable .skill bundle (Claude.ai)
├── ppm-risk-review.skill         # Installable .skill bundle (Claude.ai)
└── ag-analysis/                  # Sample outputs (DBMF, FTLS)

The ag-capital-analyst plugin

Spawns six analyst subagents in parallel, each pulling live financial data, then routes their signals through a Risk Manager and an independent QC reviewer before synthesizing a final recommendation.

Analyst	Lens
Buffett	Moat, intrinsic value, margin of safety
Growth	TAM, revenue trajectory, disruption potential
Technical	Price action, momentum, support/resistance
Fundamentals	Financial statements, ratios, valuation multiples
Sentiment	News flow, analyst opinion, short interest, social signals
Valuation	Quantitative DCF + comps, triangulated fair value

The Portfolio Manager (the orchestrator) consolidates the six signal reports, hands them to a Risk Manager subagent for risk assessment (volatility-adjusted position sizing, bull/base/bear scenarios, catalyst calendar), runs an independent QC review pass, and produces a final deliverable with a recommendation, confidence score, price target, position size, time horizon, and key risks — written for an inexperienced investor. Each run also updates a per-ticker thesis note with an append-only conviction log. Role definitions live in skills/ag-capital-analyst/references/.

Trigger phrases

"analyze AAPL", "should I buy NVDA", "find undervalued semiconductor companies", "evaluate this ETF"

Output

Each run produces a folder named {TICKER} - {Security Name}/ containing:

• buffett-signal.md, growth-signal.md, technical-signal.md, fundamentals-signal.md, sentiment-signal.md, valuation-signal.md
• all-signals.md — consolidated signals
• risk-assessment.md — Risk Manager output
• qc-review.md — independent QC review (PASS/FIX)
• final-recommendation.md — final synthesis

Plus a living _theses/{TICKER} - {Security Name}.md note that accrues a dated conviction log across runs.

In Claude Code, reports are written directly into the user's Obsidian vault so they appear alongside other investment notes. In the Anthropic Agent Skills runtime (Claude.ai), they land under /home/user/workspace/ag-analysis. See ag-analysis/ in this repo for example outputs (DBMF, FTLS).

Installation (Claude Code)

Add this repo as a marketplace, then install the plugin(s) you want:

/plugin marketplace add arkadiyg/ai-investment-analyst
/plugin install ag-capital-analyst@ai-investment-analyst
/plugin install ppm-risk-review@ai-investment-analyst

Once installed, skills auto-activate on relevant prompts — ticker/sector analysis for ag-capital-analyst; PPM/offering document review for ppm-risk-review.

Recommended data sources

The plugin is tool-agnostic — analysts use whatever live-data tools the runtime exposes (WebSearch, WebFetch, or MCP servers). It will run with web search alone, but adding a finance data source gives you structured fundamentals, real-time quotes, filings, and ownership data instead of scraped prose. Quality jumps noticeably for the Fundamentals and Technical analysts.

Source	What it adds
FMP (Financial Modeling Prep)	Real-time quotes, financial statements, ratios, DCF, analyst estimates, earnings transcripts, 13F holdings, insider trades, technical indicators
Equibles	SEC filings (with line-level reads), short interest, institutional holdings, congressional/insider trades, CFTC positioning, economic indicators
Perplexity	Higher-quality web research and sentiment, with citations