v1.13.0
English 繁體中文

Demo

Demo Overview

Real runs, real screenshots — 10-K deep dives (META · NVDA · AMD) and full-coverage demos (PLTR · RKLB)

Open Raw

Demos — Real Runs, Real Screenshots

These demos show InvestSkill working on real stocks. Every one was run for real on 2026-10-11, using SEC filings downloaded with the repo’s keyless helpers and the 2026-10-09 NASDAQ close. Each page goes step by step: the command, a screenshot of the actual output, what to notice in it, and the result. The complete, unedited outputs sit at the bottom of every page, so you can check that nothing was cherry-picked.

New here? Start with the Live Walkthrough (繁體中文). It runs four skills on Microsoft and explains the method in the most detail.

Educational only — not investment advice. Figures are as of the dates shown and go stale quickly.

Pick a demo

Demo Language What it shows Skills Headline result
Live Walkthrough — MSFT English · 繁體中文 The method, step by step: data → evaluate → red team → fact-check → learn 4 NEUTRAL · 5.5 · HOLD
META 10-K Deep Dive English Six skills chained over one annual report 6 NEUTRAL · 5.6 · HOLD
NVDA 10-K Deep Dive English Read the filing, then fact-check the reading 2 BULLISH · 6.8 · HOLD
AMD 10-K 深度解析 繁體中文 The same two steps on AMD, in Traditional Chinese 2 NEUTRAL · 5.5 · HOLD
PLTR Full Demo English Ten skills across four questions: business, price, market, risk 10 NEUTRAL · 4.5 · HOLD (stock-eval)
RKLB 完整示範 繁體中文 The same ten skills on Rocket Lab, in Traditional Chinese 10 NEUTRAL · 4.2 · HOLD (stock-eval)

Which one should I read?

  • “How do I use this?” → the Live Walkthrough.
  • “I have a 10-K and want to understand it.” → META (many skills on one filing) or NVDA / AMD (one deep read, then verify).
  • “I want everything on one stock.” → PLTR or RKLB.

How the demos were made

Every demo follows the same recipe, so you can reproduce any of them.

1 · Get primary-source data (no API key).

export EDGAR_USER_AGENT="Your Name you@example.com"
npm run fetch:edgar -- META --form 10-K     # full filing text → data/filings/META/
npm run fetch:fundamentals -- META          # reported figures → data/fixtures/META.md
# then add the latest close on the pack's `price:` line (the SEC carries no prices)

2 · Run each skill headless, with limited tools.

Demo type Tools allowed Why
10-K deep dives (META, NVDA, AMD) Read files only The filing is the whole story; no browsing means no figures from memory or stale web pages
Full demos (PLTR, RKLB) Read + web search + web fetch Price history, Form 4 trades, short interest and transcripts aren’t in filings; every web figure must be dated and sourced 
claude -p "/us-stock-analysis:financial-report-analyst META — use only data/filings/META/META_2026_10-K.txt and data/fixtures/META.md" --allowedTools Read

3 · Check the result with a second skill. Each demo ends with fact-check or result-validator run over an earlier output.

4 · Screenshot real output. Screenshots are rendered from the saved output text, with omitted lines marked “… (lines omitted for length)”. Nothing inside an excerpt is edited.

What the demos have in common

  1. Every output starts with a Data & Sources header — where each number came from, its date, and how far to trust it.
  2. Skills say “unavailable” instead of guessing. When a figure isn’t in the files (or can’t be found), the output says so.
  3. Skills disagree, and say why. On PLTR, stock-eval says HOLD, stock-valuation says SELL, and technical-analysis says BUY on a pullback: three time horizons, three answers, all explained.
  4. They catch data problems. In the 10-K demos, the analyst corrected the SEC data pack every time: an SG&A line that was really G&A (META), a net-debt line that ignored $52B of securities (NVDA), an FCF figure that included discontinued operations (AMD).
  5. Every call comes with exit conditions. Each output lists what would flip its signal.

Earlier demos

The previous demos (June–July 2026) were long single-document reports produced from data supplied in the prompt. They have been replaced by these re-runs, which use fresh SEC data, current prices and the current skill set, and show every step. The old versions remain in the git history.

Disclaimer: Educational content only. Not financial advice.