stock-analysis

Stock Analysis

Stock and company analysis plus strategy backtesting using direct recent-price feeds and the Rebyte financial data lake.

Requires Rebyte API auth — $AUTH_TOKEN and $API_URL are set up per the agent's system prompt; use them as Bearer token and base URL.

Skill layout — load the pillar you need

Pillar	When
this file	Analysis playbooks: price checks, company overviews, comparisons, fundamentals, technicals
data/SKILL.md	Data routing and mechanics: direct two-date prices, full 19-table lake catalog (US + CN), SQL patterns, news + research search, error rules. Read before fetching data.
backtesting/SKILL.md	Strategy simulation: 5-phase NautilusTrader workflow ending in a backtest result bundle
financial-templates/SKILL.md	Analysis structures (DCF, comps, memo formats) with no data calls
report-style/README.md	Kami design system for every HTML report this skill delivers
references/sec-edgar.md	SEC filings, full 10-K/10-Q text, insider (Form 4) trades via edgartools

Price/K-line charts: use the financial-charts skill (TradingView-style Lightweight Charts).

Data sources

Source	What it provides	Access
Direct recent-price APIs	Price-only OHLCV bars for the current and previous exchange-local calendar dates. US minute bars use stocks/bars with interval: "1min"; China minute bars use cn-stocks/bars_1min.	POST $API_URL/api/data/stocks/bars, POST $API_URL/api/data/cn-stocks/bars, or POST $API_URL/api/data/cn-stocks/bars_1min — see data/SKILL.md
Rebyte financial data lake	US: daily + 1-minute bars, news, SEC-filing fundamentals, splits, dividends, short data, ticker universe, IPOs. CN A-shares: daily + 1-minute bars, valuation snapshots, financial statements, money flow, unusual-move disclosures.	Read-only SQL via POST $API_URL/api/data/financial/sql — see data/SKILL.md
News archive	US equity news coverage back to 2016, searchable by meaning	POST $API_URL/api/data/research/news — see data/SKILL.md
Research library	~4,300 long-form articles from 13 investment research publications (SemiAnalysis, SemiVision, MacroCharts, Capital Wars, Citrini, Doomberg, Michael J Burry and others), 2020 to today. Primary analysis by named practitioners — use it for theses, debates, and mechanisms.	POST $API_URL/api/data/research/search, then /context or /article — see data/SKILL.md
SEC EDGAR	Full filing text (10-K, 10-Q, 8-K), filing sections, insider (Form 4) trades	edgartools Python library — see references/sec-edgar.md

Route by freshness. Use the direct APIs for "current", "today", "latest price", and recent intraday questions. Their range is fixed server-side to the current and previous calendar dates in America/New_York (US) or Asia/Shanghai (CN); callers cannot widen it. Use the lake for every older or non-price fact. The US response exposes Polygon's upstreamStatus (for example DELAYED), so never imply tick-level realtime. State the source, returned date range, latest bar timestamp, and feed status when present. During market hours, use minute bars in both markets; treat the current date's daily bar as final only after that market closes.

---

Analysis Workflows

Use direct prices for the two-date edge and lake SQL for historical or non-price steps. Exact routing and request shapes are in data/SKILL.md.

1. Quick Stock Check

User: "What's AAPL doing?" / "AAPL price"
→ Direct recent bars — quote the latest close, timestamp, date range, feed status
→ Lake daily bars (last 1 month) + ticker details for context
→ Present: latest direct price + recent trend + basic company info

2. Company Overview

User: "Tell me about NVDA" / "What does Tesla do?"
→ Ticker details + latest fundamentals period (revenue, net income)
→ Daily bars (last 3 months)
→ Direct recent bars when presenting a latest/current price
→ Recent news (5 headlines) — semantic search for themes if needed
→ Present: business summary, market position, recent performance, news themes

3. Technical Analysis

User: "Is TSLA a good buy?" / "AAPL technical analysis"
→ Daily bars (last 6 months) — trend, support/resistance
→ Direct recent intraday bars — current short-term momentum
→ Lake 1-minute bars aggregated to hourly when more than two dates are needed
→ Recent news (10 articles) — read and judge the tone yourself
→ Compute: moving averages, price range, volume trends
→ Present: trend direction, key levels, volume analysis, sentiment, outlook

4. Multi-Stock Comparison

User: "Compare AAPL vs MSFT vs GOOGL"
→ One SQL per dataset covering all tickers (WHERE ticker IN (...))
→ Direct recent bars for each ticker when current prices are part of the comparison
→ Compare: price performance, fundamentals, news flow
→ Present: side-by-side table, relative performance

5. Fundamental Deep Dive

User: "AAPL financials" / "NVDA revenue trend"
→ Fundamentals: annual (5 periods) + quarterly (4 periods)
→ Dividends (last 12) — history and implied yield vs latest direct close
→ Weekly-aggregated bars (last 2 years) — long-term price context
→ Compute: revenue growth, margin trends, EPS trend, payout ratio
→ Optional: SEC EDGAR for full 10-K text

6. Insider Activity

User: "Insider trading for TSLA" / "Are executives buying NVDA?"
→ SEC EDGAR: Form 4 filings (the lake does not carry insider trades)
→ Daily bars (last 3 months) — price context around the trades
→ Present: recent transactions, insider sentiment, correlation with price

7. Due Diligence Package

User: "Full analysis of MSFT" / "Due diligence on AMD"
→ Ticker details, 1y daily bars, annual + quarterly fundamentals,
  dividends, splits, 20 recent news items
→ Direct recent bars for the current price snapshot
→ SEC EDGAR: latest 10-K, recent 8-Ks, Form 4 insider trades
→ Present: comprehensive report (Kami-styled HTML per report-style/)

8. Sector Research

User: "Compare cloud stocks" / "Best semiconductor stocks"
→ Identify tickers (us.tickers can filter by name/type/exchange)
→ Batch daily bars + fundamentals across the set
→ Semantic news search on the sector theme
→ Present: sector overview, leaders, relative performance

9. Strategy Backtest

User: "Backtest an SMA crossover on AAPL" / "验证我的策略"
→ Switch to backtesting/SKILL.md and run its 5 phases end-to-end
→ Deliverable is the backtest bundle at /code/backtests/<slug>/

---

Trigger Patterns

ALWAYS fetch data when the user mentions any of these. Do NOT answer from memory — route to the direct price API and/or lake as specified.

User intent	Required actions
Stock symbol mentioned (AAPL, $TSLA, 000001.SZ)	Direct recent bars + lake ticker details
"current", "today", "latest price", "how is X doing"	Direct recent bars first; include marketDateRange, latest timestamp, and feed status
Historical "price" or "chart"	Lake bars for the requested range; add direct bars only when the latest edge matters
"news", "what's happening with"	News query (10+ items) or semantic search
"analyze", "research", "tell me about"	Details + bars + news
"compare", "vs", "versus"	All datasets for each stock, side-by-side
"buy", "sell", "good investment"	Bars + news + fundamentals (annual + quarterly)
"financials", "revenue", "earnings"	Fundamentals (annual + quarterly)
"dividend", "yield", "payout"	us.dividends
"split", "stock split"	us.splits
"short interest", "shorts"	us.short_interest / us.short_volume
"insider", "who's buying/selling"	SEC EDGAR Form 4 filings
"10-K", "10-Q", "SEC filing"	SEC EDGAR filings
"backtest", "回测", "strategy performance"	backtesting/SKILL.md

---

Lake SQL recipes

Auth + request format, DataFusion-style SQL patterns, the on-error rule, and the full table catalog are in data/SKILL.md. The recipes below map the common analysis needs; tickers are UPPERCASE for US, NNNNNN.SZ/NNNNNN.SH for CN.

-- Price bars, daily (raw, unadjusted; check us.splits before spanning a split)
SELECT t, o, h, l, c, v FROM us.eod
WHERE ticker = 'AAPL' AND t >= to_timestamp('2026-01-01')
ORDER BY t

-- Intraday / custom intervals: aggregate 1-minute bars
SELECT date_bin(INTERVAL '1 hour', t, TIMESTAMP '1970-01-01') AS bucket,
       min(t) AS t_open, max(h) AS h, min(l) AS l, sum(v) AS v
FROM us.bars_1m
WHERE ticker = 'AAPL' AND t >= now() - INTERVAL '5 days'
GROUP BY bucket ORDER BY bucket

-- News for a ticker (tickers is an array column)
SELECT published_utc, title, tickers FROM us.news
WHERE array_has(tickers, 'AAPL')
ORDER BY published_utc DESC LIMIT 10
-- Thematic retrieval ("news about AI chip demand"): use research/news instead
-- of ILIKE; for theses and mechanisms use research/search — see data/SKILL.md.

-- Fundamentals (SEC-filing derived; is_* income, bs_* balance, cf_* cashflow)
SELECT fiscal_year, fiscal_period, is_revenues, is_gross_profit,
       is_operating_income_loss, is_net_income_loss,
       is_diluted_earnings_per_share, bs_assets, bs_liabilities, bs_equity,
       cf_net_cash_flow_from_operating_activities
FROM us.fundamentals
WHERE array_has(tickers, 'AAPL') AND timeframe = 'annual'
ORDER BY fiscal_year DESC LIMIT 5

-- Dividends / splits
SELECT ex_dividend_date, cash_amount, frequency FROM us.dividends
WHERE ticker = 'AAPL' ORDER BY ex_dividend_date DESC LIMIT 12;
SELECT execution_date, split_from, split_to FROM us.splits
WHERE ticker = 'AAPL' ORDER BY execution_date DESC

-- Ticker details / universe screen
SELECT ticker, name, primary_exchange, type, active, cik FROM us.tickers
WHERE ticker = 'AAPL'

-- CN A-share daily bars (adj_factor included for adjusted series)
SELECT t, o, h, l, c, v, pct_chg, adj_factor FROM cn.bars_day
WHERE ts_code = '000001.SZ' AND t >= to_timestamp('2026-01-01')
ORDER BY t

-- CN valuation snapshot / financial indicators
SELECT trade_date, pe_ttm, pb, turnover_rate, total_mv FROM cn.daily_basic
WHERE ts_code = '000001.SZ' ORDER BY trade_date DESC LIMIT 20

The data/scripts/financial_cli.py helper wraps auth + the SQL endpoint:

python3 data/scripts/financial_cli.py catalog
python3 data/scripts/financial_cli.py schema us.eod
python3 data/scripts/financial_cli.py query "SELECT ... LIMIT 10"
python3 data/scripts/financial_cli.py news "Fed rate cut expectations" --ticker NVDA
python3 data/scripts/financial_cli.py research "global liquidity and central bank balance sheets"

---

SEC EDGAR Reference

For full filing text, filing sections, and insider trading data, see references/sec-edgar.md.

Quick start:

pip install edgartools

from edgar import Company, set_identity
set_identity("Rebyte Agent [email protected]")

company = Company("AAPL")
filings = company.get_filings(form="10-K")      # Annual reports
insider = company.get_filings(form="4")         # Insider trades

---

Analysis Guidelines

Computing Technical Indicators from Price Bars

The lake returns raw OHLCV. Compute indicators yourself (or in SQL):

• Simple Moving Average (SMA): average of last N closes (20-day and 50-day)
• Price trend: current close vs 20-day and 50-day SMA
• Support/Resistance: recent lows/highs from daily bars
• Volume trend: recent volume vs 20-day average volume
• 52-week range: min low / max high over 1 year of daily bars
• Corporate actions: us.eod is unadjusted — when the window spans a split, adjust with us.splits; CN daily bars carry adj_factor directly

Reading News

Lake news carries no precomputed sentiment — read the headlines/content and judge the tone yourself, noting publisher weight and recency. For thematic questions, prefer the semantic search endpoint over keyword ILIKE.

Financial Statement Analysis

• Revenue growth: YoY change across periods
• Margin trends: gross/operating/net margin over time
• Cash position: cash & equivalents vs total debt
• Earnings quality: operating cash flow vs net income (should be close)

Presenting Results

• Lead with the answer (bullish/bearish/neutral, latest sourced close + timestamp, key metric)
• Use tables for multi-stock comparisons
• Include specific numbers with dates — never vague statements
• Distinguish facts (from data) from analysis (your interpretation)
• State each source and date range explicitly; report direct-feed status and label lake data T+1
• Long-form deliverables: Kami-styled HTML per report-style/

---

Important Notes

• US tickers UPPERCASE (AAPL); CN codes suffixed (000001.SZ, 600519.SH)
• All timestamps UTC
• Direct prices cover two exchange-local calendar dates only — weekends and holidays can yield empty dates; do not widen the direct request
• Lake data is T+1 — daily tables land the prior trading day
• Direct means recent, not guaranteed tick realtime — during market hours, use US stocks/bars with interval: "1min" and CN cn-stocks/bars_1min; report Polygon's upstreamStatus, and use completed daily bars after close
• Coverage: US from 2021-06 (bars) with reference data much deeper (splits 1978→, dividends 2000→); CN daily from 1990-12 — full catalog in data/SKILL.md
• SEC EDGAR is free — no API key, but requires an identity string

---

When NOT to Use This Skill

• Simple web lookup ("What's Apple's website?") → use web search
• Live trading, tick execution, bid/ask, or streaming quotes → not supported; direct prices are OHLCV bars and may be delayed
• Personal financial advice → not qualified