finance markets

Skill

Public-market data and analysis: live and historical stock/ETF/crypto/index prices, market cap, valuation multiples, company financials (income statement, balance sheet, cash flow), earnings transcripts and estimates, analyst targets, dividends, debt, M&A, sector peers, ETF holdings, insider and institutional activity, private/pre-IPO valuations, and macro indicators (GDP, inflation, rates). Also covers filing-grounded facts — a company's officers, board, executive compensation, deal terms. Use for any question whose answer is a current or point-in-time market number, never a number from memory.

Files1
  • @skills/finance-markets/SKILL.md

Finance Markets

Requires a Kortix connector (market-data). Marketplace skill — install it when a market-data provider is configured through Kortix. Every price, fundamental, and KPI in your answer must come from that connector, not from training data.
Answer public-market questions with numbers you actually pulled — quotes, fundamentals, earnings, estimates, ratios, macro series — and show where each one came from.

The one rule that matters

Market data is stale the moment it leaves your weights. Current prices, market caps, margins, and KPIs are not in your training data, and a confidently-stated old number misleads the user worse than "I need to look that up." So:
  • Never answer a market number from memory — not even for the biggest, most familiar names.
  • Never fetch it from the open web — no search, no scraping, no URL fetches for prices or fundamentals. Web snippets are unstructured and often pre-revision. The connector returns structured, point-in-time, revisable data; the open web does not.
  • The only exception: a specific datum the connector genuinely doesn't cover (see Routing below).

Getting the connector

This skill reads from a market-data provider reached through a Kortix connector — for example a market-data API surfaced via Pipedream, or a provider you authorize with an API key. You query the provider through the connector bridge; you never hardcode a vendor's tool names.
  1. Check what's connected — list usable connectors (kortix executor connectors, or the connectors MCP tool). A market-data provider in that list means data is live.
  2. If none is connected, set it up in the same turn:
    • App / OAuth provider — mint a connect link (connect tool, or kortix executor connect <slug>). If the connector isn't on the project yet, add it first (add_connector, or kortix executor add <slug> --provider pipedream --app <app>), then connect.
    • API-key provider — mint a secret link (request_secret, or kortix secrets request <PROVIDER_API_KEY>); your code reads the key from env once it lands.
    • Surface the URL with a one-line "what this is for", end your turn, and verify on return. Don't ask the user to paste a raw key into chat or go hunting in a dashboard.
  3. Query, then cite — pull the data, then state each figure with its provenance (below).
If a user names a preferred provider ("always use for market data"), remember it and prefer it; surface a reconnect link rather than silently falling back if it's down.

Reference-date discipline

Most time-sensitive lookups (financials, earnings, price history, schedules) need an explicit as-of date. "Latest" is ambiguous and non-reproducible — never leave it implicit.
  1. Pick the reference date. Explicit in the question ("as of March 2024") → use it. Implied period ("Q4 results", "last year's revenue") → infer the date that returns it. Otherwise → today, from the session context.
  2. Map to the company's fiscal calendar, not the calendar year. Fiscal year/quarter parameters follow the company's fiscal clock. Do not assume Jan–Dec; many large names don't use it.
  3. Historical prices → pass the reference date as the explicit end date on price-history requests. Use live quotes only when the reference date is today.
Every table and report states the reference date it was built on: "As of 2026-06-29" or "TTM ending FY2025 Q2."

Provenance discipline

Treat provenance as part of the answer, not an afterthought. A response full of market numbers with no traceable source is a failure.
  • Every figure carries its source and as-of date — raw lookups (revenue, EPS, market cap, a balance-sheet line) and derived values (margins, growth, ratios) alike.
  • Derived values show their work — record the formula and the input figures behind each computed metric, so a reader can retrace it.
  • Numbers come from a real pull. If you didn't fetch it this session, you can't state it. No fabricated source labels, no "obvious" shortcuts for follow-ups.
  • Precision, not hedging. Use the exact value the connector returned — "$21.09/share", not "around $21" or "~$21.09". The data is precise; present it that way.
  • Never invent. If a value isn't available, render — or omit the row. Don't backfill with an estimate.
Note: scheduled or notification surfaces may render plainer text — there, just state the plain value with its source label; don't block a real number because a rich provenance link won't render.

Data-source routing

For each public-company question, stop at the first source that answers it:
  1. Financial statements — revenue, net income, margins, EPS, debt, cash, capex, FCF, SBC, dividends, shares, tax rate, and any ratio derivable from them (ROE, D/E, FCF margin, CAGR). For non-US tickers, expect different column naming — read the actual headers, don't assume.
  2. Earnings transcripts — far more than EPS. Forward guidance, non-GAAP metrics, and company-specific KPIs (ARPU, take rate, GMV, same-store sales, subscriber counts, load factor), segment breakdowns, M&A rationale, and management Q&A. Try transcripts before any web search — they answer roughly half of what looks like it needs a filing. For guidance-vs-actuals, pull both the guidance quarter and the results quarter.
  3. Filings — for point-in-time facts that live in SEC/EDGAR documents (officers and directors, board nominees, executive/director compensation, audit fees, deal and securities terms, proxy/8-K/10-K disclosures), do a filing lookup — never state these from memory. For post-filing news (a deal's status after a ruling), use web search instead.
  4. Private / pre-IPO — for a specific private company's profile, funding rounds, or secondary-market marks, use the connector's private-markets data if available. Only for company-specific private lookups — not public tickers, not a user's own portfolio (that's personal-finance), not broad venture commentary.

Analysis patterns

Fetch the component figures, compute locally, cite the result.
  • Profitability — gross/operating/net margin, ROE = NI / equity, ROA = NI / assets, ROIC = NOPAT / invested capital.
  • Liquidity & leverage — current ratio, quick ratio, D/E = total debt / equity, debt/EBITDA, interest coverage = EBIT / interest.
  • Comparables — pick a peer set by sector and size, pull valuation multiples (P/E, EV/EBITDA, P/S, P/B, PEG) and growth, table them against each other and a sector benchmark.
  • DCF — project FCF (operating cash flow − capex) from history, discount at WACC, add a Gordon-growth terminal value FCF_final * (1+g) / (r−g) with g ≈ 2–3%, subtract net debt, divide by shares for an implied price; compare to the live price for upside/downside.
  • Statistical — from price history compute daily returns, annualized volatility std(daily) * sqrt(252), beta vs a benchmark, correlation, Sharpe.

Formatting

  • Currency — \$0.45 · \$150.25 · \$12.5K · \$1.2M · \$150.5B · \$2.1T.
  • Percentages — price changes signed (+2.5%, -1.3%); static ratios/yields unsigned (3.5%); growth as +15% YoY.
  • Tables — right-align numbers, keep precision consistent, and pair every absolute value with a benchmark (vs sector, vs history). Lead with the insight, not a data dump.
Quick stock snapshot: price + change, market cap / P/E / div yield, day range and volume, 52-week range and position in it, a key-metrics table vs sector. Peer comparison: a valuation table and a profitability table, then 2–3 takeaways. DCF: assumptions, an FCF projection table, and a valuation bridge to the implied price.

Gotchas

  • Adjusted vs raw prices — price history is usually split/dividend-adjusted. Comparing it to a news-quoted or target price computed on raw prices will be wrong.
  • Stale financials near earnings — statement data is the last filed period; in the weeks before a release it can be 3+ months old and misleading for fast-moving names. Flag the as-of date.
  • Splits break fixed thresholds — a price alert pinned to a stored dollar level will fire phantom moves after a split. Trigger on return/percentage recomputed from fresh adjusted prices.
  • Scope — this skill is public-market only. A user's own brokerage positions, balances, or transactions belong to personal-finance. Structured multi-name screens, investor-style scoring, portfolio risk, backtests, and formal thesis stress-tests belong to investment-research (load it alongside this one).