dflow-kalshi-market-data（dflow 卡尔希市场数据）

DFlow Kalshi Market Data

Pull data about a known Kalshi market (or set of markets) — orderbook, trades, prices, candles, forecasts, in-game live data — as a snapshot, a historical range, or a live stream.

Prerequisites

• DFlow docs MCP (https://pond.dflow.net/mcp) — install per the repo README. This skill is the recipe; the MCP is the reference. Query params, pagination, exact payload schemas, WS snapshot-vs-diff semantics, and the category-specific live_data.details shapes all live there — don't guess.

Surface

All data endpoints in this skill run against the Metadata API (https://pond.dflow.net/build/metadata-api) — REST for snapshots and history, WebSockets for live streams. Call it from anywhere: a curl from the command line, a Node/Python script, a cron job, a backend, or a Next.js proxy fronting a browser UI.

If the user says "run this from my terminal", don't reach for the dflow CLI — it has no market-data subcommands. Write a short HTTP/WS script against the Metadata API instead.

Pick the shape first

Three intents, three shapes. Match the user's phrasing, then pick the endpoint:

• Snapshot ("right now", "current") → REST, one call.
• History ("last hour", "between T1 and T2", "last N trades") → REST with time / limit params.
• Stream ("live", "as it happens", "alert me when") → WebSocket subscription.

Data → endpoint map

For each dataset below, the one-liner covers all three shapes. Field-level details (exact params, pagination tokens, payload schemas) → docs MCP.

Orderbook

• Snapshot: GET /api/v1/orderbook/{ticker} or /api/v1/orderbook/by-mint/{mint} (includes sequence).
• Stream: orderbook channel (yes_bids + no_bids maps per update; no sequence on the stream payload).

Trades — two endpoints, overlapping but different scopes

• GET /api/v1/trades (and /trades/by-mint/{mint}) — the complete market print tape. All trades that hit Kalshi's orderbook, which includes DFlow onchain fills (those hit Kalshi's book too; see the "Do onchain trades show up on Kalshi's trade websocket?" FAQ). This is the default for "show me trades on this market." Stream equivalent: trades channel.
• GET /api/v1/onchain-trades (and /onchain-trades/by-market/{ticker}, /onchain-trades/by-event/{eventTicker}) — DFlow onchain fills only, with onchain-specific fields that /trades doesn't carry: wallet, transactionSignature, id, inputAmount, outputAmount, createdAt. Subset of what's on /trades, but richer per-row. No WS stream.
• Decision: *complete tape* → /trades. *Wallet-scoped activity feed, DFlow-execution analytics, tx-signature lookups* → /onchain-trades. Real-time fill detection for a specific user order → parse program events directly (see /build/prediction-markets/onchain-trade-parsing).

Top-of-book prices

• Snapshot: read yesBid / yesAsk / noBid / noAsk directly from the market object (GET /api/v1/market/{ticker} — singular) — no separate endpoint.
• Stream: prices channel.

Candlesticks (OHLCV)

• Market-level: GET /api/v1/market/{ticker}/candlesticks or /api/v1/market/by-mint/{mint}/candlesticks.
• Event-level: GET /api/v1/event/{ticker}/candlesticks.
• 5,000-candle cap per request (see Gotchas).

Forecast percentile history

• Event-level: GET /api/v1/event/{seriesTicker}/{eventId}/forecast_percentile_history (plus /api/v1/event/by-mint/{mint}/forecast_percentile_history). Kalshi's historical forecast distribution for an event.

Live data (Kalshi passthrough)

• GET /api/v1/live_data, /live_data/by-event/{ticker}, /live_data/by-mint/{mint}.
• Response includes a details object whose fields depend on the milestone type — football, soccer, tennis, golf, MMA, baseball, cricket, racing each have their own known-field sets. See live-data-details in the docs MCP before touching details.

Streaming lifecycle

Connect → subscribe → handle → reconnect. In a sentence each:

• Connect: dev is wss://dev-prediction-markets-api.dflow.net/api/v1/ws (no auth). Prod is wss://prediction-markets-api.dflow.net/api/v1/ws with x-api-key on the WS upgrade headers. REST equivalents: https://dev-prediction-markets-api.dflow.net and https://prediction-markets-api.dflow.net.
• Subscribe: send {type: "subscribe", channel: "prices" | "trades" | "orderbook", all: true | tickers: [...]}. Each channel holds its own subscription state.
• Handle: parse each message by channel, process asynchronously — don't block the read loop.
• Reconnect: exponential backoff on disconnect, and re-send every subscription after reconnect. The server doesn't remember you.

Exact message schemas (prices, trades, orderbook), heartbeat/ping behavior, and incremental-vs-snapshot semantics on the orderbook channel → docs MCP.

What to ASK the user (and what NOT to ask)

Query shape — infer if unambiguous, confirm if not:

1. Which market — ticker or outcome mint.
2. Which dataset — orderbook, trades (Kalshi vs onchain), prices, candles, forecasts, or live data.
3. Snapshot / history / stream — infer from phrasing, confirm if ambiguous.
4. History bounds / interval — time range (startTs, endTs) and periodInterval for candles; limit for trades.

Infra — always ask, never infer:

1. DFlow API key. **Ask with a clean, neutral question: *"Do you have a DFlow API key?"*** Don't presuppose where the key lives — phrasings like *"do you have it in env?"* or *"is DFLOW_API_KEY set?"* nudge the user toward env-var defaults they didn't ask for. Don't assume the user has one just because they mention the dflow CLI is configured. Surface the choice; don't silently fall back to env or to dev. It's one key for everything DFlow — same x-api-key unlocks the Trade API *and* the Metadata API, REST *and* WebSocket. If yes → prod host (https://prediction-markets-api.dflow.net REST, wss://prediction-markets-api.dflow.net/api/v1/ws WS) with x-api-key on every request (REST and the WS upgrade). If no → dev host (https://dev-prediction-markets-api.dflow.net, wss://dev-prediction-markets-api.dflow.net/api/v1/ws), rate-limited; point them at https://pond.dflow.net/build/api-key for a prod key. When you generate a script, log the resolved host + key-presence at startup so the user can see which rails they're on.

Do NOT ask about:

• RPC, wallet, signing — this skill is read-only public data.
• Settlement mint / slippage / fees — trade-side concerns; if the user pivots to placing an order off something they see here, hand off to dflow-kalshi-trading.

Gotchas (the docs MCP won't volunteer these)

• Two trade endpoints, overlapping scopes. /api/v1/trades is the complete market tape (Kalshi-offchain order flow plus DFlow onchain fills — DFlow fills hit Kalshi's book). /api/v1/onchain-trades is the DFlow-onchain subset, enriched with wallet / transactionSignature / input-output amounts. When a user says "show trades on this market" they want /trades; when they say "show this wallet's DFlow activity" they want /onchain-trades?wallet=....
• Orderbook returns only bid ladders (yes_bids, no_bids). Best YES *ask* is derived: 1 - max(no_bids keys) (a NO bid at p is a YES offer at 1-p). Same on REST and the WS channel.
• Two price scales. Probability strings ("0.4200") on orderbook + prices channels. Integer 0–10000 on /trades + trades channel, with yes_price_dollars / no_price_dollars string companions. Normalize before you compute.
• 5,000-candle cap per request, hard 400. If the range × interval would produce more than 5,000 candles, the endpoint returns a 400 with no partial result — it's Kalshi's upstream cap forwarded through DFlow. Narrow the range, widen the interval, or page yourself.
• periodInterval is in minutes, not seconds. Kalshi convention: 1 = 1-minute candles, 60 = hourly, 1440 = daily. Easy to blow past the 5,000-candle cap by assuming seconds.
• live_data.details is categorical, not generic. Fields differ per milestone type. Don't hardcode cross-category field access; branch on type and pull the known fields for that category from the MCP's live-data-details reference.
• WebSocket all: true is a firehose. Especially on prices and orderbook. Use a ticker list unless the monitor truly needs universe-wide coverage.
• WS subscriptions don't survive reconnects. After every reconnect, resend every subscribe message you had before the drop.
• Streams can go quiet in the maintenance window — Thursdays 3:00–5:00 AM ET, Kalshi is offline; expect sparse or missing WS traffic and stale REST fields.
• The CLI's stored key doesn't flow into your script's HTTP client. dflow setup stores the key for the dflow binary's own use. The Metadata API calls your script makes directly are separate — they need the key plumbed in (env, .env, flag). It's one DFlow key, but two plumbing sites any time you mix CLI invocations with direct HTTP/WS calls in the same codebase.

When something doesn't fit

For anything not covered above — full parameter lists, pagination tokens, exact WS message shapes (snapshot-vs-diff on orderbook, heartbeat cadence), candlestick interval enums, category-specific live_data.details fields, forecast-percentile response shape — query the docs MCP (search_d_flow, query_docs_filesystem_d_flow). Don't guess.

Sibling skills

• dflow-kalshi-market-scanner — find markets matching a criterion across the universe (uses these primitives, shapes them into named scans).
• dflow-kalshi-trading — place buy / sell / redeem orders on a market you're watching here.
• dflow-kalshi-portfolio — view the user's own positions and P&L.