shield_lock
金大哥
← 返回 Skill 列表
extension
分类: 数据与分析无需 API Key

昨日外盘收盘速览

Collects and summarizes yesterday's external (non-A-share) financial market performance - US, Europe, and Asia-Pacific equities, commodities, FX, bonds, crypto, and the VIX - for the most recent completed overseas trading session, then produces a structured recap with a read-through for the next A-share session. Use when the user asks for 外盘收盘 / 昨夜外盘 / 昨天美股 / 海外市场复盘 / 全球市场表现 / 海外收盘速览 / 隔夜外盘 / 外盘涨跌 / overseas market close / yesterday's global market recap, or wants a pre-open morning briefing on how external markets moved (excluding A-shares).

person作者: user_43f7505bhubcommunity
download下载资源包

外盘收盘速览（External Market Recap）

Overview

Produce a one-glance recap of how non-A-share (external) markets closed in their most recent completed trading session — typically the US session that finished overnight Beijing time, plus the Europe and Asia-Pacific (ex-A-share) sessions that ran the same calendar day. The deliverable is a structured table plus a short read-through for the upcoming A-share session.

Scope explicitly excludes A-shares, A-share ETFs, and onshore RMB cash markets. Offshore Chinese assets that trade externally (Hong Kong Hang Seng, US-listed Chinese ADRs, USD/CNH) are in scope, because they are the most direct overnight signal for A-shares.

When to Trigger

Trigger this skill on requests such as:

  • "昨天外盘怎么样" / "昨夜美股收盘" / "外盘收盘速览"
  • "海外市场复盘" / "全球市场表现" / "隔夜外盘涨跌"
  • "帮我整理一下昨天美股、欧股、大宗商品、汇率、美债、加密货币的收盘"
  • "明天A股开盘前，外盘给了什么信号"
  • "Yesterday's global market recap" / "Overseas market close summary"

If the user asks specifically about A-shares only, do not use this skill — redirect to the A-share data tools instead. If the user wants both A-share and external, run this skill for the external portion and pull A-share data separately.

Execution Workflow

Step 1 — Determine the target trading day

Run the bundled helper to resolve "yesterday" into the most recent completed external trading session. It steps back one day from today (today's overseas session is usually still open), then skips weekends and US equity holidays:

<managed-python> scripts/trading_day.py

(Run the skill's bundled scripts/trading_day.py with the managed Python runtime. An explicit date may be passed, e.g. scripts/trading_day.py 2026-08-03, to recap a past session.)

The script prints the target date (e.g. TARGET_DATE=2026-08-05), its weekday, and DAYS_BACK. Use this date as the label for every data point in the report. If the script is unavailable, fall back to: take the prior calendar day; if it lands on Sat/Sun, step back to Friday.

Note: External markets have different local holidays. The helper handles US equity holidays; for Europe/Asia sessions, if a search result shows a market was closed, record "休市" rather than fabricating a move.

Step 2 — Collect data by category via WebSearch

Gather closing levels and percentage moves. Search in focused batches (one WebSearch call per category below) so results are complete and the most recent session is captured. Use the query templates in references/coverage-list.md. Pass the resolved TARGET_DATE and phrase queries like "YYYY-MM-DD close" / "昨夜收盘" to bias toward the right session.

Cover these categories:

  1. US equities — Dow Jones, S&P 500, Nasdaq Composite; plus VIX.
  2. Europe equities — STOXX 600, DAX, FTSE 100, CAC 40, Euro Stoxx 50.
  3. Asia-Pacific ex-A-share — Nikkei 225, Hang Seng, Kospi, Taiwan Taiex, ASX 200.
  4. Commodities — Gold (COMEX), WTI & Brent crude, Copper, Silver.
  5. FX — DXY, EUR/USD, USD/JPY, USD/CNH (offshore RMB).
  6. Rates & crypto — US 10Y & 2Y Treasury yields; Bitcoin, Ethereum.
  7. Chinese ADRs (optional but recommended) — Nasdaq Golden Dragon China Index or a representative basket (BABA, PDD, JD, NIO) as the cleanest overnight read-through for A-shares.

Use the query_keyword_groups parameter of WebSearch (not multiple separate calls) when covering several instruments in one category to capture multiple angles efficiently.

Session-dating pitfall (critical): Roundups published on TARGET_DATE (Beijing morning) usually describe the prior US session — e.g. a US Tuesday close is reported Wednesday Beijing morning, so a 2026-08-05 news search returns Tuesday's tape, not the Wednesday session the skill targets. The correct session's own roundup is published the next calendar day. If a TARGET_DATE search returns a tape that contradicts the structured quote feed (see below), re-search using the next calendar day's roundup labels (e.g. 2026年8月6日 财经早餐 / 美股收盘 / 欧股收盘) to capture the true session. Cross-check the structured quote feed (basic_info / 美股行情指标 blocks) filtered to TARGET_DATE — it reflects the correct session directly and is the most reliable anchor when news labels are ambiguous.

Step 3 — Compile the recap

Build a Markdown report using the template in references/coverage-list.md. For each instrument record: name, closing level, daily change (%, and points where natural), and a one-line note (driver / context). Group by the seven categories above. Always include a header line stating the target session date and the data source basis (WebSearch, real-time at collection time).

Step 4 — Add the read-through

Append a short (3–6 bullet) "对A股开盘的指引" section: what the external tape implies for risk appetite, which A-share sectors are most exposed (e.g. tech ↔ Nasdaq/SOX, energy ↔ oil, precious metals ↔ gold, exports ↔ DXY/USD-CNH), and any notable divergence. Keep it directional and evidence-based — tie each point to a number from the table. Do not give a buy/sell call on individual A-shares; this is a market-context brief.

Step 5 — Output

  • Default: a clean Markdown report in the chat response.
  • If the user asks for a file or a dashboard-style view, offer to also render an HTML version (single root <div>, inline styles only, no external CSS/JS) using the same table data.

Quality Rules

  • No fabrication. Every number must come from a WebSearch result captured this run. If a figure is missing or contradictory across sources, mark it [MISSING] or note the discrepancy — never invent a close or a percentage.
  • Stale awareness. If the only data found is from a session older than the target date (e.g. holiday gap), state the actual date of the data and label it [STALE].
  • Source attribution. Cite the source outlet / data provider per row where possible (e.g. "Reuters", "CNBC", "MarketWatch", "Investing.com").
  • Units & signs. Always show the sign on percentage moves (+/−). State the currency for FX pairs and the unit for commodities (USD/oz, USD/bbl).
  • Exclude A-shares by design. Do not pollute the recap with Shanghai/Shenzhen/ChiNext indices; those belong to a separate A-share brief.

Resources

  • scripts/trading_day.py — resolves the most recent completed external trading session date.
  • references/coverage-list.md — full instrument checklist, WebSearch query templates per category, and the Markdown output template.