intraday-watch

name	intraday-watch
description	Intraday snapshot of a watchlist while the session is open — live quotes, move-vs-benchmark, and which names are moving beyond their own normal range. Triggers on "盘中", "现在怎么样", "实时", "盘中异动", "现在涨跌", "intraday", "live quotes", "盘中监控".

Intraday Watch

The session is still open, so nothing here is a conclusion. This skill reports what is moving right now, against the tape and against each name's own normal dispersion, and hands the reader a shortlist to look at — it does not explain intraday moves and does not tell anyone to trade.

close-recap is the same list after the close, with attribution against disclosed events. Attribution belongs there, not here: intraday, the announcement record has usually not caught up with the price, so a cause offered now is a guess wearing a citation. If the user wants "why", say the disclosed record is checked after the close and offer close-recap.

The two tools here take structured parameters, not a natural-language query

Every other 同花顺 tool in this plugin takes {"query": "..."}. These two do not:

hexin-stock.stock_highfreq_quotes
hexin-index.index_highfreq_quotes
  symbols     comma-separated, 简称 and/or code mixed: "<简称>,<代码>"
  indicators  comma-separated 中文 names: "最新价,涨跌幅,成交量,换手率"
  data_mode   "real_time" for the latest snapshot
  interval    1

The response is shaped differently too — no indicators_params. It carries:

• tables — a header row followed by one row per symbol
• sympolMap (the vendor's spelling) — how each symbol resolved: <简称> → <证券代码>
• indicatorMap — the 中文 indicator name mapped to its internal field

Read sympolMap back. It is the only confirmation that the code you passed resolved to the name you meant and not something else.

涨跌幅 is already a percentage. Verified across two tools on the same day: stock_highfreq_quotes returned values like 0.0512 and 0.018, and get_stock_performance — which declares 涨跌幅（单位：%） — returned the identical numbers at identical prices for those same names. So a 0.0512 is +0.05%, not +5.12%. Do not multiply by 100.

Workflow

Step 1: Establish whether you actually have a live quote — from the response, not a clock

data_mode: "real_time" does not fail outside trading hours. It returns the last available snapshot, stamped with when that snapshot was taken. Verified 2026-08-07: a call placed at 19:31 Beijing (well after the 15:00 close) returned rows stamped 2026-08-07 16:01:07 — a successful response carrying end-of-day data.

So, before reporting anything:

1. Read the time column off the returned rows. It is Beijing time.

2. Resolve the current date and time from the session context or the user — there is no clock tool, and the environment clock may not be Beijing (a UTC container reads 8 hours behind, which would make a stale 16:01 stamp look like the future).

3. Compare. If the stamp is not within the current session, say so in the first line and label every figure as of that stamp:

   盘中快照不可用 — 最新快照为 2026-08-07 16:01（北京时间），当前非交易时段。以下为该 时点数据，非实时报价。

Never present a stale snapshot as a live quote, and never infer "the market is closed" from the environment clock alone. A row whose stamp is minutes old during 09:30–11:30 or 13:00–15:00 Beijing is live; anything else is a snapshot with a timestamp.

Also settle the universe: a stored watchlist under watchlists/, or the names the user gave. If several lists exist and none was named, ask once.

Step 2: Pull the snapshot

One call for the whole list — symbols takes the full comma-separated set, so do not fan out per name here (unlike the daily tools).

• Indicators worth asking for: 最新价,涨跌幅,成交量,换手率. Ask for what the question needs; this tool silently drops indicators it cannot serve, so reconcile what you asked for against the header row of tables and say which ones did not come back.
• A name missing from tables is missing, not flat — report it as 源不可用 for this snapshot and name it. Suspended names behave the same way.
• There is no history in this tool (vendor: 仅支持交易日日内数据查询，不支持历史数据 查询). So an intraday high/low, an opening gap, or a comparison to yesterday's close must come from get_stock_performance (daily) — do not claim an intraday range this tool did not return.

Step 3: The tape, at the same moment

hexin-index.index_highfreq_quotes for the benchmarks the list keys off (沪深300, 上证指数, 恒生指数 as applicable), same data_mode and the same call.

• Take both timestamps and compare them. A quote stamped 10:42 read against an index stamped 16:01 is not a relative move. Where the stamps differ materially, say so and do not compute the spread.
• Move-vs-benchmark is [测算]: 名称涨跌幅 − 基准涨跌幅, with both stamps stated.

Step 4: Which moves are actually unusual

A raw ±2% list is noise on a volatile book and misses everything on a quiet one. Normalise before flagging, exactly as close-recap does:

• hexin-stock.get_risk_indicators for 波动率(年化) per name, and read its window back from indicators_params (defaults to a trailing 1 年).
• Flag a name when the move is large relative to its own normal dispersion, not against a flat threshold. State the rule you used and the vol you used.
• A move inside the name's ordinary range is reported as such and given no narrative. Intraday, that restraint matters more than at the close: the record has not caught up, so there is nothing to attribute to.

Step 5: Output

Short Markdown, in-session. This is a monitoring read, not a document — do not offer PDF, and do not build a workbook unless the user asks for one. When the user does ask to export it ("导出成 Excel", "给我个表"), build it through xlsx-author rather than by hand: a snapshot is a table of retrieved quotes, so every hardcoded cell carries its Source: <system>, <数据时点>, <indicator> comment and the workbook closes with the 来源 worksheet described there. A snapshot exported without its 数据时点 and source per cell is indistinguishable from a stale one two hours later.

**盘中快照 — [清单名]**
数据时点: [Beijing time from the `time` column] · 基准: [指数] [涨跌幅]% (同一时点)
[若非交易时段: 一行说明这是收盘后快照,非实时报价]

**超出自身常态波动的名称**
| 名称 | 最新价 | 涨跌幅% | 相对基准% | 年化波动率%(窗口) | 数据时点 |
|---|---|---|---|---|---|
|  |  |  | [测算] | [披露] |  |

**其余名称**（在常态波动范围内,不作解读）
| 名称 | 最新价 | 涨跌幅% | 数据时点 |
|---|---|---|---|

## 覆盖范围与局限
数据时点: [stamp] (北京时间) · 交易时段: [是/否]

| 检查项 | 结论 | 源 | 数据时点 |
|---|---|---|---|
| 个股实时快照 | 有记录 [n]/[N] 只 / 源不可用 [k] 只(列名) | 同花顺 hexin-stock.stock_highfreq_quotes | [stamp] |
| 基准指数同时点 | 有记录 / 时点不一致(未计算相对涨跌) | 同花顺 hexin-index.index_highfreq_quotes | [stamp] |
| 波动率(用于常态判定) | 有记录 / 源不可用 | 同花顺 hexin-stock.get_risk_indicators | [date] |

未取到的指标: [向工具请求但未返回的指标名]
本快照不含盘中最高/最低价与开盘跳空(该工具不返回),亦不含历史序列。
盘中价格随时变动;本节结论仅对上述数据时点成立。

## 来源
[n] 〔一手〕同花顺 iFinD · [工具名] · 数据时点 [stamp](北京时间);检索于 [timestamp]

Guardrails

• This skill issues no trading view. No 买入/卖出/加仓/减仓/止损, no target price, no "值得关注" that functions as a recommendation. It reports what moved and how unusual the move is against the name's own history; the portfolio manager decides.
• No intraday attribution. Do not offer a cause for an intraday move — not from news, not from sector beta, not from a pattern. Point at close-recap for attribution once the disclosed record has caught up.
• Every number carries its data timestamp, and the timestamp is the vendor's, in Beijing time. 检索于 and 数据时点 are different fields and are never substituted.
• A stale snapshot presented as live is a fabrication even though the number is real. Label it.
• Absence is absence: a name the snapshot did not return is 源不可用 with the name stated, never omitted and never carried at its last known price.