stock-kline-analysis（股票 K 线分析）

Stock K-Line Analysis Skill

Use this skill when the user gives a stock name or code and wants K-line output and/or analysis, e.g.:

• "Analyze 600519"
• "K-line and trend for 贵州茅台"
• "How is AAPL doing?"
• "Compare 000001 and 600036 by relative strength"

Defaults (baked-in)

User overrides these defaults at any time.

Market Auto-Detection Rules

If detection confidence is low, list top-2 candidates and ask once.

AkShare API Reference

Scripts

All implementation code lives in scripts/ next to this file. You can run a full analysis end-to-end with:

python .github/skills/stock-kline-analysis/scripts/run_analysis.py 000063
python .github/skills/stock-kline-analysis/scripts/run_analysis.py 贵州茅台
python .github/skills/stock-kline-analysis/scripts/run_analysis.py AAPL --out-dir /tmp/reports

Each script has a __main__ smoke-test (e.g. python fetch_kline.py 000063).

Workflow

Step 1 — Resolve Identifier

1. Apply market auto-detection rules to the raw input.
2. If code is numeric and 6-digit, run ak.stock_info_a_code_name() to confirm name.
3. If name is given, filter symbol list for closest match.
4. If ambiguous (>1 high-confidence match), show max 3 options and ask.
5. If unresolvable, report clearly and stop.

Completion check: one confirmed {code, name, market} tuple before fetching any data.

Step 2 — Fetch K-Line Data

See scripts/fetch_kline.py — fetch_all_timeframes(code, adjust="qfq") returns (df_daily, df_weekly, df_monthly), all normalized.

Key implementation notes (do NOT get wrong):

• stock_zh_a_hist returns 12 Chinese-named columns — always use explicit rename(col_map), never positional assignment.
• Windows: daily = last 182 days, weekly = last 365 days, monthly = last 3 years.

from scripts.fetch_kline import fetch_all_timeframes
df_daily, df_weekly, df_monthly = fetch_all_timeframes(code)

Fallback logic:

• On network error, retry once.
• If daily fetch is empty: report (suspended / delisted / wrong symbol / holiday) and stop.
• Weekly/monthly failures: skip that timeframe and note it in output.

Completion check: len(df_daily) > 20 and all OHLCV columns present and non-null.

Step 3 — Compute Indicators

See scripts/indicators.py — add_indicators(df) and add_tf_indicators(df_weekly, df_monthly).

Critical notes:

• bb_width = (upper − lower) / mid * 100 — result is a percentage (e.g. 12.7, not 0.127).
• RSI must use a standalone helper; do not chain .diff() twice on the same series.
• Support/resistance stored in df.attrs["support"] / df.attrs["resistance"].

from scripts.indicators import add_indicators, add_tf_indicators
df_daily = add_indicators(df_daily)
df_weekly, df_monthly = add_tf_indicators(df_weekly, df_monthly)

If fewer than 60 bars exist on daily, use all available and note the limitation. Bollinger Bands require minimum 20 bars.

Step 4 — Build K-Line Chart

Primary path — matplotlib (4-panel: K线+布林带 | 成交量 | MACD | RSI-14):

See scripts/chart.py — plot_kline(df, code, name, out_path, market_label, dpi) returns the saved path.

mplfinance is NOT installed in the base environment. chart.py calls matplotlib.use("Agg") at module level — always import before pyplot.

from scripts.chart import plot_kline
chart_path = plot_kline(df_daily, code=code, name=name, market_label="A股",
                        out_path=f"{code}_kline.png")

Fallback (text table — only if matplotlib is also unavailable), latest 20 bars:

date        close   MA20    BB_up   BB_low  RSI14   ATR%
2026-02-10  15.38   14.90   16.20   13.60   58.3    1.4%
...

Step 5 — Valuation Context

See scripts/valuation.py — fetch_valuation(code) and compute_pe_pb(result, last_close).

ak.stock_a_lg_indicator and ak.stock_a_indicator_lg do not exist — use stock_yjbb_em + stock_financial_analysis_indicator instead (both implemented in valuation.py).

from scripts.valuation import fetch_valuation, compute_pe_pb
val = fetch_valuation(code)
val = compute_pe_pb(val, last_close=float(df_daily["close"].iloc[-1]))
# val keys: eps, revenue, revenue_yoy, net_profit, net_profit_yoy,
#           book_value_per_share, roe, gross_margin, industry,
#           report_date, fin_df, pe_ttm, pb

For HK/US: skip valuation section or note it as unavailable.

Report:

• Current PE (TTM, computed from EPS), PB.
• ROE and net profit margin.
• Revenue/profit YoY growth from latest report.
• Industry classification.
• Note: historical PE percentile not available without stock_a_lg_indicator; skip that sub-bullet and state the reason.

Step 6 — Portfolio / Relative Strength Mode

Activated when user provides multiple symbols (e.g. "compare 600519 and 000858").

1. Fetch 6-month daily data for all symbols (same window as default).
2. Compute normalized 6-month return (base=100 on start date).
3. Compute 20-day rolling volatility and ATR% for each symbol.
4. Rank by: return, Sharpe-proxy (return/vol), RSI, and ATR% (lower = more stable).
5. Produce a comparison table and identify the relative leader.

Single-symbol mode: compare to its own industry index if identifiable.

Step 7 — Event-Aware Risk Overlay

See scripts/events.py — fetch_events(lookback_days, lookahead_days, min_importance) returns a list of formatted strings.

news_economic_baidu() date column is datetime.date objects — compare natively. Data lags 4–8 weeks; events.py always appends a hardcoded China macro calendar (PMI, Two Sessions, earnings windows) regardless of API success.

from scripts.events import fetch_events
event_lines = fetch_events()  # returns list[str] ready to print

Overlay on analysis:

• Note any major macro event dates near current price levels.
• Flag the applicable earnings season window relative to today.
• Highlight price behavior around large news days visible in the K-line.

If event API is unavailable, note it and manually annotate the known calendar dates above.

Step 8 — Deliver Structured Output

Return in this exact order:

[Symbol Summary]
名称/代码:           e.g. 贵州茅台 (600519) · A-Share Shanghai
分析区间:            2025-09-12 → 2026-03-12 (daily 6M, qfq-adjusted)
多周期确认:          Weekly trend: Uptrend | Monthly trend: Consolidation

[K-Line Snapshot]
最新收盘:           ¥1,580.00
1日涨跌:           +1.2% (+18.80)
MA5 / MA20 / MA60: ¥1,572 / ¥1,540 / ¥1,490   (排列多头 Bullish stack)
布林带 Bollinger:  Upper ¥1,640 | Mid ¥1,540 | Lower ¥1,440  (Width: 12.7%)
ATR-14 (波动幅):   ¥22.4 / day  (1.4% of price — moderate volatility)
20日区间:           ¥1,420 – ¥1,610
成交量 vs 20日均:   +35%  (放量)

[Technical View — 技术面]
趋势 Trend:         Daily Uptrend — MA5 > MA20 > MA60, price above all MAs
                    Weekly confirm: above weekly MA20 ✓
                    Monthly confirm: testing monthly MA20 resistance ⚠
动量 Momentum:     5D: +3.1% | 10D: +5.8% | 20D: +8.2% | Ann.Vol: 18%
MACD:              MACD line above signal, histogram expanding → bullish momentum
RSI-14:            68 — approaching overbought; momentum still intact
布林挤压 BB Squeeze: Width 12.7% — expanding (breakout in progress, not overextended)
支撑 Support:      ¥1,490 (MA60 + BB lower + prior swing low)
阻力 Resistance:   ¥1,640 (BB upper) / ¥1,650 (6M high zone)
ATR止损参考:       Trailing stop = last close − 1.5×ATR = ¥1,580 − ¥33.6 ≈ ¥1,546

[Valuation — 估值]
PE (TTM):          28x — 3-year 40th percentile (moderate)
PB:                8.2x
行业 PE 中位:      25x (white spirits industry) → slight premium to peers

[Relative Strength]  (if multi-symbol mode)
Symbol   6M Return  Vol    Sharpe  RSI   ATR%   Rank
600519   +22%       18%    1.22    68    1.4%   1st ← Leader
000858   +14%       21%    0.67    55    1.7%   2nd

[Event Overlay — 事件]
- 2026-03-15: NPC economic policy announcement (macro risk)
- 2026-04-30: Q1 earnings release window (re-rating trigger)
- No major gap days observed in 6M K-line window.

[Risk & Watchpoints — 风险]
- 多单失效: If price closes below MA20 (¥1,540) on volume → trend weakening
- 布林下轨破位: Price below BB lower (¥1,440) = volatility expansion to downside
- 超买风险: RSI near 70; daily overbought but weekly RSI 58 = room still exists
- ATR止损: Position sizing reference — 1 ATR = ¥22.4; adjust size accordingly
- 突破条件: Break above BB upper (¥1,640) + volume >+50% avg → momentum continuation

Quality Criteria

• Market mapping is stated and auditable.
• All indicator values are computed from actual fetched data, not estimated.
• Valuation section states data source and percentile basis.
• Event overlay explicitly covers ±30 days around analysis date.
• No language implying guaranteed price direction or investment advice.
• If any section fails (e.g. valuation API times out), skip it with an explicit note.

Example Prompts

• "Use stock-kline-analysis to analyze 600519."
• "Use stock-kline-analysis for 贵州茅台 — show K-line with Bollinger Bands, MACD, RSI, and ATR stop-loss."
• "Use stock-kline-analysis on AAPL — multi-timeframe trend: are daily/weekly/monthly aligned?"
• "Use stock-kline-analysis to compare 600036 and 601318 by relative strength and ATR-based risk."
• "Use stock-kline-analysis for 000858 — show Bollinger squeeze and flag any upcoming earnings event."
• "Use stock-kline-analysis for TSLA — is price near Bollinger upper band? What does ATR say about position sizing?"