yixin-stock-workflow

name

yixin-stock-workflow

description

Run a Yixin OpenAPI based A-share stock research workflow that scans fresh market/AI hot topics, queries financial and price data, applies completeness gates, builds a technical-structure TOP 5, and exports Markdown, HTML, and PNG reports. Use when the user asks to run, package, improve, or generate an HTML/image stock-picking report with Yixin search and fin_db, especially for 智能选股, 热点追踪, A股选股, 技术结构, TOP 5, 自选股候选, or report generation workflows.

Yixin Stock Workflow

Use this skill to generate an A-share research-assistant report from Yixin OpenAPI. The workflow calls:

• search: fresh market and AI/technology news
• fin_db: candidate stocks, price/trend data, growth data, and optional valuation data

The bundled script exports:

• raw JSON under data/
• Markdown, HTML, and optional PNG under reports/

Security

Never include real API keys in this skill, reports, source code, commits, screenshots, or examples.

Users must configure their own Yixin API keys before running:

{
  "search": "<search-api-key>",
  "fin_db": "<fin-db-api-key>"
}

Default location:

~/.config/yixin-api/api-keys.json

Alternative environment variables:

export YIXIN_SEARCH_API_KEY="..."
export YIXIN_FIN_DB_API_KEY="..."

Quick Start

From any project directory:

python3 /path/to/yixin-stock-workflow/scripts/run_yixin_stock_workflow.py --output-dir .

If Chrome is unavailable or image rendering is not needed:

python3 /path/to/yixin-stock-workflow/scripts/run_yixin_stock_workflow.py --output-dir . --skip-image

When installed in Codex:

python3 ~/.codex/skills/yixin-stock-workflow/scripts/run_yixin_stock_workflow.py --output-dir .

Workflow

1. Query recent market/AI-related news with search.
2. Apply local freshness filters: recent dates only, useful snippets only, low-quality sources removed, duplicate titles removed.
3. Query candidate stocks and trend data with fin_db.
4. Check trend-field completeness. If real trend fields are incomplete, automatically run narrower supplemental fin_db queries for:
   • 5/20/60-day price change
   • close price with 20/60-day moving averages
   • close price, 52-week high, and distance from the 52-week high
5. If the current date is a weekend, ask trend queries as of the latest A-share trading day rather than the calendar date.
6. Merge the base trend response and supplemental trend responses before scoring.
7. If moving-average or 52-week fields are still missing after the first supplemental pass, retry once with calculation-oriented query wording. For moving averages, split candidates into chunks before requesting daily close history so API row limits do not truncate the 60-day sample.
8. Clean candidate tables: remove duplicate stocks, sparse columns, and incomplete valuation columns.
9. Apply completeness gates:
   • PE/PB is displayed and scored only when complete enough.
   • If valuation is incomplete, hide it and use growth metrics instead.
10. Generate a technical-structure TOP 5 using available trend, moving-average, 52-week-position, volume, and growth data.
11. Export Markdown, HTML, and PNG reports.

If trend fields are incomplete after supplemental queries, the workflow must not force a technical TOP 5. It should report data coverage and downgrade the output to a candidate observation pool until real trend fields can be fetched.

A formal technical TOP 5 requires enough coverage for real price change, close price, 20-day moving average, and 60-day moving average. If only one of these dimensions is available, keep the partial data visible but do not call it a formal TOP 5.

Technical Structure Model

The TOP 5 is a research ranking, not investment advice. It combines:

• trend and momentum
• 20/60-day moving-average structure
• price position versus 52-week high
• volume/change context when available
• simplified candlestick/K-line interpretation
• simplified Chan theory structure wording
• growth metrics when valuation data is incomplete

Only real trend fields may drive technical momentum scoring:

• allowed: pct_change, avg_pct_change, explicit 涨跌幅 columns, 20/60-day moving averages, close price, and 52-week high fields
• blocked: turnover/amount, market cap (total_mv), PE/PB/PS, and other valuation or size fields

When a fin_db response mixes market price, valuation, and market-cap rows in one table, parse rows by item_key/item_name instead of treating the whole response as one metric type.

If the first broad trend query returns only close price, turnover, valuation, or market-cap fields, split the trend query into narrow requests. The supplemental requests should explicitly say not to return turnover, market cap, PE, PB, PS, or other non-trend fields when asking for涨跌幅.

If fin_db returns daily pct_change rows for a 5/20/60-day period, aggregate those daily percentage changes into a cumulative period return before scoring. Do not use the last daily row as the period return.

If fin_db returns daily close rows instead of moving-average columns, calculate 20-day and 60-day moving averages locally from the latest close-history sample. Keep the sample count visible in the displayed trend section.

Use cautious labels such as 趋势延续, 回调观察, 接近前高, 短期涨幅偏快, and 20日线可作为短线结构观察位.

Avoid language such as 买入, 卖出, 稳赚, 必涨, or target prices.

Outputs

The script prints generated paths:

report=/.../reports/<run-id>-workflow-report.md
html=/.../reports/<run-id>-workflow-report.html
image=/.../reports/<run-id>-workflow-report.png

Use the HTML report for browser review and the PNG report for sharing.

Maintenance Notes

• Keep API key handling in load_keys() only.
• Do not commit generated data/ or reports/ unless the user explicitly wants sample outputs.
• If a returned table has incomplete fields, prefer hiding the dimension over showing partial data.
• If Yixin returns stale search results, rely on local date filtering rather than trusting time_range.
• Never substitute market cap, transaction amount, or valuation fields for trend momentum. If real涨跌幅/均线/52周 fields are missing, show a data completeness note and do not publish a formal technical TOP 5.
• Report conclusions must match the actual scoring result. If no formal TOP 5 is generated, call the output a 候选观察池; do not present it as a formal ranking or confirmed list.