| name | tushare-recap-reports |
| description | Generate Tushare-based first-double and tenbagger watch recap reports. |
Tushare Recap Reports
Use this skill for the Tushare stock recap chain:
first-double: build the eligible A-share research universe and separately mark stocks whose recent half-year price move has doubled.tenbagger-watch: score the full eligible A-share universe for potential-stock follow-up.full-chain: run both reports in sequence.
The skill is self-contained under skills/tushare-recap-reports and writes HTML,
CSV, JSON, and (for full-chain) a Feishu interactive card payload. It expects
TUSHARE_TOKEN in the environment or in a repo-root .env file. Set
TUSHARE_URL to use a Tushare-compatible proxy.
The default first-double screen uses Tushare daily prices multiplied by
adj_factor and normalized by the interval-end factor (the qfq price mode), requires at least 80 available trading days,
keeps only SH/SZ/BJ A shares, and excludes ST/delisting-risk names. If an
adjustment factor or cached price series is missing, the report records a data
warning and the watch report tries to backfill the candidate's date range.
python3 skills/tushare-recap-reports/scripts/run.py full-chain
Common options:
python3 skills/tushare-recap-reports/scripts/run.py first-double --end-date 20260529 --min-pct-change 100
python3 skills/tushare-recap-reports/scripts/run.py tenbagger-watch --source-report artifacts/reports/tushare-recap-reports/first_double/latest.json
For an explicit raw-price comparison or a different coverage threshold:
python3 skills/tushare-recap-reports/scripts/run.py first-double --price-mode raw --min-trading-days 100
The potential-stock report is a scored research queue, not a tenbagger
prediction. The current scoring version is v4.0-full-universe. It starts from
the full eligible A-share universe (active A shares, non-ST, sufficient trading
history) and combines trend stage, liquidity, industry breadth, point-in-time
financial quality, valuation safety margin, benchmark market regime, and
multi-timeframe pullback structure. The half-year-double list remains a
separate retrospective validation sample; it is not an entry condition for the
research pool. Each candidate also carries an archetype, why_now, first
rejection point, and thesis-kill condition.
The A tier is gated: a candidate needs both verified quality and a usable setup.
Names outside the fundamental fetch window remain explicitly unverified rather
than being presented as core candidates. The scoring version and any
source/backfill data warnings are included in the JSON and HTML output.
The watch report now separates observed rise drivers from unverified hypotheses.
Observed drivers include industry breadth/median return, recent turnover amount,
recent acceleration, trend drawdown, positive-day ratio, benchmark regime, and
multi-timeframe price structure. The top 80 provisional names attempt to load
Tushare fina_indicator evidence as of the report cutoff date; future-announced
rows are excluded to avoid look-ahead leakage. Missing permissions or missing
filings are shown as warnings instead of being treated as a reason for the rise.
The technical ranking runs across the full universe using the daily-market cache.
To keep the job within the GitHub Actions time budget, point-in-time
fina_indicator evidence is refreshed only for the top 80 pre-financial names;
the rest are explicitly not eligible for A/B priority until that evidence is
available.
When run through the potential GitHub Actions task, the card is sent after the
report chain succeeds. The card explicitly distinguishes full-A-share coverage,
the final research queue, and the half-year-strong replay sample.
FEISHU_POTENTIAL_WEBHOOK_URL is preferred; when it is not configured, the
existing daily webhook is used as the fallback target.
Default output directory: artifacts/reports/tushare-recap-reports.
Default Tushare cache directory: artifacts/cache/tushare.
