A-Share Short-Term Decision Skill#

Implement in sequence:

1. Run short_term_signal_engine(analysis_date) for target date.
2. If needed, persist prediction with run_prediction_for_date(analysis_date).
3. Compare prediction vs actual market with compare_prediction_with_market(prediction_date, actual_date).
4. Output report with generate_daily_report(analysis_date).

Tool Contracts#

short_term_signal_engine(analysis_date=None)#

• analysis_date: YYYY-MM-DD or YYYYMMDD
• Returns weighted short-term score and recommendation status.
• Always returns friendly no_recommendation_message when no tradable candidate exists.

run_prediction_for_date(analysis_date)#

• Runs signal engine for the specified date.
• Appends decision snapshot into data/decision_log.jsonl.

compare_prediction_with_market(prediction_date, actual_date=None)#

• Loads prediction from log (or auto-generates if missing).
• Compares predicted candidates against real market closes on actual_date.
• Returns per-stock return and summary statistics.

No-Recommendation Behavior#

Required behavior:

• Never return empty output.
• If candidates is empty or signal is NO_TRADE, explicitly say: 当前暂无可执行短线买入标的.
• Include reason and next action.

Runtime#

bash

python3 main.py short_term_signal_engine --date 2026-02-12
python3 main.py run_prediction_for_date --date 2026-02-12
python3 main.py compare_prediction_with_market --prediction-date 2026-02-12 --actual-date 2026-02-13
python3 main.py generate_daily_report --date 2026-02-12

Subskills Workflow#

For recurring optimize-then-recommend flow, run:

bash

python3 subskills/config-optimization/optimize_from_aggressive.py --analysis-period "2026-02-01 to 2026-02-12"
python3 subskills/daily-recommendation/generate_daily_recommendation.py --date 2026-02-14

All generated artifacts are stored under data/.