メニュー
Use when designing leakage-safe train/test windows, universe controls, and pool-size variants for J-stock experiments that select tickers from historical entry-signal performance before out-of-sample evaluation.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Use when: generating worker-ready JSON manifests for J-stock evaluation, pos-evaluation, walk-forward-evaluate, or replay-evaluation batch execution. Resolve WebUI-equivalent defaults, split jobs by worker during generation, and stop before execution.
Use when: orchestrating an end-to-end J-stock batch evaluation workflow, or when the user explicitly names evaluation-batch-runner. Schedule evaluation-batch-generator, evaluation-batch-executor, and evaluation-result-summary in order, preserving the manifest approval boundary and allowing subagent delegation only during approved execution.
Use when: aggregating completed J-stock evaluation output directories into markdown summary reports under G:/My Drive/AI-Stock-Sync/summary without rerunning backtests, including global MDD-Win re-scoring via tools/score_evaluation_outputs.py when worker-local PRS is not comparable.
Use when generating training-window entry-signal-analysis datasets for a J-stock stock-pool experiment, especially scanning a broad/full universe with a fixed entry strategy before ticker-level pool selection.
Use when orchestrating a J-stock custom stock-pool experiment where a fixed entry condition is applied to a broad/full universe during a training window, good tickers are selected into one or more test universes, and those universes are evaluated out-of-sample against current and full-pool controls.
Use when generating approval-ready J-stock out-of-sample evaluation commands or batch manifests that compare trained selected stock pools against current production and broad/full universe controls.
| name | stock-pool-split-planner |
| description | Use when designing leakage-safe train/test windows, universe controls, and pool-size variants for J-stock experiments that select tickers from historical entry-signal performance before out-of-sample evaluation. |
Create the experiment contract before generating data or running backtests.
Resolve or ask for:
G:/My Drive/AI-Stock-Sync/config.json.140 as the primary size when matching the current pool.Prefer anchored walk-forward style when the user has no stronger preference:
2022-01-01 through 2024-12-31, test 2025-01-01 through 2025-12-31.2022-01-01 through 2025-12-31, test from 2026-01-01 through the latest available date.If data availability differs, adjust with explicit absolute dates and state the reason.
Plan these universe controls for each OOS fold:
current140: production/current 140-stock pool, passed as an explicit universe file.full: broad/full universe, with survivorship-bias caveats if it is based on a current listing file.selected_top140: tickers selected from the training window only.selected_top80, selected_top200.Return a compact plan table with:
Do not run entry-signal-analysis, write universe files, or generate evaluation manifests in this skill.