gemstar-skill

name	gemstar-skill
description	通过兼容 SKILL.md 协议的第三方智能体查询 GemStar 量化交易提醒、最新信号、目标持仓和运行状态；只允许查看或显式触发一次本地研究流程，禁止实盘下单。

GemStar Integration Skill

Use this skill when the user asks from a compatible skill host about GemStar, current holdings, target holdings, trading alerts, live signals, paper ledger, leaderboard, bundled strategies, or whether GemStar has buy/sell guidance.

Repository

GemStar lives at:

/Users/ken/workspace/GemStar

Run all GemStar commands from that directory.

Primary Commands

Current holdings, target holdings, and rebalance actions:

cd /Users/ken/workspace/GemStar
test -f artifacts/current/trade_status.md && cat artifacts/current/trade_status.md

Use the bundled script when the current directory is unknown:

bash /Users/ken/workspace/GemStar/integrations/gemstar-skill/scripts/gemstar-status.sh

Machine-readable current trade status:

cd /Users/ken/workspace/GemStar
test -f artifacts/current/trade_status.json && cat artifacts/current/trade_status.json

Latest chat-friendly alerts:

cd /Users/ken/workspace/GemStar
uv run python -m src.cli.app alerts latest --limit 5

Use the bundled script when the current directory is unknown:

bash /Users/ken/workspace/GemStar/integrations/gemstar-skill/scripts/gemstar-alerts.sh 5

Machine-readable latest alerts:

cd /Users/ken/workspace/GemStar
uv run python -m src.cli.app --output json alerts latest --limit 5

Latest leaderboard:

cd /Users/ken/workspace/GemStar
uv run python -m src.cli.app leaderboard

Bundled leaderboard strategy:

cd /Users/ken/workspace/GemStar
test -f strategies/leaderboard_quality_lowvol_v1/config.yaml && cat strategies/leaderboard_quality_lowvol_v1/config.yaml

Pipeline status:

cd /Users/ken/workspace/GemStar
uv run python -m src.cli.app status

Optional Manual Run

Only run this when the user explicitly asks to refresh or run GemStar now:

cd /Users/ken/workspace/GemStar
uv run python -m src.cli.app trade --once

This may take several minutes because it can run the deterministic production pipeline, target generation, and one live cycle. It does not invoke LLM research unless the user explicitly runs gemstar research.

Response Style

• Return concise Chinese summaries suitable for chat.
• For "当前持仓", "目标仓位", "今天建议", and "为什么买/卖", read artifacts/current/trade_status.json first; fall back to trade_status.md, then alerts.
• For "最新信号", "最新建议", or "今天信号", always report the status ref_date and file updated_at first. If ref_date is older than today's date, explicitly say the current completed GemStar status is stale and do not present it as today's latest signal.
• For "最新 leaderboard", run uv run python -m src.cli.app leaderboard and summarize rank, strategy, status, Sharpe, CAGR, max drawdown, and alpha.
• For "内置榜单策略", read strategies/leaderboard_quality_lowvol_v1/config.yaml and summarize universe, timer, factor weights, top_n, rebalance frequency, and backtest window.
• Preserve stock code and Chinese name, e.g. 300750.SZ 宁德时代.
• If no alerts exist, say there are no GemStar alerts yet and include the checked path.
• If no trade status exists, say GemStar has not generated artifacts/current/trade_status.md yet and suggest running gemstar trade --once.
• If a command fails, report the command and the short error. Do not invent trading signals.

Safety Rules

• Never place real trades.
• Never call QMT, ptrade, broker APIs, or order entry tools.
• Never modify strategy YAML, factor pool, credentials, or configuration from WeChat or any chat host.
• Do not treat GemStar status messages as investment advice; present them as system-generated alerts.
• If the user asks to buy/sell through a chat host, refuse execution and suggest checking the GemStar alert details manually.