risk-agent
๐ฏSkillfrom psh355q-ui/szdi57465yt
Analyzes and categorizes financial news risks across legal, regulatory, operational, financial, and strategic domains to generate trading signals.
Installation
npx skills add https://github.com/psh355q-ui/szdi57465yt --skill risk-agentSkill Details
Overview
# ๐ค๐น AI Trading System
Multi-AI ๊ธฐ๋ฐ ์๋ ์ฃผ์ ํธ๋ ์ด๋ฉ ์์คํ
> Claude, ChatGPT, Gemini๋ฅผ ํ์ฉํ ์์๋ธ AI ํฌ์ ํ๋ซํผ
[](https://www.python.org/)
[](https://fastapi.tiangolo.com/)
[](https://reactjs.org/)
[](https://www.postgresql.org/)
[](https://www.timescale.com/)
[](https://redis.io/)
---
๐ ํ๋ก์ ํธ ๊ฐ์
AI Trading System์ ๋ค์ค AI ๋ชจ๋ธ์ ํ์ฉํ ์ํฐํ๋ผ์ด์ฆ๊ธ ์๋ ํธ๋ ์ด๋ฉ ํ๋ซํผ์ ๋๋ค. ๋ด์ค ๋ถ์, ์๊ทธ๋ ์์ฑ, ๋ฐฑํ ์คํ , ๋ฆฌ์คํฌ ๊ด๋ฆฌ, ์ค์๊ฐ ๋ชจ๋ํฐ๋ง์ ํตํฉํ์ฌ ๋ฐ์ดํฐ ๊ธฐ๋ฐ ํฌ์ ๊ฒฐ์ ์ ์ง์ํฉ๋๋ค.
๐ฏ ํต์ฌ ๊ฐ์น
- Multi-AI ์์๋ธ: Claude Sonnet 4.5, ChatGPT-4, Gemini Pro๋ฅผ ์กฐํฉํ์ฌ ๋ ์ ํํ ์์ฅ ๋ถ์
- ์ค์๊ฐ ๋ด์ค ๋ถ์: RSS ํฌ๋กค๋ง โ AI ๋ถ์ โ ํธ๋ ์ด๋ฉ ์๊ทธ๋ ์๋ ์์ฑ
- Point-in-Time ๋ฐฑํ ์คํ : Look-ahead bias ์๋ ์ ํํ ์ ๋ต ๊ฒ์ฆ
- 2-Layer ์บ์ฑ: Redis (L1) + TimescaleDB (L2)๋ก ๋ฐ๋ฆฌ์ด๊ธ ์๋ต ์๋
- ํ๋ก๋์ ๋ ๋: ๋ชจ๋ํฐ๋ง, ์๋ฆผ, Circuit Breaker, ๋น์ฉ ์ถ์ ์๋น
---
โจ ์ฃผ์ ๊ธฐ๋ฅ
๐ฐ ๋ด์ค ๊ธฐ๋ฐ ํธ๋ ์ด๋ฉ (Phase 1-10)
- RSS ๋ด์ค ํฌ๋กค๋ง: 50+ ๊ธ์ต ๋ด์ค ์์ค ์ค์๊ฐ ์์ง
- Historical Data Seeding:
- 5๊ฐ ์์ค(NewsAPI, Google, Reuters, Yahoo, Bloomberg) ๊ณผ๊ฑฐ ๋ฐ์ดํฐ ์์ง
- yfinance ๊ธฐ๋ฐ ๊ณผ๊ฑฐ ์ฃผ๊ฐ ๋ฐ์ดํฐ(OHLCV) ๋๋ ์์ง
- Backfill API: ๋ ์ง/ํฐ์ปค ์ง์ ์๋ ์์ง ๋ฐ ์งํ๋ฅ ์ถ์
- AI ๋ด์ค ๋ถ์:
- ๊ฐ์ฑ ๋ถ์ (๊ธ์ /๋ถ์ /์ค๋ฆฝ)
- Data Explorer: ์์ง๋ ๋ด์ค/์ฃผ๊ฐ ๋ฐ์ดํฐ ์ค์๊ฐ ๊ฒ์ ๋ฐ ํํฐ๋ง UI
- ์์ฅ ์ํฅ๋ ํ๊ฐ (๋จ๊ธฐ/์ฅ๊ธฐ)
- ํฐ์ปค ๊ด๋ จ์ฑ ์ค์ฝ์ด๋ง
- ๋ฆฌ์คํฌ ์นดํ ๊ณ ๋ฆฌ ๋ถ๋ฅ (๋ฒ์ /๊ท์ /์ด์/์ฌ๋ฌด/์ ๋ต)
- ์๋ ์๊ทธ๋ ์์ฑ: ๋ด์ค โ ํธ๋ ์ด๋ฉ ์๊ทธ๋ ๋ณํ
- ์๊ทธ๋ ๊ฒ์ฆ: ๋ฐฑํ ์คํ ๊ธฐ๋ฐ ์๊ทธ๋ ํ์ง ๊ฒ์ฆ
- ์นํฐ ์ค๋กํ๋ง: ์นํฐ๋ณ ํฌ์ง์ ์ ํ ๊ด๋ฆฌ
๐ง Deep Reasoning (Phase 14)
- 3-Step Chain-of-Thought ์ถ๋ก :
- Step 1: ์ง์ ์ํฅ (Direct Impact) - 1์ฐจ ์ํ์ฃผ ํ์
- Step 2: ๊ฐ์ ์ํฅ (Secondary Impact) - ๊ผฌ๋ฆฌ์ ๊ผฌ๋ฆฌ๋ฅผ ๋ฌด๋ ์ฐ์ ๋ถ์
- Step 3: ์ ๋ต์ ๊ฒฐ๋ก (Strategic Conclusion) - ์จ์ ์ํ์ ๋ฐ๊ตด
- Hidden Beneficiary ํ์:
- ์์: "Google TPU v6" โ ๋ช ๋ฐฑํ ์ํ์ฃผ(GOOGL) + ์จ์ ์ํ์(Broadcom - TPU ์นฉ ์ค๊ณ์ฌ)
- Knowledge Graph ๊ธฐ๋ฐ ๊ธฐ์ ๊ด๊ณ ์ถ์
- Knowledge Graph:
- ํํธ๋์ญ, ๊ฒฝ์, ๊ณต๊ธ๋ง, ํฌ์ ๊ด๊ณ ๊ทธ๋ํ ์ ์ฅ
- ๊ฒฝ๋ก ํ์์ผ๋ก N-hop ๊ด๊ณ ๋ฐ๊ฒฌ
- ์ค์๊ฐ ์น ๊ฒ์์ผ๋ก ๊ด๊ณ ๊ฒ์ฆ
- Model-Agnostic AI Client:
- Gemini, Claude, GPT-4 ํต์ผ ์ธํฐํ์ด์ค
- ์ญํ ๋ณ AI ๋ชจ๋ธ ๋ฐฐ์ (Screener, Reasoning, Decision)
- ๋น์ฉ ์ต์ ํ: Gemini Flash(์คํฌ๋ฆฌ๋) + Gemini Pro(์ถ๋ก )
- A/B Backtest: ํค์๋ ๊ธฐ๋ฐ vs CoT+RAG ์ฑ๊ณผ ๋น๊ต
๐ Feature Store (Phase 2-4)
- 2-Layer ์บ์ฑ:
- L1: Redis (In-Memory, 15๋ถ TTL)
- L2: TimescaleDB (์๊ณ์ด DB, ์๊ตฌ ์ ์ฅ)
- ์ค์๊ฐ Feature ๊ณ์ฐ:
- ๊ธฐ์ ์ ์งํ: ret_5d, ret_20d, vol_20d, mom_20d
- ํ๋๋ฉํธ: pe_ratio, market_cap, dividend_yield
- AI ํฉํฐ: non_standard_risk, management_credibility
- Cache Warmer: ์ฃผ์ ์ข ๋ชฉ ์ฌ์ ์บ์ฑ์ผ๋ก ๋ ์ดํด์ ์ต์ํ
- Vector Store: SEC ๋ฌธ์ ์๋ฒ ๋ฉ & ์๋งจํฑ ๊ฒ์
๐งช ๋ฐฑํ ์คํ ์์ง (Phase 10)
- Signal Backtest Engine: ๋ด์ค ์๊ทธ๋ ๋ฐฑํ ์คํ
- Point-in-Time ๋ถ์: Look-ahead bias ์ ๊ฑฐ
- ์ฑ๊ณผ ์งํ:
- Sharpe Ratio, Sortino Ratio
- Win Rate, Profit Factor
- Maximum Drawdown
- Cumulative Returns
- ์ต์ ํ: ๊ทธ๋ฆฌ๋ ์์น๋ก ํ๋ผ๋ฏธํฐ ํ๋
- ๋น๊ต ๋ถ์: ์ฌ๋ฌ ์ ๋ต ์ฑ๊ณผ ๋น๊ต
๐ Advanced Analytics (Phase 15.5)
- Performance Attribution:
- ์ ๋ต๋ณ, ์นํฐ๋ณ, AI ์์ค๋ณ, ํฌ์ง์ ๋ณ, ์๊ฐ๋ณ ์ฑ๊ณผ ๋ถํด
- PnL ๊ธฐ์ฌ๋ ๋ถ์
- Risk Analytics:
- Value at Risk (VaR 95%, 99%)
- Conditional VaR (CVaR/Expected Shortfall)
- Maximum Drawdown & Recovery Period
- Concentration Risk (HHI Index)
- Correlation Matrix
- Stress Testing
- Trade Analytics:
- Win/Loss ํจํด ๋ถ์
- ์คํ ํ์ง (์ฌ๋ฆฌํผ์ง, ์ฒด๊ฒฐ ์๋)
- ๋ณด์ ๊ธฐ๊ฐ ์ต์ ํ
- AI ์ ๋ขฐ๋ vs PnL ์๊ด๊ด๊ณ
๐ ์๋ฆผ & ๋ชจ๋ํฐ๋ง (Phase 7-8)
- Notification System:
- Telegram, Slack, Email ์ง์
- ๋ค์ค ์ฑ๋ ๋ธ๋ก๋์บ์คํธ
- ์ฐ์ ์์ ๊ธฐ๋ฐ ๋ผ์ฐํ
- Health Monitoring:
- ์์คํ health check
- Component-level ์ํ ์ถ์
- ์๋ ๋ณต๊ตฌ ์๋
- Smart Alerts:
- ๋น์ ์ ํจํด ํ์ง
- Circuit Breaker๋ก ๊ณผ๋ถํ ๋ฐฉ์ง
- ์๋ฆผ ์ค๋ณต ์ ๊ฑฐ & ๊ทธ๋ฃนํ
- ๋น์ฉ ์ถ์ :
- AI API ํธ์ถ ๋น์ฉ ์ค์๊ฐ ์ถ์
- ์ผ๋ณ/์๋ณ ์ฌ์ฉ๋ ๋ฆฌํฌํธ
- ์์ฐ ์๋ฆผ
๐ ๋ฆฌํฌํ (Phase 15)
- Daily/Weekly/Monthly Reports:
- ํธ๋ ์ด๋ฉ ์ฑ๊ณผ ์์ฝ
- AI ์ฌ์ฉ ๋น์ฉ
- ๋ฆฌ์คํฌ ๋ฉํธ๋ฆญ์ค
- PDF Export: ์ ๋ฌธ์ ์ธ ๋ฆฌํฌํธ ์์ฑ
- CSV Export: ๋ฐ์ดํฐ ๋ถ์์ฉ ๋ด๋ณด๋ด๊ธฐ
๐ ์ธ์ฆ & ๋ก๊น (Phase 7)
- API Key ๊ด๋ฆฌ: ๊ณ์ธต์ ๊ถํ (Read/Write/Execute)
More from this repository10
Generates and validates trading signals by analyzing news sentiment, performing multi-step AI reasoning, and identifying direct and indirect market impacts across different stocks.
Assists in automated error detection, root cause analysis, and intelligent debugging for the AI trading system's complex multi-model architecture.
Automates complex institutional research workflows by generating structured reports, analyzing organizational data, and mapping stakeholder relationships.
report-writer-agent skill from psh355q-ui/szdi57465yt
pm-agent skill from psh355q-ui/szdi57465yt
Manages and optimizes investment portfolios by dynamically analyzing news, generating trading signals, and recommending strategic stock allocations using multi-AI ensemble reasoning.
Automates and manages cross-platform notifications, routing alerts through multiple channels with intelligent delivery and priority handling
signal-generator-agent skill from psh355q-ui/szdi57465yt
Evaluates trading strategy performance by analyzing historical market data, computing key metrics, and generating comprehensive backtesting reports
Extracts and analyzes sentiment, market signals, and strategic insights from CEO speeches to generate automated trading recommendations.