Menu
Use when the user requests broad investment research, thesis development, bull and bear analysis, evidence comparison, source-grounded company or market research, or synthesis that goes beyond a single data specialist.
Installer avec Codex ou Claude Copiez ce prompt, collez-le dans Codex, Claude ou un autre assistant, puis laissez-le vérifier la page du skill et l'installer pour vous.
Basé sur la classification professionnelle SOC
Central router for OBaI's MCP servers. Trigger this skill whenever the user asks a financial-markets question and the OBaI MCP servers (localhost ports 8001-8010) are available: stock prices, quotes, technicals, fundamentals, valuation, SEC filings, insider trades, earnings, dividends, news catalysts, options chains and Greeks, stock screening, portfolio analysis, ETF holdings, treasury rates, trading-strategy design and backtesting, qualitative company research, Polymarket prediction markets, or Coinbase spot crypto. This skill maps the question to the right OBaI specialist skill and MCP server, then enforces grounding and synthesis rules. Read the matching specialist skill before calling that server's tools.
Real-time and historical stock market data via the OBaI market-data MCP server (http://localhost:8002/mcp). Use for current prices, quotes, OHLCV candles, technical indicators (RSI, SMA, EMA, ADX), market movers, sector snapshots, pre/after-hours quotes, short volume, market hours, and commodity/futures prices (gold, oil, gas). Read before calling any market_data_* tool.
Use the OBaI CLI to answer financial and stock market questions by running `obai query` commands with `--session` for conversation memory. Trigger this skill whenever the user asks about stock prices, earnings, fundamentals, options, market movers, portfolio analysis, portfolio risk, screening, dividends, SEC filings, insider trades, backtesting strategies, commodity or futures prices, company deep dives, competitive positioning, management quality, product sentiment, thematic research, or any financial market question. The agent should autonomously run `obai query "<question>" --json --session <id>`, parse the JSON output, and present the answer — without asking the user to run commands themselves. Always use --session to maintain context across related queries. Also trigger when the user asks to check OBaI server health, run evaluations, or compare stocks.
Autonomous paper trading bot that manages a stock portfolio on Alpaca's paper trading platform. Use this skill when the user asks to: run the daily trading routine, check the portfolio or account status, execute or close trades, evaluate strategy signals, screen stocks, get technical analysis, review trading performance, deploy a new strategy, run portfolio risk analysis, or validate strategy robustness. Trigger on mentions of paper trading, Alpaca, autotrader, buy/sell stocks, portfolio check, market hours, risk limits, strategy signals, or daily routine.
Coinbase spot crypto research via the OBaI crypto MCP server (http://localhost:8010/mcp). Use for Coinbase spot product lookup, crypto OHLCV candles, order books, latest trades and bid/ask, single-product spot strategy backtests, and Coinbase paper-ledger strategy artifacts. Excludes derivatives, DeFi, on-chain analysis, other exchanges, and live order placement. Read before calling any crypto_* tool.
Polymarket prediction-market analysis via the OBaI prediction-markets MCP server (http://localhost:8009/mcp). Use for market discovery, event odds, YES/NO executable pricing and depth, price history, trade flow, holder concentration, trader leaderboards, wallet analysis, trade decision memos, calibration/longshot-bias analysis, rule backtests, edge estimation, and Kelly sizing. Excludes equity strategy backtesting (use obai-strategy). Read before calling any prediction-market tool.
| name | obai-research-routing |
| description | Use when the user requests broad investment research, thesis development, bull and bear analysis, evidence comparison, source-grounded company or market research, or synthesis that goes beyond a single data specialist. |
Use this skill to decide when the Hub should involve research_analysis and how to combine it with other specialists.
Research routing is for broad evidence synthesis, not for replacing specialist data tools.
Use this skill when the user asks for a research-style answer that requires multiple evidence types, source comparison, or thesis construction.
Relevant intent categories include:
Do not use this skill — route to the listed specialist instead — for:
market_data_analysisscreener_lookupevents_news_analysisfundamentals_analysisoptions_analysisportfolio_analysisstrategy_analysisprediction_market_analysisResearch is not a substitute for specialized current data when the user asks for current state.
When a question needs both structured data and qualitative synthesis, call research_analysis alongside the relevant specialist instead of choosing one or the other.
When structured-data specialists cannot answer a qualitative or thematic question, route to research_analysis rather than answering from model memory. Web synthesis beats training knowledge.
research_analysis runs semantic web search, which needs a company name (not a ticker symbol) to retrieve relevant sources.
Before calling research_analysis:
screener_lookup firstDo not pass a bare ticker symbol when the search target is a company name — the semantic search will return weak results.
Use research_analysis when the answer needs broad interpretation, thesis framing, or source comparison.
Use direct specialists when the answer needs concrete data from a specific domain.
The Hub may combine research with:
Do not call every specialist by default.
Before calling research_analysis, identify:
Pass compact context, not raw transcript bloat.
Research output is usually evidence-supplier output, not terminal output.
The Hub may synthesize research output with other specialists using obai-stock-synthesis unless a terminal specialist output contract applies.
When research conflicts with another specialist:
Do not let research synthesis override: