> ## Documentation Index > Fetch the complete documentation index at: https://docs.doomberg.me/llms.txt > Use this file to discover all available pages before exploring further. # Starter Prompt > The starter_research MCP prompt — its parameters, the full template text it returns, and ready-to-paste example prompts for your agent. The starter prompt is a ready-to-use research directive you paste into your agent to kick off a full Ithaca research session. It tells the agent to follow the canonical workflow — open a session, research the market, propose a strategy, backtest it, and report — for a specific symbol and question. ## The starter\_research prompt The starter prompt is parameterized by two inputs: | Parameter | Type | Required | Description | | ---------- | ------ | -------- | -------------------------------------------------------------------------------------------------------- | | `symbol` | string | yes | The ticker to research (e.g. `NVDA`, `AAPL`, `TSLA`) | | `question` | string | yes | The research question or thesis to investigate (e.g. "Is now a good time to enter a momentum position?") | ## Full template text When invoked, the starter prompt returns this template text for the agent to follow: ```text theme={null} Research {symbol} using Ithaca and answer: {question} Follow the canonical Ithaca workflow: 1. Call session_context with this prompt to open a traced research session. 2. Research {symbol} using data tools. At minimum: - market_get_ohlcv: pull 1 year of daily price history - market_screener: screen peers in the same sector - fundamentals_get: pull fundamentals metric cards - insider_activity: check SEC Form 4/5 insider transactions - congress_trades: check STOCK Act congressional trading disclosures - analyst_ratings: pull Wall Street consensus and price targets - technical_indicators: compute RSI, MACD, Bollinger Bands, ATR - vol_realized: compute annualized realized volatility Use search_skills to discover additional tools if you need them (e.g. options, sentiment, macro, short interest). 3. Synthesize your findings into a thesis. Note: - Price action and momentum vs. sector - Fundamental health (growth, margins, valuation) - Insider sentiment (net buy/sell) - Congressional trading signals - Analyst consensus and target upside - Technical levels (overbought/oversold, trend) - Volatility regime 4. Call propose_strategy with a declarative StrategySpec that captures your thesis. Use a closed-schema spec — no free-form fields, no code. Choose a strategy id (e.g. rank_returns, mean_reversion) and params that fit the regime you identified. 5. Call backtest_run with the validated spec and range="1y". If using hosted MCP, call run_skill with skill_id="backtest.run" instead, then subscribe_run to follow progress. 6. Report your findings to the human. Include: - Thesis summary - Key evidence (with provenance — note if any data is stale or partial) - Strategy spec (name, universe, signal, construction, risk limits) - Backtest results (CAGR, Sharpe, max drawdown, win rate, trade count) - Risk gate decision (pass/fail) - Recommendation and caveats CONSTRAINTS: - You can READ data and PROPOSE strategies. You cannot deploy or promote. - Every tool call is traced. The human watches live in the Ithaca web UI. - Strategies are declarative specs, not code. Never submit executable code. - If a tool returns an error, report it and retry or pivot. Never fabricate data. - Note the provenance of every data point (source, freshness, coverage). ``` ## How to use it Pick an example prompt from the section below, or write your own using the template. Replace `{symbol}` and `{question}` with your target. Open Codex, Claude Desktop, Cursor, or whichever agent you installed Ithaca into. Paste the prompt. The agent will read the server instructions (from the MCP `initialize` response) and follow the canonical workflow. Open [app.doomberg.me](https://app.doomberg.me) → **Sessions**. Your agent's research session streams live via SSE. Every tool call, argument, and result appears in real time. When the agent finishes, it posts a narrative report in the chat. Review the thesis, evidence, strategy, and backtest stats. If you want to promote the strategy to paper trading, do it from the web UI — the agent cannot. ## Example prompts ### Basic momentum research ```text theme={null} Research NVDA using Ithaca and answer: Is the 60-day momentum thesis still intact, and should I backtest a monthly rebalanced long-only momentum strategy? Follow the canonical Ithaca workflow: 1. Call session_context with this prompt. 2. Research NVDA: market_get_ohlcv (1y daily), market_screener (Technology sector), fundamentals_get, insider_activity, congress_trades, analyst_ratings, technical_indicators, vol_realized. 3. Synthesize a thesis. 4. Call propose_strategy with a rank_returns momentum spec (lookback=60, top=1, monthly rebalance, 5 bps cost). 5. Call backtest_run with range="1y". 6. Report findings with stats and a recommendation. ``` ### Value + insider convergence ```text theme={null} Research AAPL using Ithaca and answer: Are insiders buying while the stock is cheap relative to fundamentals? If so, propose a mean-reversion strategy. Follow the canonical Ithaca workflow: 1. Call session_context with this prompt. 2. Research AAPL: market_get_ohlcv, fundamentals_get, fundamentals_pit (point-in-time), insider_activity, institutional_holdings, analyst_ratings, technical_indicators. 3. Check if insiders are net buyers and if P/E is below sector median. 4. If the thesis holds, call propose_strategy with a mean_reversion spec (lookback=20, z-score entry, monthly rebalance). 5. Call backtest_run with range="2y". 6. Report findings. Note whether insider buying preceded price recoveries historically. ``` ### Short squeeze screen ```text theme={null} Research GME using Ithaca and answer: Is there a short squeeze setup? Check short interest, days-to-cover, FTDs, and recent price action. Follow the canonical Ithaca workflow: 1. Call session_context with this prompt. 2. Research GME: market_get_ohlcv, short_interest, short_volume, short_ftd, short_screener, insider_activity, congress_trades, technical_indicators, vol_realized, market_tide (breadth). 3. Assess squeeze risk: high short interest %, high days-to-cover, rising FTDs, technical breakout, volatility expansion. 4. If a setup exists, call propose_strategy with a rank_returns momentum spec (short lookback=10, top=1, weekly rebalance) to ride the squeeze. 5. Call backtest_run with range="6mo". 6. Report findings. Flag the extreme risk profile and small sample size. ``` ### Macro regime check ```text theme={null} Research the market using Ithaca and answer: What is the current macro regime, and which sectors should I tilt toward? Follow the canonical Ithaca workflow: 1. Call session_context with this prompt. 2. Pull macro data: macro_dashboard, economic_calendar, vix_term_structure, commodity_prices, forex_rates, market_tide, market_movers. 3. Identify the regime (risk-on / risk-off / transition) from breadth, VIX shape, and commodity/FX moves. 4. Screen sectors using market_screener for the regime (defensive vs. cyclical). 5. Call propose_strategy with a sector-rotation spec over the top 2 sectors. 6. Call backtest_run with range="1y". 7. Report the regime, sector tilt, and backtest results. ``` ### Options volatility trade ```text theme={null} Research TSLA using Ithaca and answer: Is implied vol rich or cheap relative to realized vol, and should I backtest a vol-selling strategy? Follow the canonical Ithaca workflow: 1. Call session_context with this prompt. 2. Research TSLA: market_get_ohlcv, vol_realized, vol_term_structure, vol_vrp (variance risk premium), vol_iv_rank, vol_anomaly_score, vol_character (Hurst exponent), technical_indicators, options_greeks (for current ATM). 3. Assess: is IV > RV (vol expensive)? Is IV rank high? Is vol trending or mean-reverting (Hurst)? 4. If vol is rich, call propose_strategy with a mean_reversion spec that captures vol mean-reversion via the underlying. 5. Call backtest_run with range="1y". 6. Report the vol regime, VRP, IV rank, and backtest results. ``` ## Tips for writing your own prompts Be specific about the **symbol** and the **question**. "Research NVDA" is too open-ended — the agent will pull everything. "Research NVDA and assess whether 60-day momentum is intact" gives the agent a clear thesis to test. Name the tools you want if you have a preference. The agent can discover tools via `search_skills`, but naming them (e.g. "check insider\_activity and congress\_trades") ensures they are included. Specify the backtest range. The default is `1y`, but for mean-reversion strategies a longer range (`2y` or `3y`) gives more regime diversity. For momentum, `1y` is usually enough. Ask for provenance in the report. The agent should note if any data is stale or partial — this is part of the server instructions, but reminding it ensures the report flags data quality issues. ## Related The canonical workflow in detail — session\_context, research, propose, backtest, report. Browse the full 60+ tool catalog to name in your prompts. Discover the 50+ data skills — market, fundamentals, congress, insider, vol, options, and more. The declarative strategy schema the agent assembles in step 4.