菜单
Run a Goldman Sachs + Citadel combined fundamental and technical analysis on a specific ticker - a quick analytical deep-dive. Use when the user asks about a specific stock, wants a deep dive, or asks for entry/exit points. For a full bull/bear multi-agent debate hand off to adversarial-research.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Run a parallel bull/bear adversarial research pipeline on a specific ticker. Use when the user asks for "adversarial research", "bull bear analysis", "deep research on [ticker]", a bull-bear debate on X, or wants a rigorous debate-style investment thesis. (Bare "should I buy X?" belongs to pre-trade-check.) Then spawns parallel sub-agents over live data.
Monthly rebalance + DCA prompt for the long-term core ETF sleeve only (broad-market index ETFs) - NOT single-stock adds (use cash-deployment) and NOT max-Sharpe reweighting (use optimize-allocation). Use when the user asks "should I rebalance?", "where should I deploy my paycheck?", "DCA recommendation", "is my allocation drifting?", "auto-invest plan", or on the 1st of each month. Computes target vs actual core allocation, suggests DCA amount per ETF, accounts for tax-account routing.
Plan how to deploy uninvested cash across existing portfolio holdings. Use when the user asks "where do I put my cash?", "I have $X to invest", "deploy my cash", "what should I buy with my settled funds?", or "add to my best names". Routes capital to existing tickers ranked by setup quality and avoids concentration pile-on. Fetches portfolio + technicals + fundamentals via aifolimizer MCP.
Run a one-shot morning portfolio digest that ties together health, alerts, macro regime, crowding risk, concentration, and earnings calendar into a single decision-ready brief. Use when the user asks for "morning briefing", "daily digest", "what's happening today?", "what changed overnight?", or "give me the morning rundown". Composes existing MCP tools - no new data sources.
Harvard Endowment-style dividend income analysis - yield, payout-ratio safety, DRIP projection, and tax-account placement for income holdings. Use when the user asks about dividends, passive income, DRIP, yield, payout ratio, dividend safety, or "what should I hold for income?".
Run a JPMorgan-style pre-earnings analysis on a specific ticker. Use when the user asks about an upcoming earnings report, "should I hold through earnings?", "what's the expected move?", or names a stock and earnings in the same query. Fetches portfolio via aifolimizer MCP.
| name | stock-analysis |
| description | Run a Goldman Sachs + Citadel combined fundamental and technical analysis on a specific ticker - a quick analytical deep-dive. Use when the user asks about a specific stock, wants a deep dive, or asks for entry/exit points. For a full bull/bear multi-agent debate hand off to adversarial-research. |
Every numeric claim (price, P/E, RSI, target, FCF, weight) MUST come from a tool call in THIS run. After fetching, restate the verified figures in a short "Verified data" block and cite ONLY those numbers downstream. If a figure is not in any tool response, say "not available" - never estimate, recall, or invent it. WebSearch is allowed only for narrative (earnings quotes, upgrades), not for numbers that a tool already provides.
Before fetching market data, load prior decisions on this ticker so the verdict stays consistent across sessions:
mcp__aifolimizer__get_ticker_decision_history with ticker=TICKER, max_decisions=5 - prior actions, outcomes, reflectionsmcp__aifolimizer__get_ticker_reflection with symbol=TICKER, n=3 - prior recs + realized alphamcp__aifolimizer__get_cross_ticker_lessons with max_lessons=3 - portfolio-level win/loss patternsReconciliation rule: if a prior decision exists and your new read flips it, state explicitly WHY it changed (new data / catalyst / price move). Never silently contradict a logged decision - that drift is exactly what this prevents.
mcp__aifolimizer__get_profile - account types, cash balances, total capital. Frame tax placement recommendation at endmcp__aifolimizer__get_portfolio - confirm ticker is in portfolio + get cost basis and current weightmcp__aifolimizer__get_fundamentals with symbols=[ticker] - P/E, EPS, dividend yield, market cap, analyst target, institutional ownership, betamcp__aifolimizer__get_technicals with symbols=[ticker] - SMA20/50/200, RSI, MACD, Bollinger Bands, trend signalmcp__aifolimizer__get_news_headlines with ticker=ticker - recent newsmcp__aifolimizer__get_positioning_signals with symbols=[ticker] - crowding score, institutional ownership, short interest, headline velocity. Flag "edge already priced" before issuing buymcp__aifolimizer__get_insider_sentiment with ticker=ticker - insider MSPR (net buying-pressure) trend; feeds fundamental item 6 (insider trend). US-listed onlymcp__aifolimizer__get_finnhub_news with ticker=ticker - news bull/bear tally + net_sentiment; cross-check the get_news_headlines narrative for divergencemcp__aifolimizer__get_recent_filings with ticker=ticker - recent material SEC filings; flag any 8-K filed in the last 5 days as event risk before issuing a call. US-listed onlymcp__aifolimizer__get_factor_exposure with ticker=ticker - dominant style factor (value/momentum/quality/size); use to pick which INVESTOR LENS applies
11b. (US tickers only, when Buffett lens applies) Call mcp__aifolimizer__get_dcf_valuation with symbol=ticker for the FCF history (owner-earnings anchor) and mcp__aifolimizer__get_sec_financials with symbols=[ticker] for the 3-4yr revenue/income/EPS trend (capital-allocation read). Skip for .TO names - EDGAR has no Canadian filingsget_profile - never hardcodeget_insider_sentiment avg_mspr + net_signal (bullish/bearish/neutral)pivot_levels.s1/s2 (support) and pivot_levels.r1/r2 (resistance) from technicals data directly. These are classic floor pivots from the last closed bar. Do NOT invent levels.volume_score (current vol / 20d avg). >1.5 = above-avg conviction, >2.0 = surge, <0.5 = low-conviction move. Buyer vs seller dominancestage (1=basing, 2=uptrend, 3=distribution, 4=decline), minervini_score /7. Score ≥5 = institutional-quality setuppct_from_52w_high and pct_from_52w_low from technicals datatechnical_score /1.0 (0.40×Minervini + 0.25×trend + 0.20×RSI position + 0.10×MACD + 0.05×volume). ≥0.65 = strong setup, 0.45-0.65 = mixed, <0.45 = weakpivot_levels.s1 as initial support; stop-loss below pivot_levels.s2; profit target at pivot_levels.r1 (conservative) or r2 (extended)Apply the 2 most relevant lenses for this stock type. Skip inapplicable ones - state why in one line.
Graham (Deep Value): P/E < 15? Debt/equity < 1? Positive net current assets? 3/3 pass = "Graham would buy at this price." 1-2/3 = note which criteria miss. 0/3 = "Too expensive for value mandate."
Buffett (Quality Moat + Owner Earnings + Management): Three-part read.
profit_margin × revenue / market_cap-style qualitative read.get_dcf_valuation fcf_history (US only). State FCF yield = latest FCF / market_cap as %. If FCF runs persistently below net income, flag "accounting earnings overstate cash - lower quality." If FCF ≥ net income and growing: "earnings are real cash."payout_ratio sustainable (<60% mature co, <80% REIT/utility)? Share count discipline - in get_sec_financials, EPS growing faster than net income = buybacks (good when cheap); EPS lagging net income = dilution (flag). Net insider buying from get_insider_sentiment = alignment. Rate: rational allocator / mixed / value-destroyer.Lynch (GARP): Compute PEG = pe_ratio / (eps_growth_yoy × 100). PEG < 1.0 = undervalued grower, 1.0-2.0 = fairly priced growth, > 2.0 = growth already priced in. State PEG explicitly. If eps_growth_yoy null, state "PEG unavailable."
Druckenmiller (Macro Momentum): Does get_macro_snapshot (rates/CPI/CAD-USD) support this sector's tailwind? Does the chart (stage 2 uptrend + volume confirmation) validate the macro thesis? Risk/reward ≥ 3:1 AND macro + chart aligned → "Druck would size up." Misalignment → "Wait for macro confirmation before entering."
Lens selection guide (use as default, override with judgment):
get_positioning_signalsconsensus_flag=True, downgrade confidence by 1 notch and state "AI/retail consensus already long; late entry has negative expected alpha"contrarian_flag=True AND fundamentals + technicals strong, upgrade confidence by 1 notch>2.0 = retail attention surge, late-cycle; <0.5 = forgotten name, potential setupCall mcp__aifolimizer__log_recommendation with action (BUY/HOLD/SELL/ADD/TRIM), conviction (HIGH/MED/LOW), target_pct + stop_pct (percent from entry - entry captured live at call time), 1-line rationale, skill="stock-analysis". Feeds forward win-rate / track-record loop.
get_fundamentals cached 6h - analyst_target can be stale within trading day; flag if last update >24h.get_technicals cached 1h - entry zones/stop-loss stale on high-volatility days. Mention timestamp.analyst_target from MCP or derive from explicit valuation math you show.minervini_score requires all 7 sub-criteria - if any field null, score invalid; state "incomplete data".pct_from_52w_high/low from technicals - use directly, do NOT recompute from price guess.crowding_score uses 4 weighted signals; if 3+ inputs null (common for small caps/TSX), label unreliable - state "positioning data sparse".pivot_levels null for symbols with <2 trading days of data (new listings, halted). State "pivot data unavailable" rather than guessing.volume_score null when volume data missing (common for some TSX ETFs). Do not comment on volume conviction in that case.get_insider_sentiment / get_recent_filings are US-only (Finnhub/EDGAR) - for .TO tickers they return no_api_key/no_cik; state "US-only data unavailable for TSX name", don't fabricate.get_finnhub_news sentiment is a crude keyword tally, not NLP - use as a tie-breaker, not a primary signal. get_factor_exposure low R² (<0.2) = factor model doesn't fit; skip the lens-selection use.technical_score weights are fixed (40/25/20/10/5). Treat as screening signal, not a precise model output.get_dcf_valuation / get_sec_financials via EDGAR). For .TO names these return empty - fall back to payout_ratio + the cash-flow narrative in fundamental item 2; state "FCF/share-count detail unavailable for TSX name", do NOT fabricate FCF.get_dcf_valuation returns a note when latest FCF is negative - in that case skip the FCF-yield claim and say "owner earnings negative this period; cyclical or reinvestment-heavy - check capex."