菜单
Use this skill when the user asks about anomalies, risk flags, style drift, attribution drift, concentration alerts, or anything unusual in attribution patterns. Also use proactively when generating reports to check for HIGH severity items before presenting findings.
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
Use this skill when the user asks to "prepare an RFP response", "draft an RFP", "respond to an RFP", or needs to compile firm capabilities, track record, investment process, and compliance content for a proposal. Also use when the user mentions "Request for Proposal", "mandate proposal", "tender response", or "client proposal".
Use this skill when the user asks about a borrower's default probability, credit risk drivers, PD scores, risk ratings, or wants to run what-if scenarios on credit metrics like "what if revenue drops 20%" or "recalculate PD with adjusted inputs". Also use when explaining why a borrower received a particular risk rating, SHAP analysis, or comparing risk across borrowers.
Use this skill when the user asks which factors are driving returns, wants to compare factor performance across periods, or requests ad-hoc factor exploration like "which factors are working this quarter", "compare momentum vs value", or "factor correlations". Also use when the user needs analysis beyond what pre-computed factor views provide.
Use this skill when the user wants to optimize portfolio weights, find the efficient frontier, construct a maximum Sharpe or minimum variance portfolio, or rebalance with specific constraints like position caps, sector limits, or ESG minimums. Also use when the user says "what's the optimal allocation" or "rebalance for minimum risk".
Use this skill when the user asks to "generate an attribution report", "prepare attribution for the board", "write up quarterly performance", or wants a comprehensive multi-section attribution narrative. Orchestrates multiple tools into a complete report with audience-appropriate depth.
Use this skill to adapt response depth, formatting, and language to the target audience. Activate when the user specifies an audience like "board memo", "client letter", "PM deep-dive", or when generating reports for different stakeholders, even if they don't explicitly name an audience tier.
| name | attribution-anomaly-scan |
| description | Use this skill when the user asks about anomalies, risk flags, style drift, attribution drift, concentration alerts, or anything unusual in attribution patterns. Also use proactively when generating reports to check for HIGH severity items before presenting findings. |
Trigger when user asks: "anomalies", "risk flags", "attribution drift", "any alerts", "style drift", "concentration risk", "anything unusual", "scan portfolios", "risk scan", "what should I be concerned about"
| Rule | What It Detects | Threshold |
|---|---|---|
| FACTOR_DRIFT_ALERT | Factor exposure >2sigma from trailing average | z-score > 2.0 |
| CONCENTRATION_ALERT | Single sector >60% of active return | share > 0.6 |
| SINGLE_FACTOR_DOMINANCE | One factor >70% of systematic return | share > 0.7 |
| STYLE_INCONSISTENCY | Strategy name vs actual exposure mismatch | Value fund with Growth tilt |
| ATTRIBUTION_SPIKE | Monthly return >2sigma from trailing | z-score > 2.0 |
| ALLOCATION_DRIFT_ALERT | Sector weight >2sigma from 6M rolling average | z-score > 2.0 |
| SELECTION_REVERSAL_ALERT | Selection effect flips sign vs trailing 3M | sign change + magnitude > 10bps |
| WEIGHT_CONCENTRATION_ALERT | Single sector >40% of portfolio weight | weight > 0.40 |
| CLASSIFICATION_SENSITIVITY_ALERT | Sector vs country attribution diverge significantly | divergence > 2% |
Tool: anomaly_detector
Query: Latest month, all portfolios, ordered by anomaly severity (HIGH first)
Format by severity:
HIGH (3+ flags): Full detail with specific flag names, affected sectors/factors, and magnitudes MEDIUM (1-2 flags): Count + most significant flag LOW (0 flags): Count only ("X portfolios have no anomalies")
"Found [N] HIGH severity items across [M] portfolios. I can:
Which would be most useful?"
Tool: brinson_analyzer
Query: Recent 3-6 months for the flagged portfolio to show the attribution trend and when things changed
Tool: brinson_analyzer (rolling analytics via the view)
Query: Month-by-month exposure or weight history showing the drift building
Tool: scenario_sensitivity or backtest_historical_stress
Apply relevant stress scenario to the portfolio with drifted exposures
## Attribution Anomaly Scan — [Date]
### HIGH Severity ([N] portfolios)
| Portfolio | Flags | Key Detail |
|-----------|-------|-----------|
| [Name] | [Flag1], [Flag2], [Flag3] | [Most actionable detail] |
### MEDIUM Severity ([N] portfolios)
- [Portfolio]: [Primary flag] — [1-line summary]
### LOW Severity ([N] portfolios)
No action required.
### Recommended Actions
1. [Highest priority action for most severe item]
2. [Second priority]