Wind Alice Fund Compare

Problem

Public fund comparisons often stop at net-value curves and manager bios. Practitioners still need a defensible way to decide whether Fund A should replace Fund B, which peer fund offers a better risk-adjusted trade-off, or what quantitative evidence should go into portfolio review and due diligence. This skill targets that fund-research workflow. It uses Wind Alice terminal data to produce a structured comparison report for 2–5 public funds, rather than a generic narrative.

How It Works

The core flow converts the user’s intent into a natural-language request, then calls the Wind Alice Agent through the alice-fund-compare CLI:

• Input: fund names or codes (for example, 000001.OF) plus the comparison angle, such as risk-return fit or holding concentration.
• Comparison dimensions: performance, risk metrics, portfolio structure, and management evaluation.
• Output: the agentResult.value payload, typically including key conclusions, fund-specific strengths, main risks, allocation advice, and comparison tables.
• Key controls: use --prompt to pass the question, --no-wait to let the CLI poll until the task reaches a terminal state, -d to choose the report attachment directory, and --continue-session only when follow-up context should be reused.

The report is research-oriented. It supports objective analysis, and it can also emphasize one side when the prompt asks a directional question such as “why is A better than B?” Data limitations are explicitly noted in the affected sections.

Boundaries

• Best suited to comparing 2–5 public funds; depth may decline beyond three funds.
• Full report files may be downloaded silently. When delivering results, present only agentResult.value; do not paste attachment content.
• Requires Node.js 18+ and WIND_API_KEY. Prefer storing the key in ~/.wind-alice/config.env; by default the CLI does not read environment variables.
• On Windows, if Chinese output appears garbled, read logs and reports as UTF-8 first; do not retry by changing the prompt or manually polling tasks.