Agent Skills
› openai/plugins
› fund-screener
fund-screener
GitHub基于Morningstar数据筛选基金和ETF，支持按类别、评级、费用等条件过滤。包含严格的数据合规限制，执行标准化查询、结果验证与排序，并生成包含披露信息和优化建议的结构化报告。
触发场景
用户需要根据特定财务指标筛选基金或ETF
用户希望获取符合特定风险或收益标准的投资标的列表
安装
npx skills add openai/plugins --skill fund-screener -g -y
SKILL.md
Frontmatter
{
"name": "fund-screener",
"description": "Use when screening funds or ETFs by Morningstar category, ratings, fees, assets, returns, or risk."
}
Fund Screener
Screen funds using the connected Morningstar app as the data source.
Guardrails
- Use only data returned by the Morningstar app in the current session.
- Do not infer missing values, create synthetic scores, predict performance, or give investment advice.
- Show unavailable values as
N/Aand distinguish missing data from tool failure. - Supported universes are ETFs, open-end funds, and closed-end funds.
- Treat filters as AND logic unless the user explicitly asks for alternatives; run separate passes for OR logic and deduplicate results.
Workflow
Before running a real screen, read references/full-workflow.md. It preserves Morningstar's partner-authored rules for criteria confirmation, normalization, datapoints, result validation, output tables, disclosures, and follow-up suggestions.
- Collect the screening criteria in one pass: universe, category, medalist rating, star rating, expense ratio, assets, and any user-specified filters.
- Normalize user terms against Morningstar-supported datapoints and values before screening. If a close match is likely, ask for confirmation before running.
- Run the screen, targeting a useful result set of roughly 10 to 20 funds.
- Validate fund status and remove inactive, merged, or liquidated funds.
- Enrich surviving results with category, active/passive status, benchmark, expense ratio, assets, inception date, medalist rating, star rating, returns, category ranks, and risk metrics when available.
- Rank results by the user's priority. If no priority is given, sort by assets descending and then expense ratio ascending.
Output
Use this order:
- Disclosure banner.
- Criteria used table, one row per active filter only, including normalized terms.
- Result count and exclusions.
- Snapshot table.
- Performance table.
- Risk and category-rank table.
- Three concise, data-backed refinements such as tighter expense ratio, higher rating floor, narrower category, or universe changes.
If the screen returns zero or too few results, identify the likely binding criteria and ask what to relax.
版本历史
- 11c74d6 当前 2026-07-19 09:40
