Stock Screener

WorkBuddy PandaData 数据源覆盖

本技能在 WorkBuddy 专家包中运行时，所有实时和历史金融数据必须来自已连接的 pandadata Connector。若下文、参考资料或脚本提到 Python panda_data SDK、 AkShare、Tushare、网页抓取、直连 HTTP 或本地凭证，以本节为准：不得使用这些 方式获取正式数据。

1. 数据型任务必须先完成 auth_status 和至少一次真实的 call_pandadata 业务调用；在收到 Connector 返回前，禁止输出分析、排名、数字结论或“无数据”。本 Skill 的流程不得绕过 主 Agent 的最低接口调用清单。
2. 先按本 Skill 的 references/pandadata-interface-contracts.md 选择已登记方法和参数。只要任务能映射到已登记接口， 就直接通过 call_pandadata 传入该业务方法和 params；常规调用前不得执行接口检索。
3. 仅当已登记方法被 Connector 明确报告为参数契约不兼容、字段契约变化或调用失败时， 才对该方法调用一次 get_method_doc，修正参数后最多重试一次。
4. 仅当本地接口表没有匹配项，或 Connector 明确报告方法不存在/不受支持时，才调用 search_methods 动态发现接口；不得靠猜测连续试用名称相近的 get_* 方法。
5. 只有 call_pandadata 实际返回 0 行时才允许写“无数据”。必须先完成一次复查调用：校验 最新交易日与代码格式，放宽日期窗口，移除非必填过滤条件，或使用登记的备用参数；仍为 0 行才可如实报告，并保留两次调用回执。0 行不得触发 search_methods。
6. 不向 call_pandadata 添加未登记参数或顶层行数限制；记录实际方法、参数、数据日期、 频率、复权口径、行数、空值和错误状态。
7. 包内脚本只可处理 Connector 已返回的数据或执行纯本地计算与校验，不得自行联网取数。

最终答案必须包含“数据调用回执”表：接口、实际参数、状态、行数、数据日期范围和关键字段。 缺少回执表示任务未完成，必须继续调用工具而不是结束回答。

权限不足、配额限制、空结果、延迟发布和字段缺失都必须明确披露，不得切换到其他数据源 或用模型推断补数。

{
  "version": 1,
  "task": {
    "placeholder": "请输入自然语言选股条件，例如：沪深300 中连续三年分红、低估值且排除高质押股票",
    "required": true
  },
  "fields": [
    {
      "key": "universe",
      "label": "股票池",
      "type": "select",
      "default": "000300.SH",
      "options": [
        { "value": "000300.SH", "label": "沪深300" },
        { "value": "000905.SH", "label": "中证500" },
        { "value": "399006.SZ", "label": "创业板指" },
        { "value": "000852.SH", "label": "中证1000" },
        { "value": "000016.SH", "label": "上证50" },
        { "value": "all_a", "label": "全部 A 股" }
      ]
    },
    {
      "key": "date",
      "label": "筛选日期",
      "type": "date",
      "help": "留空时使用最近可用交易日；历史筛选严格按当时已披露数据"
    },
    {
      "key": "limit",
      "label": "结果数量",
      "type": "number",
      "default": "20"
    }
  ],
  "prompt_template": "{{#task}}任务与材料：\n{{task}}\n\n{{/task}}{{#attachments}}用户上传的材料（已放入工作区）：\n{{attachments}}\n\n{{/attachments}}在 {{universe}} 股票池中解析并执行自然语言 A 股筛选条件。{{#date}}以 {{date}} 为筛选时点。{{/date}}未指定日期时以最近可用交易日为筛选时点。将条件拆为指标、运算符、阈值、窗口和排序规则，严格避免前视偏差；逐层记录接口、参数、数据截止日、剩余数量与缺失值，最多返回 {{limit}} 只具备实际命中值和来源证据的股票，输出中文报告。"
}

Stock Screener

Use this skill to turn natural-language A-share screening requirements into a reproducible Pandadata query plan, execute the filters layer by layer, and return a sourced stock list with the actual values behind every matched condition.

Creator, Maintainer, And Scope

• Creator: abgyjaguo (https://github.com/abgyjaguo).
• Maintainer: abgyjaguo for the QuantSkills community.
• Repository: https://github.com/quantskills/skill-stock-screener.
• License: GNU General Public License v3.0 only (GPL-3.0-only).
• Scope: research-oriented A-share screening from public Pandadata interfaces. The skill is not official investment advice, a certified data product, or a guarantee of screening performance.

Core Rules

• Use the local pandadata skill for every real data call. Before calling an API, inspect that skill's method index or search helper to confirm exact parameters, fields, date conventions, and return shape.
• Never invent Pandadata methods, fields, credentials, factor definitions, or unsupported screening conditions. If a condition cannot be mapped to documented data, say so and offer the nearest auditable proxy.
• Treat screening date as a first-class input. Default to the latest available trading day, but ask for clarification when the user needs a historical point-in-time screen.
• Avoid look-ahead bias. For financial statements, dividends, forecasts, pledge data, and shareholder events, use only rows whose disclosure or announcement date is not later than the screening date.
• Keep every output reproducible: preserve original user criteria, normalized atomic filters, method names, parameters, data cutoff dates, row counts, and missing-data notes.
• End formal screening reports with: 本筛选结果基于公开数据与规则化条件生成，仅供研究参考，不构成任何投资建议。

Workflow

1. Parse the request into atomic filters: metric, operator, threshold, time window, universe, date basis, and ranking or limit rule. Restate ambiguous business meaning before execution, such as whether 连续3年分红 means three fiscal years with cash dividends or three calendar years with ex-dividend events.
2. Build the starting universe with get_trade_list, then apply explicit universe filters such as exchange, board, industry, concept, index membership, ST exclusion, listing-age, suspension, or user-provided symbols.
3. Read references/screener-guide.md when planning any non-trivial screen. Use it for the condition map, execution order, output schema, and QA checklist.
4. Sort filters by selectivity and API cost. Prefer bulk universe/status/industry filters first, then bulk market and financial filters, then per-symbol or sparse event calls such as pledge, unlock, shareholder change, and top-holder checks.
5. Smoke-test each unfamiliar method on one date range or a small symbol set before full-market execution. Record field names and row counts before expanding.
6. Execute filters layer by layer. After each layer, record the remaining stock count, eliminated count, API method, parameters, data date or report period, and any skipped rows.
7. Return a compact Chinese Markdown report by default: criteria interpretation, screening funnel, final table, evidence columns, missing-data caveats, saved result path when a file is created, and the fixed disclaimer.

Common Method Map

Output Standards

• Include only stocks that can be traced to evidence. If a final name remains because data is missing rather than passing a condition, mark it as 数据缺失 and keep it out of strict-pass counts unless the user explicitly allows missing-data inclusion.
• Show actual values, not only pass/fail flags. Use columns such as 代码, 名称, 条件命中值, 报告期/数据日, 方法, and 备注.
• Save machine-readable results only when useful for reruns or when the user asks. Use screens/<YYYY-MM-DD>-<slug>.json in the user's working project, not inside the installed skill folder unless the skill folder is also the project workspace.
• For ranked screens, separate hard filters from ranking metrics. State tie-breakers, sort direction, and whether the result is top N, percentile, or threshold based.

Resource Guide

• references/screener-guide.md: detailed condition-to-method map, normalized filter schema, default execution order, JSON result schema, and QA checklist.

Cross-Agent Use

• Codex and Claude Code can load this folder directly as a skill named $stock-screener through SKILL.md.
• Cursor should use agents/cursor-rule.mdc as the project rule adapter and keep the full skill folder under .cursor/skills/stock-screener.
• Hermes and OpenClaw should use agents/portable-loader.md when they do not natively discover SKILL.md folders.