Skip to content

Releases

v0.4.1 — P0 核心版本

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 29 Oct 08:27

简体中文

摘要

交付 v0.4.1 P0 核心范围:二维能力路由与共享 evidence/result 契约、A 股单证券研究路径、market_overview_research@2.0 固定清单与 ETF 基础行情。

变更

  • 二维能力路由与共享的 evidence/result 契约。
  • A 股单证券研究路径。
  • market_overview_research@2.0 的固定能力清单,以及明确标记为 unsupported 的全市场槽位。
  • ETF 基础行情。
  • core/installed 两套验证与分发边界。
  • 可选、由环境驱动的 Tushare 能力探测。

边界

  • 全市场筛选在本版明确延期且不受支持:market_screen_research 仍在 source/artifact I/O 之前被拒绝。
  • Universe、factor、ranking、candidate-research 与 Result Artifact 模块保留为证据谱系,不构成公共筛选能力。
  • 延期工单 #133、#139、#140 作为历史 provenance 保留,并标记为本版本不计划交付。未来的筛选需要合格的三所历史 Universe 来源、合格的批量因子 owner、PIT/修订证据、installed 端到端验证与同候选发布证据。

验证

P0/installed 配置 77 项测试通过;Ruff check/format 通过;严格 mypy 1.17.1 覆盖 97 个源文件通过;compileall、Skill 校验与 DAG Definition Index 校验通过。

链接

English

Summary

Delivers the v0.4.1 P0 core scope: two-dimensional capability routing with shared evidence/result contracts, the A-share single-security research path, the market_overview_research@2.0 fixed inventory and ETF base market data.

Changes

  • Two-dimensional capability routing with shared evidence/result contracts.
  • The A-share single-security research path.
  • The market_overview_research@2.0 fixed capability inventory, with whole-market slots explicitly marked unsupported.
  • ETF base market data.
  • core and installed validation and distribution boundaries.
  • Optional, environment-driven Tushare capability detection.

Limitations

  • Whole-market screening is explicitly deferred and unsupported in this version: market_screen_research is still rejected before source or artifact I/O.
  • Universe, factor, ranking, candidate-research and Result Artifact modules are retained as provenance and do not constitute public screening support.
  • Deferred tickets #133, #139 and #140 remain historical provenance and are marked as not planned for this version. Future screening needs qualified three-exchange historical Universe sources, a qualified batch factor owner, PIT/revision evidence, installed end-to-end validation and same-candidate release evidence.

Verification

P0 and installed profiles pass 77 tests; Ruff check/format passes; strict mypy 1.17.1 passes across 97 source files; compileall, Skill validation and DAG Definition Index validation pass.

Links

v0.4.0 — 市场总览 @2.0

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 17 Oct 01:36

简体中文

摘要

将综合市场研究升级为 market_overview_research@2.0,新增可证明的沪、深、北三所交易会话与固定四指数完成日观测,保留唯一公共入口 scripts/entrypoint.py run --request。

变更

  • 综合市场方案保留 15 个能力槽位:4 项 core、8 项 standard 按层级和适用条件执行,全市场成交额、全市场两融及全市场盘中 3 项明确为 unsupported。
  • 在免费匿名来源边界内退出公共筛选、排名、候选深研及其公共结果包工作流。原有单证券盘中快照和完整交易日复盘继续保留。
  • 完成不可变 Operation Evidence、精确归一化、有界 provider lane、时点化 Universe 和 Result Artifact 恢复基础;内部模块不等于新增公共筛选入口。
  • 修复规范内容错误重复、逐应用覆盖、时区比较、依赖归属及跨批次安装事实合并,保留失败关闭和完整原始验证身份。
  • 更新中英文安装入口、变更日志、文档与案例索引,安装命令固定到 v0.4.0 标签。

边界

  • 来源结果保留 complete 15、observed_empty 1、partial 2、credential_blocked 1、unavailable 25。执行覆盖和安装成功不代表全部来源当前可用;日历与四指数证明也不等于全市场数据完整性。
  • 两项盘中实时来源验证按已批准范围豁免,仅具备安装与离线证明;本次发布不等待开盘。
  • 实际来源证明保留冻结提交 2d65e84ae4431fb3fe99bdd0fb765493471410c8,不会改写为发布提交。候选审计中的发布前状态属于其记录时点,正式发布状态以本 Release 和 v0.4.0 标签为准。

验证

  • 已验收实现 e9e129dcc1bf9c070d5566a9585e0e88dadd45ab:Python 3.12 全量 1,706 项测试通过,最终 Standards/Spec 审查均无阻塞。
  • 正式发布整理仅修改仓库文档和对应分发断言,134 个安装文件、运行代码、锁定依赖和原始 JSON 证据保持不变;发布整理的 55 项分发测试、发布矩阵、格式和定义检查通过。
  • 完整清单 46 owners/61 应用;适用来源 44 owners/59 应用全部取得真实执行记录。28 个有效场景,46/46 安装加载,701/701 字段去向。
  • 三所日历与 SSE:000001、SZSE:399001、SZSE:399006、BSE:899050 四指数的真实收盘验证对应 2026-10-16。
  • 发布提交:076ae3e24d9b6d1e973bb07d2455a787fab9dbea。Annotated v0.4.0 标签对象:3673df149e3640c0b40ce57bcc456ce9917e00cc,与远端 main 指向同一发布提交。
  • 从远端标签进行深度 1 干净克隆后:55/55 分发测试通过;实际复制安装 19 项锁定依赖,installation_health 为 complete。验证在禁网隔离环境使用真实锁定 wheel,未调用数据来源。

链接

English

Summary

Upgrades combined market research to market_overview_research@2.0, adds provable SSE, SZSE and BSE trading sessions plus fixed four-index completed-day observations, and keeps the single public entry scripts/entrypoint.py run --request.

Changes

  • The combined market plan keeps 15 capability slots: four core and eight standard execute by tier and applicability, while whole-market turnover, whole-market margin and whole-market intraday are explicitly unsupported.
  • Public screening, ranking, candidate deep research and their public result-package workflows exit within the free anonymous source boundary; single-security intraday snapshots and complete-session replay remain.
  • Complete immutable Operation Evidence, exact normalisation, bounded provider lanes, time-bound Universe and Result Artifact recovery foundations; internal modules are not new public screening entries.
  • Fix duplicate canonical-content records, per-application coverage, timezone comparison, dependency ownership and cross-batch installation fact merging, keeping fail-closed behaviour and full original verification identity.
  • Update bilingual install entry points, changelog, docs and example index, pinning install commands to the v0.4.0 tag.

Limitations

  • Source results remain complete 15, observed_empty 1, partial 2, credential_blocked 1, unavailable 25. Execution coverage and a successful install do not mean every source is currently available; the calendar and four-index proof is not whole-market data completeness.
  • Two intraday live-source checks are exempted under the approved scope and carry install and offline proof only; this release does not wait for a market open.
  • The actual source proof keeps the frozen commit 2d65e84ae4431fb3fe99bdd0fb765493471410c8 and is not rewritten as the release commit. Pre-release state in the candidate audit belongs to its recording time; the formal state is this Release and the v0.4.0 tag.

Verification

  • Accepted implementation e9e129dcc1bf9c070d5566a9585e0e88dadd45ab: Python 3.12 full suite 1,706 tests pass, final Standards/Spec review unblocked.
  • Release packaging changed documentation and the corresponding distribution assertions only; 134 installed files, runtime code, locked dependencies and raw JSON evidence are unchanged. The 55 distribution tests, release matrix, format and definition checks pass.
  • Full inventory 46 owners / 61 applications; applicable sources 44 owners / 59 applications all have real execution records. 28 valid scenarios, 46/46 installed loading, 701/701 field provenance.
  • Real close verification for the three-exchange calendar and SSE:000001, SZSE:399001, SZSE:399006, BSE:899050 corresponds to 2026-10-16.
  • Release commit 076ae3e24d9b6d1e973bb07d2455a787fab9dbea; annotated v0.4.0 tag object 3673df149e3640c0b40ce57bcc456ce9917e00cc points at the same release commit as remote main.
  • A depth-1 clean clone of the remote tag passes 55/55 distribution tests; the copied install resolves 19 locked dependencies with installation_health reporting complete. Verification used real locked wheels in a network-isolated environment and called no data sources.

Links

v0.3.0 — 统一入口与 12 类任务

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 18 Aug 08:29

简体中文

摘要

将 A 股研究能力收敛到唯一公共入口 scripts/entrypoint.py run --request,统一交付 12 类版本化 ResearchTask、6 个固定研究方案和 42 份可执行请求示例。

变更

  • 新增完整交易日分时复盘,以及证据约束的 Agent 层情景分析边界。
  • 新增综合单标的与综合市场研究方案;局部失败、空集、冲突和不支持能力均保留在统一结果中。
  • 完成研究内容、资金事件和市场信号的 owner-local 契约闭合,保留身份、时点、单位、分页、截断、空集和字段去向。
  • 默认安装锁定正式运行依赖,并提供独立的安装健康检查。
  • 综合市场方案明确保留 5 项 v0.3.0 不支持能力,不使用单只证券、指数或榜单伪造全市场事实。
  • README、Skill 引用文档、中文案例和 docs/research 已统一到 v0.3.0。

验证

  • Python 3.12 全量测试:1066/1066。
  • 发布请求画像:27/27。
  • 正式来源 operation:44/44 已实际尝试。
  • 默认安装加载:44/44。
  • active-request 字段去向:740/740。
  • 远端 v0.3.0 标签干净克隆验证:78/78。
  • Ruff、严格 mypy、Skill 校验、compileall、release matrix 与差异检查全部通过。

链接

English

Summary

Collapses A-share research behind the single public entry scripts/entrypoint.py run --request, delivering 12 versioned ResearchTasks, six fixed research plans and 42 executable request examples.

Changes

  • Add complete-trading-day intraday replay plus an evidence-constrained Agent-layer scenario boundary.
  • Add combined single-security and combined market research plans; partial failures, empty sets, conflicts and unsupported capabilities stay in the unified result.
  • Close owner-local contracts for research content, capital events and market signals, preserving identity, as-of time, units, pagination, truncation, empty sets and field provenance.
  • Lock production runtime dependencies on default install and provide a separate installation health check.
  • The combined market plan explicitly keeps five v0.3.0-unsupported capabilities and never fabricates whole-market facts from a single security, index or ranking.
  • README, Skill reference docs, Chinese examples and docs/research are unified on v0.3.0.

Verification

  • Python 3.12 full suite: 1066/1066.
  • Release request profiles: 27/27.
  • Production source operations: 44/44 actually attempted.
  • Default install loading: 44/44.
  • active-request field provenance: 740/740.
  • Clean clone of the remote v0.3.0 tag: 78/78.
  • Ruff, strict mypy, Skill validation, compileall, release matrix and diff checks all pass.

Links

v0.2.0 — 研究级盘中行情快照

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 03 Aug 15:04

简体中文

摘要

新增通过稳定 entrypoint.py run --request seam 提供的研究级盘中行情快照。

变更

  • 新增 intraday_market_signal,只接受当前北京时间交易日的一只规范 SSE/SZSE A 股。
  • 使用通达信同一观测作为快照基准,腾讯作为必须成功的独立核心价格交叉观测。
  • 明确开盘/收盘集合竞价、连续竞价、午间休市、六十秒新鲜度、停牌与昨收语义。
  • 任一来源失败、陈旧、错证券、会话或价格类型不兼容、核心价格冲突时失败关闭。
  • 两个来源 operation 仍为实验来源,正常结果最高为 limited,单次 live probe 不构成生产准入。
  • mootdx==0.11.7 仅为盘中与 F10 能力范围内的锁定可选依赖;缺失时只阻断相关能力。
  • 中英文 README 固定安装到 v0.2.0 标签,新增不保存来源响应的盘中请求案例和独立发布审计。

验证

发布提交:697f0db。发布前验证 54 项聚焦测试、521 项 Python 3.12 全量离线测试、Ruff(138 个文件)、strict mypy(48 个源文件)、Skill validator 和 diff check;远程标签干净克隆的盘中/分发测试、strict mypy 与 Skill validator 均通过。

English

Summary

Adds a research-grade intraday market snapshot behind the stable entrypoint.py run --request seam.

Changes

  • Add intraday_market_signal, accepting only one canonical SSE/SZSE A-share for the current Beijing-time trading day.
  • Use the same TDX observation as the snapshot basis, with Tencent as a mandatory successful independent core-price cross-check.
  • Define opening/closing auction, continuous trading, lunch break, sixty-second freshness, suspension and previous-close semantics.
  • Fail closed on source failure, staleness, wrong security, incompatible session or price type, or core-price conflict.
  • Both source operations remain experimental with normal results capped at limited; a single live probe is not production admission.
  • mootdx==0.11.7 remains a locked optional dependency for intraday and F10 scope only; when missing it blocks only those capabilities.
  • Pin the bilingual README install to the v0.2.0 tag and add an intraday request case that stores no source responses plus a standalone release audit.

Verification

Release commit: 697f0db. Before release, 54 focused tests, 521 Python 3.12 offline tests, Ruff (138 files), strict mypy (48 source files), the Skill validator and diff check were re-verified; the clean-clone intraday/distribution tests, strict mypy and the Skill validator on the remote tag all pass.

v0.1.1 — 呈现与发布修订

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 03 Aug 15:03

简体中文

摘要

兼容的文档与 Agent 呈现修订,不改变公共 ResearchTask CLI、schema_version: "1.0"、计算方法或来源资格。

变更

  • 明确事实、Agent 推断、研究判断、条件触发位、署名观点与投资行动建议的边界。
  • 精简安装 Skill 入口,并补充中英文安装、F10 与跨平台 Python 3.12 说明。
  • 将 mootdx==0.11.7 保持为 F10 能力范围内的锁定可选依赖。
  • 更新蓝色光标与工业富联案例,展示 limited 结果下仍需保留的证据、风险和失效条件。
  • 修复 strict mypy 对可选整数字段的类型收窄,不改变运行行为。

验证

发布提交:0200fe1。发布前重新验证 467 项 Python 3.12 离线测试、Ruff、strict mypy(44 个源文件)、Skill validator;远程标签干净克隆的分发测试、strict mypy 与 Skill validator 均通过。

English

Summary

Compatible documentation and agent-presentation revision; the public ResearchTask CLI, schema_version: "1.0", calculation methods and source eligibility are unchanged.

Changes

  • Make the boundary between facts, agent inference, research judgement, conditional triggers, attributed opinion and investment action explicit.
  • Simplify the Skill install entry and add bilingual install, F10 and cross-platform Python 3.12 guidance.
  • Keep mootdx==0.11.7 as a locked optional dependency within F10 scope.
  • Update the BlueFocus and Foxconn cases to show which evidence, risks and invalidation conditions must be retained for limited results.
  • Fix strict mypy narrowing for optional integer fields without changing runtime behaviour.

Verification

Release commit: 0200fe1. Before release, 467 Python 3.12 offline tests, Ruff, strict mypy (44 source files) and the Skill validator were re-verified; the clean-clone distribution test, strict mypy and the Skill validator on the remote tag all pass.

v0.1.0 — A股研究技能

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 02 Aug 18:54

简体中文

摘要

首个完整产品版本,交付 evidence-first 的 A 股研究任务与四套固定研究流程。

变更

  • 行情与估值:最近 N 日双源 OHLCV、ETF 行情、自动单票估值和 2–10 标的同口径比较。
  • 研究内容:个股/行业/主题研报、财务材料、F10、新闻公告、市场快讯和互动易。
  • 资金与事件:北向披露边界、个股/板块资金流、龙虎榜、解禁、两融、大宗、股东户数和分红。
  • 市场信号与衍生品:题材、板块、行业轮动、涨跌停、监控、异动、热度,以及四类 ETF 期权。
  • 四套流程:单票估值、批量对比、主题研报、新标的八步研究。

边界

  • 证券身份、研究日期、公开时间、获取时间、单位、口径、来源定位和计算谱系均进入版本化 JSON。
  • 无合格独立 fallback 时失败关闭;空响应、schema 漂移和覆盖不完整不会被解释为“没有数据”。
  • 免费来源仍是实验 operation,结果最高为 limited;Release 不把它们描述成已独立核验事实。
  • 主题研报有无凭据的东财标题关键词基线;iWencai 仅为可选增强。F10 需要可选的 mootdx。

安装与升级

git clone --depth 1 --branch v0.1.0 --single-branch https://github.com/RedHeartSecretMan/a-share-research-skill.git

验证

  • 464 项离线测试通过。
  • Ruff check/format、strict mypy(44 个源码文件)、Skill validation、diff check 通过。
  • 23 个能力场景、四套流程、蓝色光标与工业富联 Demo 完成真实来源 smoke。
  • 从公开远程 v0.1.0 tag 干净克隆,复制完整 Skill 后通过校验与 Python 3.12 隔离 CLI smoke。

详细能力、限制与现场哈希见仓库的 v0.1.0 发布审计。

English

Summary

First complete product version: evidence-first A-share research tasks and four fixed research flows.

Changes

  • Prices and valuation: dual-source OHLCV for the last N sessions, ETF quotes, automatic single-security valuation and same-basis comparison across 2–10 securities.
  • Research content: single-stock, industry and theme reports, financial material, F10, news and announcements, market flashes and investor Q&A.
  • Capital and events: northbound disclosure boundaries, single-stock and sector fund flows, dragon-tiger lists, lock-up releases, margin trading, block trades, shareholder counts and dividends.
  • Market signals and derivatives: themes, sectors, industry rotation, limit moves, monitoring, unusual moves, heat, and four ETF option classes.
  • Four flows: single-security valuation, batch comparison, theme research, and the eight-step new-security flow.

Limitations

  • Security identity, research date, publication time, retrieval time, units, basis, source location and calculation lineage all enter versioned JSON.
  • Fail-closed without a qualified independent fallback; empty responses, schema drift and incomplete coverage are never reported as "no data".
  • Free sources remain experimental operations with results capped at limited; this release does not describe them as independently verified facts.
  • Theme research keeps an unsourced Eastmoney headline-keyword baseline; iWencai is optional enhancement only. F10 requires the optional mootdx.

Install and upgrade

git clone --depth 1 --branch v0.1.0 --single-branch https://github.com/RedHeartSecretMan/a-share-research-skill.git

Verification

  • 464 offline tests pass.
  • Ruff check/format, strict mypy (44 source files), Skill validation and diff check pass.
  • 23 capability scenarios, the four flows and the BlueFocus/Foxconn demos complete real-source smoke runs.
  • A clean clone from the public v0.1.0 tag, with the full Skill copied in, passes validation and an isolated Python 3.12 CLI smoke.

A detailed capability, limitation and live-hash record is in the repository's v0.1.0 release audit.

v0.0.1 — 可信证据内核预览

Choose a tag to compare

@RedHeartSecretMan RedHeartSecretMan released this 02 Aug 08:54

简体中文

摘要

可信证据内核预览:用可复核的证据链解析 A 股证券身份、最近已完成交易日收盘价与单票估值。

变更

  • 通过上交所/深交所与巨潮实验来源交叉解析 A 股证券身份。
  • 通过交易所日线与腾讯日线交叉核验最近已完成交易日的未复权收盘价。
  • 校验调用者提供的研究证据包。
  • 根据合格证据计算总市值、PE TTM 与 PB MRQ,并保留计算谱系。
  • 以 scripts/entrypoint.py 作为 Skill 唯一公共运行入口。
  • 蓝色光标与工业富联两个身份及单日收盘价联网验证案例。

边界

该版本不会从代码前缀或 opaque orgId 猜测交易所。实验来源结果最高为 limited;身份歧义、来源冲突或关键证据不足时返回 blocked。它尚不包含完整行情序列、研报、资金、筹码、公告、市场信号或综合个股研究能力,不代表产品 v0.1.0。

English

Summary

Evidence-kernel preview: an auditable evidence chain for A-share security identity, the last completed session's close price and single-security valuation.

Changes

  • Cross-resolve A-share security identity from SSE/SZSE and CNINFO experimental sources.
  • Cross-check the last completed session's unadjusted close against exchange daily bars and Tencent daily bars.
  • Validate the caller-supplied research evidence package.
  • Compute total market cap, PE TTM and PB MRQ from qualified evidence while keeping the calculation lineage.
  • Use scripts/entrypoint.py as the Skill's only public runtime entry point.
  • Two networked identity and single-day close cases: BlueFocus and Foxconn Industrial Internet.

Limitations

This version never infers an exchange from a code prefix or an opaque orgId. Experimental sources cap results at limited; identity ambiguity, source conflicts or missing key evidence return blocked. It does not yet cover full price series, research reports, capital flows, chips, announcements, market signals or comprehensive single-security research, and is not product v0.1.0.