@v587d/capital-generation 2.7.4 → 2.7.6

package/selected-skills/trading-skills/UPSTREAM.md

@@ -31,18 +31,21 @@
      两者都不进 catalog、也不进发布包（见 Packaging），按"不进 catalog 的正文一律不动"原样保留。
   2. **删除上游的 `.gitignore`。** 被排除的 ignore 规则会在本仓继续生效，并把"按 commit 重导"的快照
      与所钉 commit 弄成两种样子（InvestSkill 那一轮就是三个文件被静默挡掉）。vendored 快照不把 ignore 规则带进宿主仓库。
-  3. **取数与执行面口径改写**：8 份 `SKILL.md` + 2 份 `references/calculation-helpers.md`，
-     共 17 处句子级改动（+34 / −13 行）。分析结论、判定档位词汇（`must-watch` / `fragile` / `too concentrated` 一类）、
+  3. **取数与执行面口径改写**：10 份 `SKILL.md` + 3 份 `references/calculation-helpers.md`，
+     共 20 处句子级改动（+46 / −20 行）。分析结论、判定档位词汇（`must-watch` / `fragile` / `too concentrated` 一类）、
      输入清单、输出顺序与 `references/` 的框架、阈值文档一字未改。
      本颗**没有** `web search` 与记忆回填那一族（`web search` 零命中；每颗都写"缺什么就点名缺、别编"，
      `example-mode` 指使用者自带的示例数据并要求标注，不是逃生口），换的是另外三种形态：
-     - **谁去取数（6 处）**：`catalyst-map`、`watchlist-review`、`evidence-gap-check`、`thesis-validation`、
+     - **谁去取数（7 处）**：`catalyst-map`、`watchlist-review`、`evidence-gap-check`、`thesis-validation`、
        `execution-plan-check`、`portfolio-concentration` 各有一句 `Do not fetch live data unless the user
        explicitly asks to pair this skill with …`。前半句的意图（别自己去堆一堆数据）保留，
-       后半句改成"确实缺一个日期或一个字段时，把标的、字段、as-of 日期写成 brief 交给宿主取数通道——
+       后半句改成"确实缺一个日期或一个字段时，把标的、字段、as-of 日期写成交给宿主取数通道——
        角色分流的宿主里就是负责检索与行情的角色——用回来的东西继续，剩下的照旧标为未知"。
        顺带这三句原本点名的 `earnings-preview` / `macro-event-analysis` 不在可选清单里（见下面的收录判断）。
-     - **没有 shell 就算不了（2 + 2 处）**：`journal-pattern-analyzer:49` 与 `risk-reward-sanity-check:52` 的
+       本轮收录 `position-management` 时同一句按同一改法处理：这颗原本靠这句守住"不自己堆数据、
+       要不要拉一份行情是使用者点的事"，闸门只认 `fetch live` 与 `pair this skill` 两个词形，
+       所以改完必须自己核对语义还在——现在它写的是"这份复查不出网，缺什么由主 Agent 决定要不要委派去取"。
+     - **没有 shell 就算不了（2 + 4 处）**：`journal-pattern-analyzer:49` 与 `risk-reward-sanity-check:52` 的
        `For agents that support code execution, use [references/calculation-helpers.md] …` 改成
        "本插件的分析角色没有 shell，也不跑 helper 脚本：统计量交给宿主数据侧（角色分流的宿主里委派做统计与质量校验的角色），
        三笔以内可以在答复里手算并把算式写出来，缺输入的条目留空不补默认值"；
@@ -50,29 +53,40 @@
        改写成"这些量是什么、怎么取"（公式按上游 `_lib/calculations.py` 的实现核对：
        `risk_reward_ratio` = |target−entry| / |entry−stop| 每单位、`expectancy` = 胜率×均盈 − 败率×均亏，
        `kelly_fraction` 只是参照值，不作为交给使用者的仓位）。
+       本轮收录 `position-sizing` 时同一族又落两处：它的 `SKILL.md:52` 与 `references/calculation-helpers.md:3`
+       原本也是那句"能跑代码的宿主去用共享模块"——这颗自己不带取数，所以要改的只有这一处形状：
+       按 `references/methodology.md` 的基础算式（`position size = allowed trade risk / per-unit realized risk`）
+       在答复里把式子写出来，波动率一类统计量交给 `data_junior`。
      - **点名的下一步 skill 未必存在（7 处）**：`catalyst-map`、`watchlist-review`、`evidence-gap-check`、
        `thesis-validation`、`journal-pattern-analyzer`、`execution-plan-check`、`portfolio-concentration` 的
        `Next Skill` 清单原句保留，各加同一句前置条件与退路（InvestSkill 那轮的改法）：
-       点名的 skill 只有使用者显式开启才存在于会话里，本插件明确不提供的那两类（下单、给仓位）永远不存在，
-       只能说成选项、不能替使用者跑。
+       点名的 skill 只有使用者显式开启才存在于会话里，本插件明确不提供的那一类（替使用者下单与改单）永远不存在，
+       只能说成选项、不能替使用者跑。这句原本写的是"（下单、给仓位）两类"——本轮起仓位测算已收录，
+       留着就成了假话，七处与第六颗同一句一起改成执行面口径。
      **没有新增 `Retrieval:` 出处档**：上游根本没有这个枚举（§8.6 第 3 条不适用），
      不为了套前几颗的改法新造上游没有的模板字段。
 
-- 收录判断（19 颗里 9 颗进 catalog，10 颗留在快照里当可审查材料）：
+- 收录判断（19 颗里 11 颗进 catalog，8 颗留在快照里当可审查材料，其中 `etoro` 连发布包都不进）：
   - **进**：`catalyst-map`、`watchlist-review`、`evidence-gap-check`、`thesis-validation`、
     `risk-reward-sanity-check`、`execution-plan-check`、`portfolio-concentration`、
-    `journal-pattern-analyzer`、`post-trade-review`。
+    `journal-pattern-analyzer`、`post-trade-review`，本轮再加 `position-sizing` 与 `position-management`。
     `post-trade-review` 一字未动就过闸门（全文无取数、无脚本、无出图，只在 `references/review-rubric.md`
     里举了一条"高冲击数据日把仓位减 25%"的过程规则例子 → `riskNote` 写明属于过程推演）。
+    新收那两颗的输出形状：`position-sizing` 索取 account equity / risk budget / entry / stop，给
+    `Position Recommendation`（股数、合约数、占账户风险的百分比）；`position-management` 复查在手持仓，
+    给 `hold, trim, tighten risk, close, or wait` 与 `Best Next Action`。两颗都不取数、不碰凭据、不产文件，
+    所以要动的只有上面那处"能跑代码的宿主"退路句与那一句护栏，判定逻辑一字未改。
+  - **本轮判据纠正的记录**：这两颗当初按"给仓位数字与持仓加减动作"整颗不收，写的理由是
+    「评级、买卖动作和仓位决策属于明确冲突，不能直接集成」——那是把 §8.6 第 7 条读成了内容禁令。
+    本插件不收的从来是**执行面**：没有券商接口，不替使用者下单或改单（第 4 条；`etoro` 至今按这条不收，
+    也不进发布包）。仓位数字与加减建议在正文里属于方法推演，收录之后由 catalog 的 `riskNote` 承担口径，
+    正文的"不据它执行"那句也跟着写在 `Next Skill` 前置条件里。同一条误读在第六颗挡掉三颗
+    （见 `selected-skills/claude-for-financial-services-cn/UPSTREAM.md`），两处一起纠正。
   - **不进 · 执行面**：`etoro`。它是券商公开 API 的操作手册：`SKILL.md:18` "**Keys (request from the user on install)**"
     与逐步生成密钥的说明、`x-api-key` / `x-user-key` 请求头、`curl -X POST …/trading/execution/…market-open-orders/by-amount`
     带 `"IsBuy": true, "Leverage": 1`、"Use **non-demo execution endpoints** for real trading"、按 `positionId` 平仓，
     末尾还指到一个 MCP server。§8.6 第 4 条（执行面）与第 7 条（真实下单）双重排除；本插件不接券商、不下单、
     凭据只进 credentials 域（§5）。这颗连发布包都不进（`.npmignore`），全量仍住在 git 里。
-  - **不进 · 仓位与买卖动作**：`position-sizing`（索取 account equity，输出 `Position Recommendation`：
-    股数／合约数／账户风险百分比）、`position-management`（`hold, trim, tighten risk, close, or wait` + `Best Next Action`）。
-    与 `docs/reference/external-skill-scan-2026-10.md` 观察名单里既写的同一口径：
-    「评级、买卖动作和仓位决策属于明确冲突，不能直接集成」。
   - **不进 · 接管会话路由**：`pre-trade-check`、`earnings-trade-prep`、`portfolio-risk-review`、`post-trade-debrief`
     四颗 workflow。上游自己的 "Workflow routing" 段按结果分流到 11–12 颗子 skill 并给出
     `ready` / `reject` / `hold through with conditions` 一类决断，与本插件的角色路由冲突
@@ -85,9 +99,9 @@
     另外 `earnings-preview` 与 catalog 里已收录的 InvestSkill 同名那颗**撞车**——provider 的启用集合只按 `name` 认，
     `test/persona.test.mjs` 钉住 name 全局唯一，收进来必然出问题。
   - `test/selected-skills-provider.test.mjs` 用「skills 根目录 − 排除清单」派生本颗的可选清单，
-    `test/persona.test.mjs` 另把 `etoro` / `position-sizing` / `position-management` / `pre-trade-check` 明确断为不进 catalog。
+    `test/persona.test.mjs` 另把 `etoro` / `pre-trade-check` 明确断为不进 catalog（执行面与会话路由那一类）。
 - 复核方法：`git clone` 所钉 commit 后，把上游 `skills/<domain>/<name>` 与快照 `skills/<name>` 逐颗 `diff -r`——
-  差异只应包含以上三类（10 份文件、17 处句子级改动；其余 11 颗与 `_lib/`、`LICENSE` 逐字节相同）。
+  差异只应包含以上三类（13 份文件、20 处句子级改动；其余 9 颗与 `_lib/`、`LICENSE` 逐字节相同）。
   `test/selected-skills-content.test.mjs` 扫正文**结果**、本文件记**理由**，两边任一漂回去都会红。
 - Packaging: 这颗带 `.npmignore`，挡的是非运行内容：`docs/`（上游文档站与 18 份 walkthrough）、
   `scripts/`（上游的 catalog 生成与仓库校验 python）、`.github/`、`catalog.json`（上游自己的索引，与本仓
@@ -104,7 +118,7 @@
 - Runtime status: `watchlist` / 默认关闭的显式实验。只有使用者在「插件」→ Capital Generation →「精选 Skills」
   里逐格打开，它们才对**之后新建**的 Capital 根 Agent 可见；子 Agent 不继承，上游脚本从不执行。
   能力标签带来的新词是 `trade-discipline`（交易纪律 / Trade discipline），市场口径继续用 `us-stock`；
-  `riskNote` 里逐条写明各颗不做什么（不下单、不改单、不给仓位数字、不预测表现）。
+  `riskNote` 里逐条写明各颗不做什么（不预测表现；本插件没有券商接口，不据这些结论下单或改单）。
 - Review status: 2026-11-06 复查上游、许可证与维护状态。内容按美股与境外交易习惯写（SPY／QQQ／ES 期货、
   美股时段与盘前盘后、FINRA／Fidelity／Schwab 口径的集中度阈值、EURUSD 与 CFD／杠杆工具），
   **不声称适用 A 股**，也不得直接当作 A 股规则；它提供的是过程纪律而不是信号或收益承诺，

package/selected-skills/trading-skills/skills/catalyst-map/SKILL.md

@@ -104,7 +104,7 @@ Always include:
 - what requires deeper prep now versus background monitoring
 - whether the next step is `earnings-preview`, `macro-event-analysis`, `market-regime-analysis`, `thesis-validation`, or no further action yet
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices
 

package/selected-skills/trading-skills/skills/evidence-gap-check/SKILL.md

@@ -105,7 +105,7 @@ Always include:
 - the single highest-value next research step
 - whether the user should move next to `thesis-validation`, `earnings-preview`, `macro-event-analysis`, `market-regime-analysis`, `watchlist-review`, or stay in research mode
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices
 

package/selected-skills/trading-skills/skills/execution-plan-check/SKILL.md

@@ -106,7 +106,7 @@ Always include:
 - whether the plan should proceed, be resized, be restructured, or wait
 - whether the user should move next to `portfolio-concentration`, `position-sizing`, or back to `risk-reward-sanity-check`
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices
 

package/selected-skills/trading-skills/skills/journal-pattern-analyzer/SKILL.md

@@ -105,7 +105,7 @@ Always include:
 - one or two process changes to test next
 - whether the next step is `post-trade-review`, a revised rule in the journal, or no additional skill yet
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices
 

package/selected-skills/trading-skills/skills/portfolio-concentration/SKILL.md

@@ -104,7 +104,7 @@ Always include:
 - any special risk from employer stock, narrow products, or illiquid holdings
 - whether the user should move next to `position-sizing`, reduce exposure first, or review a fuller portfolio snapshot
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices
 

package/selected-skills/trading-skills/skills/position-management/SKILL.md

@@ -48,7 +48,7 @@ Use the user's materials first.
 
 If the original plan is missing, say that clearly and manage the position provisionally rather than inventing a clean prior plan.
 
-Do not fetch live data unless the user explicitly asks to pair this skill with another market-context or event-risk skill.
+Nothing here goes and gets market material on its own: work from what the user handed over. When the review genuinely needs a current quote or a fresh event, that is the main agent's call to delegate through the host's single retrieval entry (naming the ticker, the fields and the as-of date); whether a market-context or event-risk companion is worth pulling in is the same routing call, not this skill's.
 
 ## Analysis process
 

package/selected-skills/trading-skills/skills/position-sizing/SKILL.md

@@ -49,7 +49,7 @@ If any key input is missing, state what is missing and stop rather than invent i
 
 Use [references/methodology.md](references/methodology.md) for sizing conventions and caveats. Use [assets/trade-plan-template.md](assets/trade-plan-template.md) when the user wants a reusable planning format.
 
-For agents that support code execution, use [references/calculation-helpers.md](references/calculation-helpers.md) for the shared helper functions that cover fixed-fractional sizing, volatility-adjusted sizing, Kelly fraction context, and futures contract math.
+Use [references/calculation-helpers.md](references/calculation-helpers.md) for the quantities behind fixed-fractional sizing, volatility-adjusted sizing, Kelly fraction context, and futures contract math. This host has no shell: work each one from the base formula in [references/methodology.md](references/methodology.md) and show the arithmetic in the reply, handing volatility-type statistics to the data side (delegate `data_junior`).
 
 ## Output structure
 

package/selected-skills/trading-skills/skills/position-sizing/references/calculation-helpers.md

@@ -1,6 +1,8 @@
 # Calculation Helpers
 
-For code-execution environments, reuse the shared repo helper at `skills/_lib/calculations.py`.
+这里没有能跑代码的宿主：本插件的分析角色不带 shell，也不复用上游那份共享计算模块（它不进发布包）。
+下面每个名字对应一个可以手算的量——按 [methodology.md](methodology.md) 的基础算式在答复里把式子写出来算，
+需要波动率一类的统计量时交给宿主数据侧（本插件里是委派 `data_junior`），不在这份附属文件里自己算历史序列。
 
 Relevant functions for this skill:
 

package/selected-skills/trading-skills/skills/thesis-validation/SKILL.md

@@ -108,7 +108,7 @@ Always include:
 - the most important dependency or catalyst
 - whether the idea should move next to `market-regime-analysis`, `risk-reward-sanity-check`, `position-sizing`, or more research
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices
 

package/selected-skills/trading-skills/skills/watchlist-review/SKILL.md

@@ -104,7 +104,7 @@ Always include:
 - what additional prep each top name needs
 - whether the next step is `earnings-preview`, `macro-event-analysis`, `market-regime-analysis`, `thesis-validation`, or no further action yet
 
-A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and the ones this plugin deliberately does not offer — placing orders, setting position size — are not reachable at all. Say what you would check next and let the user decide.
+A skill named above is an option for the user, not a step this role can take: only the skills the user switched on exist in this session, and what this plugin deliberately does not offer — placing or amending an order on the user's behalf — is not reachable at all. Say what you would check next and let the user decide.
 
 ## Best practices