@topmindspace/topmind-skills 4.15.3 → 4.15.5

package/CHANGELOG.md

@@ -4,6 +4,23 @@ topmind-skills 的版本记录。GitHub 只保留最近 2 个 Release，本文
 
 版本真源：`topmind-pack.json`。`package.json`、每个 `SKILL.md` 的 `metadata.version`、`evals/evals.json` 与之一致。
 
+## 4.15.5 — 2026-10-16
+
+### 安装器
+
+- 默认升级：包里版本更新（或已装版本缺失）就直接覆盖，不备份，打印 `Upgraded <skill> <旧> -> <新>`；同版本跳过，打印 `Already up to date`（`--force` 重装）；已装版本比包里新时拒绝降级，需 `--force`。
+- symlink 模式：之前装过的同名 skill 普通目录（SKILL.md name 一致）直接换成软链；无关目录才需要 `--force`，并给一行提示。
+- 新增 `tests/installer-upgrade.test.mjs`；README 中英文、INSTALL.md、帮助文本同步。
+
+## 4.15.4 — 2026-10-16
+
+### 优化
+
+- 捕获说明不再写「待办一律 L1」。建议阶梯只留 `shared/auto-suggest.md`：材料 L0 落盘，显式行动语待办在 auto 下同轮写入，其余建议点头再落。
+- `topmind-capture` 不再复制一套抓取 L1/L2/L3 表。`shared/long-url-capture.md` 标明那些是抓取档，不是建议阶梯。
+- `skills.md` 的包内计数改回 10。公众号 `topmind-wechat-post` 仍在外部写作包，退役的 `topmind-wechat` 不进包。
+- Router 的错误处理改指向 `shared/writeback-receipt.md`，不再留第二份清单。
+
 ## 4.15.3 — 2026-10-10
 
 ### 优化

package/INSTALL.md

@@ -92,7 +92,7 @@ npx @topmindspace/topmind-skills add topmindspace/topmind-skills -p
 # 单目录（旧 --dest 仍可用）
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills --dest ~/.claude/skills
 
-# 升级（读 receipt）
+# 升级（读 receipt）：包里更新 → 直接覆盖，不备份，打印 Upgraded <skill> <旧> -> <新>；同版本 → Already up to date；已装更新 → 拒绝降级，需 --force
 npx @topmindspace/topmind-skills update -g
 
 # 查看
@@ -162,7 +162,7 @@ npx @topmindspace/topmind-skills doctor
 | `--skill <id>` | 只装列出的 skill id |
 | `--locale en-US` | locale overlay（当前未提供 overlay，回退中文正文） |
 | `-n` / `--dry-run` | 只打印计划 |
-| `--force` | symlink 模式下替换非软链目标 |
+| `--force` | 允许降级（已装版本比包里新）、同版本重装、symlink 模式下替换无关目录。之前装过的同名 skill 普通目录（SKILL.md name 一致）在 symlink 模式下会直接换成软链，不需要 --force |
 
 ### Agents
 

package/README.en.md

@@ -9,7 +9,7 @@ Portable AI skill pack: the same content contract on Claude Code / Codex / OpenC
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills -g
 # subset of agents
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills -g -a claude-code -a mimocode
-# upgrade
+# upgrade (default: a newer packaged skill overwrites in place, no backup, "Upgraded <skill> <old> -> <new>"; same version is skipped "Already up to date"; a newer installed version is refused unless --force)
 npx @topmindspace/topmind-skills update -g
 
 # 2) Community CLI (fast try; omits shared/ — see INSTALL.md)

package/README.md

@@ -9,7 +9,7 @@
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills -g
 # 只装给部分智能体
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills -g -a claude-code -a mimocode
-# 升级
+# 升级（默认：包里更新就直接覆盖，不备份，打印 Upgraded <skill> <旧> -> <新>；同版本跳过 Already up to date；已装更新拒绝降级，需 --force）
 npx @topmindspace/topmind-skills update -g
 
 # 2) 社区 CLI（快速试用；不含 shared/ — 见 INSTALL.md）

package/README.zh-CN.md

@@ -9,7 +9,7 @@
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills -g
 # 只装给部分智能体
 npx @topmindspace/topmind-skills add topmindspace/topmind-skills -g -a claude-code -a mimocode
-# 升级
+# 升级（默认：包里更新就直接覆盖，不备份，打印 Upgraded <skill> <旧> -> <新>；同版本跳过 Already up to date；已装更新拒绝降级，需 --force）
 npx @topmindspace/topmind-skills update -g
 
 # 2) 社区 CLI（快速试用；不含 shared/ — 见 INSTALL.md）

package/bin/install-skills.mjs

@@ -32,7 +32,7 @@
  *   --skill <id>        Only install these skill ids (repeatable / comma-separated)
  *   --global / -g       Alias: --host claude-code (user-level ~/.claude/skills)
  *   --dry-run           Print plan only
- *   --force             Replace non-symlink destinations when using symlink mode
+ *   --force             Downgrade / reinstall same version / replace unrelated folder in symlink mode
  *
  * Examples:
  *   # Like `npx skills add topmindspace/topmind-skills -g`
@@ -250,7 +250,12 @@ Options:
   --skill <id>         Only install these skill ids (repeatable / comma-separated)
   --locale <code>      Install locale overlay (e.g. en-US); falls back to topmind_LOCALE env
   -n, --dry-run        Print plan only
-  --force              Replace non-symlink destinations when using symlink mode
+  --force              Allow downgrading a newer installed skill, reinstall an up-to-date one,
+                       or replace an unrelated folder in symlink mode
+Upgrade behaviour: a newer packaged skill overwrites the installed copy in place
+  (no backup, "Upgraded <skill> <old> -> <new>"); the same version is skipped
+  ("Already up to date"); a newer installed version is refused unless --force.
+  In symlink mode a previous plain-folder install of the same skill is replaced.
 
 Examples:
   node bin/install-skills.mjs add topmindspace/topmind-skills -g
@@ -701,12 +706,75 @@ async function applyLocaleOverlay(overlayDir, targetDir) {
   }
 }
 
-async function installCopy(skillsRoot, dest, entries, locale) {
+/** Read name + version from <dir>/SKILL.md frontmatter (metadata.version or top-level version). */
+async function readSkillMeta(dir) {
+  try {
+    const t = await fs.readFile(path.join(dir, "SKILL.md"), "utf8");
+    const fm = (t.match(/^---\r?\n([\s\S]*?)\r?\n---/) || [])[1] || "";
+    const name = ((fm.match(/^name:\s*(.+?)\s*$/m) || [])[1] || "").replace(/^["']|["']$/g, "");
+    const v =
+      fm.match(/^metadata:\s*\n(?:[ \t]+.*\n)*?[ \t]+version:\s*(.+?)\s*$/m) || fm.match(/^version:\s*(.+?)\s*$/m);
+    return { name, version: v ? v[1].replace(/^["']|["']$/g, "") : "" };
+  } catch {
+    return null;
+  }
+}
+
+function parseVer(v) {
+  const m = String(v || "").trim().replace(/^v/, "").match(/^(\d+)\.(\d+)\.(\d+)/);
+  return m ? m.slice(1, 4).map(Number) : null;
+}
+
+/** a vs b: 1 / 0 / -1. Missing or unparseable b counts as oldest (upgrade). */
+function compareVersions(a, b) {
+  const pa = parseVer(a);
+  const pb = parseVer(b);
+  if (!pb) return 1;
+  if (!pa) return -1;
+  for (let i = 0; i < 3; i++) if (pa[i] !== pb[i]) return pa[i] > pb[i] ? 1 : -1;
+  return 0;
+}
+
+/** Existing non-symlink dst is a previous install of this skill (same SKILL.md name) or an installer-owned file. */
+async function isPreviousInstall(src, dst) {
+  const st = await fs.lstat(dst).catch(() => null);
+  if (!st) return false;
+  const sst = await fs.lstat(src).catch(() => null);
+  if (sst && sst.isFile() && st.isFile()) return true;
+  if (!st.isDirectory()) return false;
+  const a = await readSkillMeta(src);
+  const b = await readSkillMeta(dst);
+  return !!(a && b && a.name && a.name === b.name);
+}
+
+async function installCopy(skillsRoot, dest, entries, locale, { force = false } = {}) {
   await fs.mkdir(dest, { recursive: true });
+  // Pre-pass: never silently downgrade a skill that is newer than the packaged one.
+  const plan = new Map();
+  const newer = [];
+  for (const name of entries) {
+    const src = path.join(skillsRoot, name);
+    const dst = path.join(dest, name);
+    const sm = await readSkillMeta(src);
+    if (!sm) continue;
+    const dm = (await fs.lstat(dst).catch(() => null))?.isSymbolicLink() ? null : await readSkillMeta(dst);
+    if (!dm) continue;
+    const cmp = compareVersions(sm.version, dm.version);
+    plan.set(name, { cmp, old: dm.version || "unknown", now: sm.version || "unknown" });
+    if (cmp < 0) newer.push(`${name} ${dm.version} > ${sm.version}`);
+  }
+  if (newer.length && !force) {
+    fail(`installed version is newer than packaged (${newer.join(", ")}); not downgrading — pass --force to downgrade`);
+  }
   for (const name of entries) {
     const src = path.join(skillsRoot, name);
     const dst = path.join(dest, name);
     if (!(await pathExists(src))) continue;
+    const pv = plan.get(name);
+    if (pv && pv.cmp === 0 && !force) {
+      log(`  = Already up to date: ${name} ${pv.now}`);
+      continue;
+    }
     const st = await fs.lstat(src);
     if (await pathExists(dst)) await fs.rm(dst, { recursive: true, force: true });
     if (st.isDirectory()) {
@@ -727,7 +795,9 @@ async function installCopy(skillsRoot, dest, entries, locale) {
     } else {
       await fs.copyFile(src, dst);
     }
-    log(`  + ${name}`);
+    if (pv && pv.cmp > 0) log(`  ↑ Upgraded ${name} ${pv.old} -> ${pv.now}`);
+    else if (pv) log(`  + Reinstalled ${name} ${pv.now} (--force)`);
+    else log(`  + ${name}`);
   }
 }
 
@@ -739,8 +809,8 @@ async function installSymlink(skillsRoot, dest, entries, { force }) {
     if (!(await pathExists(src))) continue;
     if (await pathExists(dst)) {
       const st = await fs.lstat(dst);
-      if (st.isSymbolicLink() || force) await fs.rm(dst, { recursive: true, force: true });
-      else fail(`${dst} exists and is not a symlink (pass --force)`);
+      if (st.isSymbolicLink() || force || (await isPreviousInstall(src, dst))) await fs.rm(dst, { recursive: true, force: true });
+      else fail(`${dst} is an unrelated folder (no matching SKILL.md name) — pass --force to replace it`);
     }
     await fs.symlink(src, dst, "junction");
     log(`  ~ ${name} → ${src}`);
@@ -794,9 +864,9 @@ async function linkPackEntries(canonicalRoot, agentRoot, entryNames, { force })
     if ((await pathExists(dst)) || (await isBrokenSymlink(dst))) {
       const st = await fs.lstat(dst).catch(() => null);
       if (st && st.isSymbolicLink()) await fs.rm(dst, { recursive: true, force: true });
-      else if (force) await fs.rm(dst, { recursive: true, force: true });
+      else if (force || (await isPreviousInstall(src, dst))) await fs.rm(dst, { recursive: true, force: true });
       else {
-        log(`  ! ${dst} exists (not symlink) — skip (pass --force to replace)`);
+        log(`  ! ${dst} is an unrelated folder (no matching SKILL.md name) — skipped; pass --force to replace it`);
         continue;
       }
     }
@@ -845,7 +915,7 @@ async function installViaMatrix(skillsRoot, plan, opts) {
   }
 
   // 1) Full pack body → canonical
-  await installCopy(skillsRoot, canonical, plan.entries, opts.locale);
+  await installCopy(skillsRoot, canonical, plan.entries, opts.locale, opts);
 
   // 2) Private agent dirs: symlink entries (default) or copy
   for (const t of targets) {
@@ -854,7 +924,7 @@ async function installViaMatrix(skillsRoot, plan, opts) {
       continue;
     }
     if (linkMode === "copy") {
-      await installCopy(skillsRoot, t.root, plan.entries, opts.locale);
+      await installCopy(skillsRoot, t.root, plan.entries, opts.locale, opts);
     } else {
       await linkPackEntries(canonical, t.root, plan.entries, opts);
     }
@@ -1014,7 +1084,7 @@ async function cmdAdd(opts, sourceOverride) {
       }
       await installSymlink(skillsRoot, dest, plan.entries, opts);
     } else {
-      await installCopy(skillsRoot, dest, plan.entries, opts.locale);
+      await installCopy(skillsRoot, dest, plan.entries, opts.locale, opts);
     }
 
     await writeInstallReceipt(dest, {

package/evals/evals.json

@@ -1,7 +1,7 @@
 {
   "skill_name": "topmind",
   "daily_entry": "topmind",
-  "version": "4.15.3",
+  "version": "4.15.5",
   "evals": [
     {
       "id": 1,

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@topmindspace/topmind-skills",
-  "version": "4.15.3",
+  "version": "4.15.5",
   "description": "topmind Agent Skills pack — spec-valid Markdown skills for capture, organize, write, memory, maintain, loop.",
   "type": "module",
   "license": "MIT",

package/shared/long-url-capture.md

@@ -8,6 +8,8 @@
 
 ## 分层策略
 
+本节的 L1 / L2 / L3 / L3+ 是**抓取档**（怎么把网页变成 Markdown）。它们不是 [`auto-suggest.md`](./auto-suggest.md) 的建议阶梯 L0 / L1 / L2。
+
 ```text
 L1  静态 HTTP fetch + Readability / 启发式 / GitHub raw   ← 默认
 L2  Desktop 增强渲染（隐藏 Chromium offscreen） ← SPA 空壳 / 「增强渲染」

package/shared/trigger-disambiguation.md

@@ -36,7 +36,7 @@
 | X 长文 / 推特长文 | topmind-x-article（外部可选）vs write vs x | 写 X Articles 长文与一键复制 → `topmind-x-article`；未安装 → `topmind-write`。短帖不走 x-article |
 | 研究报告 / 调研报告 | organize vs topmind-research vs topmind-presentation | 要找证据、核事实 → `topmind-research`；对工作区已有材料做分析 → `topmind-organize`；要做成可翻页 HTML / PPTX → `topmind-presentation`。「做一份 X 的研究报告」先 research，回执建议 presentation，不自动链式；对应技能未安装 → organize / write |
 | 复盘 / 汇报材料 | loop vs organize vs topmind-presentation | 工作区巡检 → loop；本周动态 → organize「整理本周」；要出汇报幻灯片 → `topmind-presentation`（未安装 → `topmind-write`） |
-| 交接 / 导出记忆 / 记忆同步 | topmind-handoff（外部可选）vs memory vs write | 跨工具打包或接收交接包 → `topmind-handoff`；更新本工作区「我的情况」→ `topmind-memory`；导出稿件 → `topmind-write`。handoff 未安装 → 回执说明，不自动打包 |
+| 交接 / Handoff / 做个交接 / 工作交接 / 导出记忆 / 记忆同步 | topmind-handoff（外部可选）vs memory vs write | 只说做个handoff、做个交接、工作交接、Handoff、交接、记忆同步 → `topmind-handoff` 的一轮（先整合再写回）。点名导出或接收仍只做那一件。更新本工作区「我的情况」→ `topmind-memory`；导出稿件 → `topmind-write`。handoff 未安装 → 回执说明，不自动打包 |
 | 配图 / 封面 / 长图 / 海报 | topmind-cover vs topmind-poster vs topmind-presentation（均外部可选） | 文章头图、封面、单张配图 → `topmind-cover`；长图、信息海报、多段竖版 → `topmind-poster`；可翻页演示 → `topmind-presentation`；均未安装 → 回执说明，不硬生成 |
 
 **原则**：动作动词定 action；修饰词定 confidence。无法判定 → capture（先存后整）。

package/skills.md

@@ -1,7 +1,7 @@
 # topmind Skills Package Index (包索引)
 
 > [!NOTE]
-> **每日唯一入口**：`topmind` (router)。其余 6 个 action、2 个 optional connector、1 个 optional 记账模块、1 个 optional 公众号 write 子技能（共 11）均由 router 调度。  
+> **每日唯一入口**：`topmind` (router)。包内还有 6 个 action、2 个 optional connector、1 个 optional 记账（连 router 共 10 个目录）。公众号 `topmind-wechat-post` 在外部写作包，不计入本包。  
 > **通用与可移植**：兼容 Claude Code / OpenCode / Codex / Hermes 等 Agent Host。  
 > **版本真源**：`topmind-pack.json`（可执行 `npm run versions` 查看完整版本号）。  
 > 详见权威说明：[Skills README](./README.md)（简体中文） · [English](./README.en.md) · [安装与部署指南 (INSTALL.md)](./INSTALL.md)。
@@ -32,7 +32,7 @@
 2. **渐进式披露 (Progressive Disclosure)**：Discovery (`name+description`) → Activation (`SKILL.md`) → Deep Procedures (`shared/` / `references/`)。
 3. **文件系统即真源**：不引入独立的数据库或 Agent state 依赖。
 4. **遵从 6 条核心规约**：见 [`shared/project-model-brief.md`](./shared/project-model-brief.md)。
-5. **自动建议阶梯**：L0 零提问落盘 · L1 默认准备建议（分类/待办/记忆/专题）· L2 确认后写。见 [`shared/auto-suggest.md`](./shared/auto-suggest.md)。待办是 `memory/todo.md` 卫星，不是第六概念。
+5. **自动建议阶梯**：真源 [`shared/auto-suggest.md`](./shared/auto-suggest.md)（材料 L0 落盘；显式行动语待办在 auto 下同轮写入；其余建议点头再落；L2 才动记忆/删档/建专题）。待办是 `memory/todo.md` 卫星，不是第六概念。
 
 👉 **详细架构、Frontmatter Schema 与安装方法请参阅**：
 - [Skills README](./README.md)（简体中文） · [English](./README.en.md)

package/topmind/SKILL.md

@@ -9,14 +9,14 @@ compatibility: >-
   topmind workspace with {NN-Name}/ categories. Host file tools primary; UTR optional.
   Install with shared/ sibling for progressive disclosure links.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "router"
   entrypoint: "true"
   triggers: "topmind, 知识库, 工作区, 待办, 有什么要做的, 要做的事, todo, todo list"
   tags: "router, entrypoint, topmind"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 
@@ -100,7 +100,7 @@ Which category? Which topic (or loose note)? Which object? Which action? Which s
 - 高信心类别、无专题 → `{大类}/*.md`，建议是否升级专题  
 - 低信心类别 → **role:buffer**（常为 `00-Inbox/`）+ L1 分类备选（最多 3）  
 - 跳过 `hidden` 类别  
-- 行动语 / 稳定个人信息 → L1 待办提取、记忆候选（不打断落盘）
+- 行动语 / 稳定个人信息 → 不打断落盘；写入或建议只按 [`../shared/auto-suggest.md`](../shared/auto-suggest.md)
 
 ### Minimum Context
 
@@ -149,11 +149,7 @@ UTR 可选（MCP primary+danger 共 29；注册表 38 = 8 域 / 38 命令，见
 
 ## Error Handling
 
-写入失败时：
-- 磁盘满 / 权限不足 → 报错 + 已备回复制位置；不静默丢数据
-- 路径过长 / 非法字符 → 提示用户缩短专题名
-- protection:locked → 内容编辑允许（任务级首写快照）；可恢复删/归档允许；永久删 locked 仅用户
-- 工作区不存在 topmind.yaml → 按默认契约解释 + 回执标注「默认契约」
+写入失败不得静默丢数据。优先级与 locked 语义见 [`../shared/writeback-receipt.md`](../shared/writeback-receipt.md)。工作区没有 `topmind.yaml` 时按默认契约解释，回执标注「默认契约」。
 
 
 

package/topmind-capture/SKILL.md

@@ -8,21 +8,21 @@ description: >-
 license: MIT
 compatibility: topmind workspace. Host file tools or optional UTR capture-note.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "capture"
   entrypoint: "false"
   triggers: "收集, 收进, 保存, 剪藏, 记一下, 临时存放, 随手记, 想法, 灵感, 暂存沉淀, 记录, 备忘, 存一下, 暂存, 笔记, 抄下来, 摘录, 记一下要做, 别忘了, 要做, capture, note, save, Note it"
   tags: "capture, inbox, note, material"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 
 # topmind Capture
 
 先存材料（用户动词：**记一下**）。不负责重塑、交付、诊断或外部连接。  
-**想记就记 / 随用随记**：默认零提问落盘；分类、待办、记忆一律走 [`../shared/auto-suggest.md`](../shared/auto-suggest.md) 的 L1 建议，不弹多选。
+**想记就记 / 随用随记**：材料默认零提问落盘。分类、待办、记忆的写入或建议只按 [`../shared/auto-suggest.md`](../shared/auto-suggest.md)，本文件不另写一套阶梯，也不弹多选。
 
 ## Activation checklist
 
@@ -31,7 +31,7 @@ metadata:
 3. 按内容性质选 **真实目录名**（role；禁止写死编号）  
 4. **日常/动态** → 追加到当前周期本；专题明确 → 专题根；不清 → Inbox  
 5. 写 frontmatter；白话回执（如「已记到每周一本」）；**不**自动 organize/write  
-6. **同轮 L1**：扫描行动语 / 稳定个人信息 / 低信心分类 → 回执附最多 3 条建议（见下）  
+6. **同轮建议阶梯**：行动语 / 稳定个人信息 / 低信心分类只按 [`../shared/auto-suggest.md`](../shared/auto-suggest.md)（不在这里复述）  
 
 ## When NOT to use
 
@@ -71,18 +71,11 @@ low cat                   → role:buffer（Inbox）
 
 **低信心时不要只丢 Inbox 就完事**：仍落 buffer，回执附分类备选（最多 3 个，真实目录名），如「更像 ① 20-研究 ② 10-动态，说一声即可移」。详见 [`../shared/auto-suggest.md`](../shared/auto-suggest.md)。
 
-### 待办与记忆（L1，不打断捕获）
+### 待办与记忆（不打断捕获）
 
-捕获正文只写材料；下列副作用走建议阶梯，**不**为分类/待办向用户提问：
+捕获正文只写材料。待办、记忆、分类备选的 L0/L1/L2 只以 [`../shared/auto-suggest.md`](../shared/auto-suggest.md) 为准，**不**为分类/待办向用户提问。
 
-| 信号 | L0 | L1 |
-|------|----|----|
-| 显式行动语（要做 / 记得 / 别忘 / 待办 / 截止 / todo） | 材料照常落盘 | 同轮提取到 `memory/todo.md`（auto 可直接写+回执；confirm 则建议） |
-| 隐含行动语（「周三截止」「下周要交」） | 材料照常落盘 | 建议提取 1–N 条待办，点头再写 |
-| 稳定个人信息（偏好 / 目标 / 进行中） | 材料照常落盘 | 建议进「我的情况」（写入仍须确认） |
-| 低信心类别 | Inbox | 分类备选 |
-
-清单是语义平面卫星（`memory/todo.md`），**不是**新概念、**不**新建顶层 `todo/`。去重按文本；拒绝过的条目不要反复再提。
+清单是语义平面卫星（`memory/todo.md`），**不是**新概念、**不**新建顶层 `todo/`。去重按文本；拒绝过的条目不要反复再提。**禁止**静默写 `memory/profile.md` 或 `memory/topics/`。
 
 UTR：`capture-note`（自动识别周期本）。Desktop：`ingestInbox` dest.mode=`stream`（默认）。
 
@@ -101,17 +94,9 @@ route_reason: …
 status: todo
 ```
 
-长 URL / 网页：[`../shared/long-url-capture.md`](../shared/long-url-capture.md)。
-
-| 层 | 谁 | 能力 |
-|----|----|------|
-| L1 | Desktop `fetchUrl` / host fetch | Mozilla Readability + 启发式 |
-| L1-GH | 同上 · GitHub md/README | raw 直取 + 相对图改写（`lib/github-md.mjs`） |
-| L2 | Desktop 增强渲染 | SPA 空壳（ephemeral Chromium） |
-| L3 | 手动粘贴 | 永远可用 |
-| L3+ | 浏览器扩展 Clip Bridge | 活 DOM Readability → Desktop 清洗落盘 |
+长 URL / 网页：[`../shared/long-url-capture.md`](../shared/long-url-capture.md)。那里的 L1/L2/L3 是**抓取档**，不是建议阶梯。
 
-Agent host 能 fetch 则提取正文，否则保留 URL + 摘录；推荐日常剪藏走 L3+。GitHub 仓库 / Markdown 文件优先走 raw/README，不 HTML scrape。
+Agent host 能 fetch 则提取正文，否则保留 URL + 摘录。GitHub 仓库 / Markdown 文件优先走 raw/README，不 HTML scrape。
 
 ### 本地文档 / Office / 邮件 / 统一捕获
 
@@ -132,7 +117,7 @@ Agent host 能 fetch 则提取正文，否则保留 URL + 摘录；推荐日常
 3. 保留 provenance；长文标注截断  
 4. 按保存设置写入（见 [`../shared/writeback-receipt.md`](../shared/writeback-receipt.md)）  
 5. 回执：路径、目标、信心、下一步（不自动触发其他 skill）  
-6. L1 建议：待办提取 / 记忆候选 / 分类备选（见 [`../shared/auto-suggest.md`](../shared/auto-suggest.md)）
+6. 待办 / 记忆 / 分类备选：只按 [`../shared/auto-suggest.md`](../shared/auto-suggest.md)，不在回执外再写一套阶梯
 
 **边界**：只写材料笔记。**禁止**改 `topic.md`、禁止自动 organize、禁止静默写 memory/profile 或 memory/topics。存完可在回执**建议**「需要时整理 / 记入待办 / 更新我的情况」——仅用户面，不自动链式。
 

package/topmind-ledger/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: Host file tools or Kernel writeback. Desktop optional mini-app. Feishu/lark-cli is not required.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "memory"
   entrypoint: "false"
   triggers: "记账, 记一笔, 花了, 存入, 查看账单, 账户余额, bookkeeping, log expense"
   tags: "ledger, bookkeeping, memory, optional"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-loop/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: topmind workspace. Progress under .topmind/loop/*. Host LLM runs the walk; UTR optional.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "loop"
   entrypoint: "false"
   triggers: "跑一遍 loop, loop 一下, 整体体检, 工作区复盘, 把工作区跑一遍, 巡检, 继续 loop, 从断点继续, 全面检查, 复盘, review, audit, walk"
   tags: "loop, audit, walk, resumable, cyclic"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-maintain/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: topmind engine and/or workspace. Prefer read-only diagnostics first.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "maintain"
   entrypoint: "false"
   triggers: "快速体检, 体检, 诊断, doctor, 健康检查, 清理, 修复, 快速检查, 检查, 状态, workspace check, scan, fix, maintain"
   tags: "maintain, doctor, cleanup, audit, health-check"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-memory/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: workspace core profile (memory/profile.md) + periodic reflections.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "memory"
   entrypoint: "false"
   triggers: "写入专题记忆, 更新我的情况, stable memory, 追加确认结论, 记住这个, 记住我, 加到专题记忆, 沉淀结论, 提炼到记忆, update profile, memory"
   tags: "memory, stable, core-profile, 我的情况"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-organize/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: topmind workspace with stream period notes, topics, or Inbox.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "organize"
   entrypoint: "false"
   triggers: "整理本周, 理顺本周, 周复盘, 整理, 组织, 研究, 分析, 对比, 提取, 梳理, 归纳, 证据, 实体, 质量审查, 总结, 概括, 汇总, 提纲, 分类整理, 找关联, 总结成笔记要点, organize, summarize, analyze, weekly review"
   tags: "organize, research, analyze, stream, weekly-review"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-pack.json

@@ -4,8 +4,8 @@
   "label": "topmind",
   "description": "topmind skill pack: single daily entry topmind + progressive disclosure (shared/ + references/), Agent Skills spec-valid frontmatter (custom fields under metadata, checked by agentskills validate), capture-first template-driven categories, optional connectors topmind-weread and topmind-x, optional topmind-ledger (记账). None of the optional skills is an extra user concept. WeChat (公众号) and other writing skills live in the external topmind-writing-skills pack (topmind-wechat-post etc.) and are routed only when installed. File-system-first content truth; UTR optional.",
   "daily_entry": "topmind",
-  "version": "4.15.3",
-  "updated": "2026-10-08",
+  "version": "4.15.5",
+  "updated": "2026-10-16",
   "license": "MIT",
   "locales": {
     "default": "zh-CN",

package/topmind-weread/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: Requires WeRead Skill API key (wrk-*). Desktop WereadService preferred; host may call gateway manually.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "connector"
   entrypoint: "false"
   triggers: "微信读书, weread, 划线同步, 读书笔记, 阅读统计, 书架, weread sync, sync highlights"
   tags: "weread, reading, highlights, sync, capture"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-write/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: topmind workspace. Delivery role category (often 88-交付 / 88-Delivery).
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "write"
   entrypoint: "false"
   triggers: "写, 写作, 起草, 续写, 修订, 润色, 交付, 改写, de-AI, 输出, 出稿, 定稿, 发布, 导出, 排版, write, draft, deliver, publish, export"
   tags: "write, draft, revise, polish, deliver"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---
 

package/topmind-x/SKILL.md

@@ -6,14 +6,14 @@ description: >-
 license: MIT
 compatibility: Desktop needs Bearer (read) and/or xurl (post). Agent hosts use official X MCP + xurl bridge.
 metadata:
-  version: "4.15.3"
+  version: "4.15.5"
   action_category: "connector"
   entrypoint: "false"
   triggers: "发推, 推特, twitter, x.com, 发帖, 推文, post tweet, x platform"
   tags: "x, twitter, social, mcp, agent"
   author: "TopMindSpace"
   homepage: "https://github.com/topmindspace/topmind-skills"
-  updated: "2026-10-08"
+  updated: "2026-10-16"
   degradation: "../shared/capability-degradation.md"
 ---