feat: 网站 CJK 友好强调、构建时更新共享查询 Skill、飞书样式导入修复（0.7.86）

CHANGELOG.md

@@ -2,6 +2,12 @@
 
 All notable changes to Context are documented here.
 
+## 0.7.86 - 2026-10-26
+
+- Website emphasis now follows the CJK-friendly amendment to CommonMark through `markdown-it-cjk-friendly`, replacing 0.7.85's built-in rule. Any East Asian wide or full-width character beside a `*` run, including full-width punctuation, counts like whitespace when deciding whether the run opens or closes emphasis; `_` and runs with no CJK character beside them keep the CommonMark result. Clean content renders as in 0.7.85.
+- `context build` keeps the shared `context-knowledge-query` Skill current. When a kb template's copy differs from the shipped Skill in the same language, the build replaces it and reports a `package-query-skill-refreshed` agent hint, so a package no longer declares one Context version while carrying an older version's Skill. Package-specific search hints stay in `others/context/search-guide.md`; a package that needs its own query Skill adds it under a different name.
+- Lark capture merges adjacent runs of the same inline style and keeps whitespace outside emphasis markers, so `<b>A</b><b> </b><b>B</b>` becomes `**A B**` instead of `**A**** ****B**`. Paragraphs inside table cells and list items stay on separate lines (`<br>` in tables) instead of running together. Existing snapshots change only when they are captured again.
+
 ## 0.7.85 - 2026-10-25
 
 - Website bold and italic text now closes after CJK punctuation. CommonMark only ends `**…**` after punctuation such as `？`, `）` or `、` when whitespace or more punctuation follows, so Chinese like `**用 MCP 还是知识包？**不在…` showed the asterisks literally. The site renderer now treats a Chinese, Japanese or Korean letter on the outer side of the markers like whitespace. Delimiters with no CJK letter beside them render exactly as before, and sites built with an earlier CLI re-render on the next build.

package.json

@@ -1,6 +1,6 @@
 {
   "name": "context",
-  "version": "0.7.85",
+  "version": "0.7.86",
   "packageManager": "bun@1.3.9",
   "repository": {
     "type": "git",

packages/context-cli/context-workflow/provider.yaml

@@ -1,6 +1,6 @@
 schema: agent-graph.provider.v1
 id: c4a/context
-version: 0.7.85
+version: 0.7.86
 name: Context workflow
 description: Internal work contract for Context knowledge workspaces.
 graphs:

packages/context-cli/context-workflow/resources/manuals/guides/package-outputs.md

@@ -278,7 +278,9 @@ records establish what the package actually says. Package authors edit the sourc
 template when project-specific terminology, entry points or task workflows are
 needed: put terms, synonyms, common questions and preferred entry pages in
 `others/context/search-guide.md`, which the Skill reads first, and keep the
-shared Skill unchanged. Editing the Skill makes it package-specific; rename its
+shared Skill unchanged. Context manages the shared Skill: `context build`
+replaces a template copy that differs from the shipped one, so edits to it do
+not survive. A package that needs its own query Skill adds it under a different
 directory and `name` so it installs separately. When a page's text does not
 settle a question, the Skill follows the page's `Sources` entries: repository
 links are read at their recorded commit through authorized host tools, document
@@ -290,7 +292,11 @@ Workspaces created before the rename keep `skills/knowledge-query/` in their
 template. `context build` replaces it with `context-knowledge-query` in the old
 Skill's language, adds `others/context/search-guide.md` when it is missing, and
 reports the replacement as an agent hint. Move any package-specific wording the
-old Skill carried into the search guide.
+old Skill carried into the search guide. Every build also compares
+`skills/context-knowledge-query/` with the shipped Skill in the same language
+and replaces an older or edited copy, reporting a `package-query-skill-refreshed`
+agent hint, so a package never declares one Context version while carrying
+another version's Skill.
 
 When approved pages reference materialized resources, Context keeps their
 production copies in content-addressed `knowledge/assets/` paths and bundles

packages/context-cli/context-workflow/resources/manuals/reference/package-templates.md

@@ -247,7 +247,8 @@ The default kb template includes:
   attribution, follows repository sources at their recorded commit when the page
   text is not enough, and reports gaps instead of inventing unsupported answers.
   It uses no package-specific template values, so every package renders the same
-  file. The
+  file, and `context build` replaces a template copy that differs from the
+  shipped Skill in its language. The
   default entry OKF root is `wikis/`; packages that select additional internal
   collections expose
   `guides/`, `rules/`, or `feats/` indexes when those roots are selected.

packages/context-cli/package.json

@@ -1,7 +1,7 @@
 {
   "name": "@c4a/context-cli",
   "description": "Local runtime and Agent integration for traceable knowledge production",
-  "version": "0.7.85",
+  "version": "0.7.86",
   "type": "module",
   "license": "MIT",
   "engines": {
@@ -71,6 +71,7 @@
     "handlebars": "^4.7.8",
     "ink": "^5.0.0",
     "jiti": "^2.7.0",
+    "markdown-it-cjk-friendly": "3.0.0",
     "mdast-util-to-string": "^4.0.0",
     "mermaid": "11.12.0",
     "react": "^18.3.1",

packages/context-cli/src/__tests__/documentLarkXmlV062.test.ts

@@ -46,6 +46,25 @@ describe("Lark Docx XML readable projection", () => {
     expectClosedCounts(projection.fidelity);
   });
 
+  test("merges split style runs and keeps paragraphs inside table cells apart", () => {
+    const projection = projectLarkDocxXml({
+      sourceUrl: "https://example.larkoffice.com/wiki/source-token",
+      xml: [
+        "<title>Quarterly review</title>",
+        "<p><b>整体总结：oncall</b><b> </b><b>问题数维持在低位</b>，完成了<b>API</b><b> </b><b>封装</b>。</p>",
+        "<p>Plain <i>split</i><i> italic</i> and<b> padded </b>text.</p>",
+        "<table><tr><td>Area</td><td>Result</td></tr>",
+        "<tr><td>质量</td><td><p><b>评分</b>: 0.7</p><p><b>总结</b>:</p><ul><li>details</li></ul></td></tr></table>",
+      ].join("\n"),
+    });
+
+    expect(projection.markdown).toContain("**整体总结：oncall 问题数维持在低位**，完成了**API 封装**。");
+    expect(projection.markdown).toContain("Plain *split italic* and **padded** text.");
+    expect(projection.markdown).toContain("| 质量 | **评分**: 0.7<br>**总结**:<br>");
+    expect(projection.markdown).not.toContain("****");
+    expectClosedCounts(projection.fidelity);
+  });
+
   test("distinguishes inline and token-backed diagrams without guessing from prose", () => {
     const projection = projectLarkDocxXml({
       sourceUrl: "https://example.larkoffice.com/docx/source-token",