@gobing-ai/knowledge-kit 0.0.26 → 0.0.28

package/.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
     },
     "plugins": [
         {
-            "version": "0.0.26",
+            "version": "0.0.28",
             "name": "kk",
             "source": "./plugins/kk",
             "description": "knowledge-kit — a set of toolkit for knowledge management"

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@gobing-ai/knowledge-kit",
-    "version": "0.0.26",
+    "version": "0.0.28",
     "description": "An ingest → create → publish content pipeline CLI (Bun).",
     "repository": {
         "type": "git",

package/plugins/generations/dailynews-gen/dist/index.js

@@ -22064,19 +22064,22 @@ function stripLeadingTitleEcho(body, title) {
   const wanted = title.replace(TITLE_ECHO_NOISE_RE, "");
   if (wanted.length < 4)
     return body;
-  let matched = "";
-  let end = -1;
-  for (let i = 0;i < body.length && matched.length < wanted.length; i++) {
-    const kept = (body[i] ?? "").replace(TITLE_ECHO_NOISE_RE, "");
-    if (kept === "")
-      continue;
-    matched += kept;
-    if (matched.length >= wanted.length)
-      end = i + 1;
+  let text = body;
+  for (;; ) {
+    let matched = "";
+    let end = -1;
+    for (let i = 0;i < text.length && matched.length < wanted.length; i++) {
+      const kept = (text[i] ?? "").replace(TITLE_ECHO_NOISE_RE, "");
+      if (kept === "")
+        continue;
+      matched += kept;
+      if (matched.length >= wanted.length)
+        end = i + 1;
+    }
+    if (end === -1 || matched !== wanted)
+      return text;
+    text = text.slice(end).replace(/^[\uFF0C\u3002\u3001\uFF1A:\uFF1B;!\uFF01?\uFF1F\u2026\s\u300C\u300D\u300E\u300F\u3010\u3011]+/, "");
   }
-  if (end === -1 || matched !== wanted)
-    return body;
-  return body.slice(end).replace(/^[\uFF0C\u3002\u3001\uFF1A:\uFF1B;!\uFF01?\uFF1F\u2026\s\u300C\u300D\u300E\u300F\u3010\u3011]+/, "");
 }
 var TAG_METADATA_LINE_RE = /^\s*\**\s*\u6807\u7B7E\s*\**\s*[\uFF1A:\uFF0C,]/;
 function stripTagMetadataLines(body) {

package/plugins/generations/dailynews-gen/src/script-builder.ts

@@ -232,20 +232,29 @@ const TITLE_ECHO_NOISE_RE = /[，。、：:；;!！?？…\s「」『』【】]/
  * character prefix compare, never a fuzzy/LLM similarity match (anti-pattern list). The body is
  * returned untouched whenever the normalized prefix does not equal the normalized title, and a
  * body that is *only* the restatement collapses to the empty string (the lead alone carries it).
+ *
+ * 20260929 dogfood: plan bodies can restate the title MORE THAN once — a bare first line plus a
+ * restated line carrying a parenthetical extension (`title（来源标注日期为…）`). A single pass
+ * strips only the first, so the second becomes the new head and the VoiceScript validator rejects
+ * the segment. Keep stripping until the head no longer echoes; every accepted pass strictly
+ * shortens the body, so the loop terminates.
  */
 export function stripLeadingTitleEcho(body: string, title: string): string {
     const wanted = title.replace(TITLE_ECHO_NOISE_RE, '');
     if (wanted.length < 4) return body; // a short title would match too much of the body
-    let matched = '';
-    let end = -1;
-    for (let i = 0; i < body.length && matched.length < wanted.length; i++) {
-        const kept = (body[i] ?? '').replace(TITLE_ECHO_NOISE_RE, '');
-        if (kept === '') continue; // drift punctuation between title and echo
-        matched += kept;
-        if (matched.length >= wanted.length) end = i + 1;
+    let text = body;
+    for (;;) {
+        let matched = '';
+        let end = -1;
+        for (let i = 0; i < text.length && matched.length < wanted.length; i++) {
+            const kept = (text[i] ?? '').replace(TITLE_ECHO_NOISE_RE, '');
+            if (kept === '') continue; // drift punctuation between title and echo
+            matched += kept;
+            if (matched.length >= wanted.length) end = i + 1;
+        }
+        if (end === -1 || matched !== wanted) return text;
+        text = text.slice(end).replace(/^[，。、：:；;!！?？…\s「」『』【】]+/, '');
     }
-    if (end === -1 || matched !== wanted) return body;
-    return body.slice(end).replace(/^[，。、：:；;!！?？…\s「」『』【】]+/, '');
 }
 
 /**

package/plugins/generations/episode-plan-gen/dist/index.js

@@ -22101,7 +22101,14 @@ function docTimestamp(doc2) {
   }
   return null;
 }
+var CHINA_POLITICS_RE = /\u4E60\u8FD1\u5E73|\u674E\u514B\u5F3A|\u674E\u5F3A|\u515A\u4E2D\u592E|\u653F\u6CBB\u5C40|\u4E2D\u5357\u6D77|\u653F\u5E9C\u5DE5\u4F5C\u62A5\u544A|\u56FD\u53F0\u529E|\u5916\u4EA4\u90E8\u53D1\u8A00\u4EBA|\u4E24\u4F1A/;
+function isChinaPolitics(doc2) {
+  return CHINA_POLITICS_RE.test(`${doc2.title ?? ""}
+${(doc2.body ?? "").slice(0, 2000)}`);
+}
 function passesGates(doc2, planDate) {
+  if (isChinaPolitics(doc2))
+    return false;
   const meta3 = asRecord(doc2.metadata);
   const tags = Array.isArray(meta3.tags) ? meta3.tags.filter((t) => typeof t === "string") : [];
   if (tags.some((tag) => NON_NEWS_TAGS.has(tag.toLowerCase())))

package/plugins/generations/episode-plan-gen/src/index.ts

@@ -227,7 +227,19 @@ function docTimestamp(doc: Doc): number | null {
 }
 
 /** Freshness + non-news gates applied before scoring/dedup (archived or off-topic items never enter the pool). */
+/**
+ * Platform-safety gate (operator rule 2026-09-29): China domestic politics stories never enter
+ * the plan. Keyword heuristic — high-precision markers + top-leader names only; widen after
+ * reviewing false negatives in the audit trail rather than pre-emptively (overblocking kills
+ * legitimate AI-industry coverage).
+ */
+const CHINA_POLITICS_RE = /习近平|李克强|李强|党中央|政治局|中南海|政府工作报告|国台办|外交部发言人|两会/;
+function isChinaPolitics(doc: Doc): boolean {
+    return CHINA_POLITICS_RE.test(`${doc.title ?? ''}\n${(doc.body ?? '').slice(0, 2000)}`);
+}
+
 function passesGates(doc: Doc, planDate: string): boolean {
+    if (isChinaPolitics(doc)) return false;
     const meta = asRecord(doc.metadata);
     const tags = Array.isArray(meta.tags) ? meta.tags.filter((t): t is string => typeof t === 'string') : [];
     if (tags.some((tag) => NON_NEWS_TAGS.has(tag.toLowerCase()))) return false;

package/plugins/generations/image-gen/dist/index.js

@@ -23435,7 +23435,12 @@ header = ${curlQuote(`x-goog-api-key: ${apiKey}`)}
           "-d",
           "@-",
           url2
-        ], { input: JSON.stringify(body), encoding: "utf8", timeout: 310000 });
+        ], {
+          input: JSON.stringify(body),
+          encoding: "utf8",
+          timeout: 310000,
+          stdio: ["pipe", "pipe", "pipe"]
+        });
       } catch (error51) {
         const e = error51;
         const details = (typeof e.stderr === "string" ? e.stderr.trim() : "") || e.message || "curl request failed";

package/plugins/generations/image-gen/src/providers/google.ts

@@ -174,6 +174,9 @@ export const createGoogle: ProviderFactory = (options?: ProviderOptions): ImageP
             // the key and the proxy password into logs. `-sS` (not `-s`) keeps curl's own diagnostic on
             // stderr, so `details` no longer falls back to that argv-bearing message.
             //
+            // `stdio: ['pipe', 'pipe', 'pipe']` is explicit because Bun's `execFileSync` otherwise
+            // tees the child's stderr to the parent's stderr even though it also captures it.
+            //
             // File IO goes through the `@gobing-ai/ts-runtime` seam (`no-direct-fs-io`): the seam is
             // text-only with no `mode` argument, and the file lives in `os.tmpdir()`, which is a
             // per-user directory (0700 on macOS).
@@ -199,7 +202,12 @@ export const createGoogle: ProviderFactory = (options?: ProviderOptions): ImageP
                         '@-',
                         url,
                     ],
-                    { input: JSON.stringify(body), encoding: 'utf8', timeout: 310_000 },
+                    {
+                        input: JSON.stringify(body),
+                        encoding: 'utf8',
+                        timeout: 310_000,
+                        stdio: ['pipe', 'pipe', 'pipe'],
+                    },
                 );
             } catch (error: unknown) {
                 const e = error as { message?: string; stderr?: string | Buffer };

package/plugins/kk/plugin.json

@@ -1,5 +1,5 @@
 {
     "name": "kk",
-    "version": "0.0.26",
+    "version": "0.0.28",
     "description": "knowledge-kit Claude Code plugin: skills, commands, subagents, hooks, and rules"
 }

package/plugins/kk/scripts/kk-workflow-stages.ts

@@ -1233,7 +1233,9 @@ function dailyPublishXhs(args: string[]): void {
  * outcome never enters `publish_status` (surfdash + podcast remain the publication contract).
  */
 function dailyPublishWechat(args: string[]): void {
-    const [workDir = '', runDate = '', pluginsPath = '', live = 'false'] = args;
+    const [workDirRaw = '', runDate = '', pluginsPath = '', live = 'false', collectionArg = ''] = args;
+    // Metadata carries file paths consumed by the plugin child process (cwd = plugin dir) - never emit relatives.
+    const workDir = resolve(workDirRaw);
     const article = join(workDir, '22-article', `${runDate}_06_article_content.json`);
     const input = join(workDir, '31-publish', `${runDate}_20_publish-wechat_content.json`);
     const result = join(workDir, '31-publish', `${runDate}_20_publish-wechat_result.json`);
@@ -1250,8 +1252,35 @@ function dailyPublishWechat(args: string[]): void {
             return;
         }
         const packaged = requireJson(article, 'publish-wechat');
+        // MP draft enrichment (operator preference 2026-09-29): 原创+作者, 赞赏, 合集, 原文链接
+        // (the surfdash post), AI 生成声明, cover. The brand strings are this daily show's constants.
+        const cover = join(workDir, '23-cover', `${runDate}_07_cover_cover.png`);
+        const audio = join(workDir, '30-audio', `${runDate}_11_generate_content.mp3`);
+        const wechatmp: Record<string, unknown> = {
+            original: true,
+            aiGenerated: true,
+            author: '冰原奔狼',
+            collection: collectionArg || '每日AI资讯',
+            ...(isFile(cover) ? { coverPath: cover } : {}),
+            // Audio block at the article top (operator request 2026-09-29): the daily MP3 goes to
+            // the 素材库 under the article title; the plugin reuses an existing same-title material.
+            ...(isFile(audio)
+                ? {
+                      audioPath: audio,
+                      audioTitle: typeof packaged.title === 'string' ? packaged.title : '',
+                      audioCoverPath: isFile(cover) ? cover : '',
+                  }
+                : {}),
+        };
+        // WeChat-only closing line (operator request 2026-09-29): point readers/listeners back to
+        // the show's platforms. Appended to the wechat input body only, never the master article.
+        const outro = '\n\n欲听往期节目，可至小宇宙、Apple Podcast、YouTube 等平台搜索【咸话咸说】。';
         sted('write', input, () =>
-            writeJsonText(input, { ...packaged, metadata: { ...metadataOf(packaged), publishStatus: want } }),
+            writeJsonText(input, {
+                ...packaged,
+                body: `${typeof packaged.body === 'string' ? packaged.body : ''}${outro}`,
+                metadata: { ...metadataOf(packaged), publishStatus: want, wechatmp },
+            }),
         );
         // `soft` keeps the WeChat target non-blocking: the plugin still writes its own failed Result
         // and prints the failing step (session preflight, editor paste, save control), which is what
@@ -1575,6 +1604,17 @@ function dailyPublishPrep(args: string[]): void {
         stampId3Tags(audioFile, typeof article.title === 'string' ? article.title : '', runDate, artist, album);
         const resolvedCover = resolve(coverPath);
         const hasCover = existsSync(resolvedCover);
+        // Daily-show itunes convention: season = year, episode = day-of-year. run_date arrives
+        // dashed (`$(date +%F)`) or digits-only (formatRunDate); map.ts forwards numbers only,
+        // so a malformed run_date omits both tags instead of failing the publish.
+        const digits = runDate.replaceAll(/\D/g, '');
+        const season = Number(digits.slice(0, 4));
+        const month = Number(digits.slice(4, 6));
+        const day = Number(digits.slice(6, 8));
+        const episode =
+            digits.length === 8 && month >= 1 && month <= 12 && day >= 1 && day <= 31
+                ? Math.round((Date.UTC(season, month - 1, day) - Date.UTC(season, 0, 1)) / 86_400_000) + 1
+                : 0;
         // podcast-pub's documented contract reads metadata.coverImage (map.ts); options.coverPath
         // alone was silently dropped by the mapper — episodes published without a cover.
         const merged = {
@@ -1584,6 +1624,7 @@ function dailyPublishPrep(args: string[]): void {
                 audioFile,
                 runDate,
                 ...(hasCover ? { coverImage: resolvedCover } : {}),
+                ...(episode > 0 ? { itunes: { season, episode } } : {}),
             },
             options: {
                 audioFile,
@@ -1765,7 +1806,7 @@ function dailyPublishPartial(args: string[]): void {
  * Idempotent per post; commits + pushes the surfdash checkout (SURFDASH_AUTO_PUSH=0 keeps commit-only).
  */
 function dailyPatchSurfdash(args: string[]): void {
-    const [workDir = '', runDate = '', surfdashRoot = ''] = args;
+    const [workDir = '', runDate = '', surfdashRoot = '', pluginsPath = ''] = args;
     if (!surfdashRoot || !existsSync(surfdashRoot)) {
         console.log(`patch-surfdash: surfdash root '${surfdashRoot}' not found — skip`);
         return;
@@ -1866,6 +1907,39 @@ function dailyPatchSurfdash(args: string[]): void {
                 changed++;
             }
         }
+        // 20260929 prod feedback: CMS cards (homepage hero/tag/search) read EmDash rows whose
+        // featured_image is snapshotted at translate-sync time — i.e. the pre-patch
+        // `./assets/cover.png`, which the deployed worker does not serve (404). The patch only
+        // rewrites files, so re-sync the EmDash rows from the patched files through the same
+        // `kk executor` path as translate-sync (its plugin child gets EMDASH_API_TOKEN from kk
+        // config; a direct orchestrator spawn gets 403 from prod dev-bypass). Runs before the
+        // changed===0 early-return so re-runs heal rows an earlier commit left stale
+        // (20260925/20260929). Fixture tests pass no plugins-path and must not shell out.
+        if (pluginsPath === '') {
+            console.log('patch-surfdash: no plugins-path — skip emdash re-sync');
+        } else {
+            const syncIn = join(workDir, '31-publish', `${runDate}_23_patch-surfdash_sync.json`);
+            const syncResult = join(workDir, '31-publish', `${runDate}_23_patch-surfdash_sync-result.json`);
+            sted('write', syncIn, () =>
+                writeJqText(syncIn, { body: '', options: { operation: 'sync', postPath: join(slugDir, 'index.md') } }),
+            );
+            const status = kk(
+                ['executor', 'run', 'surfdash-pub', '--in', syncIn, '--out', syncResult, ...pluginsArgs(pluginsPath)],
+                { unset: NESTED_SPUR_UNSET, soft: true },
+            );
+            const result = isFile(syncResult) ? readJson(syncResult) : null;
+            const ok =
+                status === 0 &&
+                result !== null &&
+                (result.ok === true || (result.targets as Array<{ ok?: boolean }> | undefined)?.[0]?.ok === true);
+            if (!ok) {
+                console.log(
+                    `patch-surfdash: WARNING: emdash re-sync failed (exit ${status}) — CMS cards keep the stale cover; fix: kk executor run surfdash-pub --in ${syncIn} --out ${syncResult} --plugins-path ${pluginsPath}`,
+                );
+            } else {
+                console.log('patch-surfdash: emdash rows re-synced from patched files');
+            }
+        }
         if (changed === 0) {
             console.log('patch-surfdash: nothing to change');
             return;

package/plugins/kk/workflows/kk-daily-ai-voice.yaml

@@ -76,6 +76,8 @@ vars:
   # live — 发表 only on an ad-hoc wechat_live="true" run, deliberately, by hand.
   wechat_enabled: "true"
   wechat_live: "false"
+  # 合集 tag for the MP draft (passed to wechatmp-pub); other shows/collections override per workflow.
+  wechat_collection: "每日AI资讯"
   plan_max_items: "20"
   plan_translate_enabled: "true"
   article_write_enabled: "true"
@@ -191,7 +193,7 @@ states:
           # ADR-115 error tier: keep this prompt <= 1000 chars (0133 R3).
           input: |
             Read ${vars.work_dir}/21-plan/${vars.run_date}_03_plan_plan.json — metadata.docs (Doc[]) is the translated plan; don't edit it.
-            Write a morning-news prose article in natural simplified Chinese: title line "# 每日 AI 资讯 — ${vars.run_date}", an intro paragraph (2–3 sentences), then one section per item (## numbered title, 2–4 sentences of flowing prose, source link [来源：<label>](<sourceUri>) inline), a closing paragraph tying items together.
+            Write a morning-news prose article in natural simplified Chinese: title line "# 每日 AI 资讯 — ${vars.run_date}", an intro paragraph (2–3 sentences), then one section per item (## numbered title, 2–4 sentences of flowing prose, source link [来源：<label>](<sourceUri>) inline), a closing paragraph tying items together. 内容合规（平台审核）：全文不出现中国现任或历任党、国家、政府最高领导人的姓名，必须提及时以职务代称（如「国家主席」「国务院总理」）；不书写中国国内政治相关议题。
             Narrative register — smooth transitions between items, no list-like bulleting, no scaffold labels (「可做角度」「背景」「潜在影响」), no hedging (「读者宜另行核实」「仍待验证」). Keep facts/URLs from the plan verbatim; product/model names as-is; 「」 quotes.
             Output the markdown to ${vars.work_dir}/22-article/${vars.run_date}_06_article-write_body.md.
           expectFile: ${vars.work_dir}/22-article/${vars.run_date}_06_article-write_body.md
@@ -255,7 +257,7 @@ states:
           input: |
             Read ${vars.work_dir}/21-plan/${vars.run_date}_03_plan_plan.json — a JSON array of docs; use each doc's `id`; don't edit it. Also read the style contract ${vars.work_dir}/24-script/${vars.run_date}_08_script-write_style.md and obey its char caps and banned phrases.
             For each doc write a `lead` (a spoken bridge into the title, not containing it) and an `insight` (why it matters, grounded only in that doc's facts).
-            Write an `intro` hooking today's actual themes and an `outro`; no greeting or sign-off — the builder adds both.
+            Write an `intro` hooking today's actual themes and an `outro`; no greeting or sign-off — the builder adds both. 内容合规：不出现中国现任、历任最高领导人姓名（以职务代称），不涉及中国国内政治。
             Output only the JSON object {"version":1,"intro":"…","outro":"…","items":[{"id":"…","lead":"…","insight":"…"}]} with one item per doc id to ${vars.work_dir}/24-script/${vars.run_date}_08_script-write_plan.json.
           expectFile: ${vars.work_dir}/24-script/${vars.run_date}_08_script-write_plan.json
           allowAppend: false
@@ -355,7 +357,7 @@ states:
           input: |
             Read ${vars.work_dir}/21-plan/${vars.run_date}_03_plan_plan.json — a JSON array of news docs; don't edit it.
             Take item 1 only (its "title" and "metadata.brief") — the lead story.
-            Write a Xiaohongshu note: "title" (≤20 chars) and "body" (Chinese note ≤1000 chars, ending with 3-5 hashtags).
+            Write an XHS note: "title" (≤20 chars) and "body" (Chinese note ≤1000 chars, ending with 3-5 hashtags). 不提领导人姓名，不涉国内政治。
             Write exactly ${vars.xhs_cards} cards: card 1 role "cover", the rest "content", except a final "closing". Each card carries "role", "copy" (Chinese on-card headline ≤12 chars), "points" (2-4 short Chinese bullets), "layout", "subject" (one concrete English visual scene), "sourcePoint" (the lead-story fact that card uses).
             Ground every "copy", "points" and "sourcePoint" in that story alone — no invented facts. No text, letters, numbers, logos or real people in "subject".
             Output strictly valid JSON {"title": "...", "body": "...", "cards": [...]} to ${vars.work_dir}/25-xhs/${vars.run_date}_18_xhs_plan.json.
@@ -383,7 +385,7 @@ states:
         onError: continue
         options:
           command: >-
-            kk stage daily-publish-wechat "${vars.work_dir}" "${vars.run_date}" "${vars.plugins_path}" "${vars.wechat_live}"
+            kk stage daily-publish-wechat "${vars.work_dir}" "${vars.run_date}" "${vars.plugins_path}" "${vars.wechat_live}" "${vars.wechat_collection}"
 
   - id: publish
     description: "Classify the sequenced surfdash → podcast outcomes full|partial|none into publish-status.txt (exit 0 in all three — routing is transition-owned; task 0118: per-target classification over the merged result.json)"
@@ -398,12 +400,12 @@ states:
           var: publish_status
 
   - id: patch-surfdash
-    description: "20260917 #4/#6 — patch published surfdash posts: podcast cover URL into image/og_image frontmatter + podcast cross-link (zh/en/ja), commit+push"
+    description: "20260917 #4/#6 — patch published surfdash posts: podcast cover URL into image/og_image frontmatter + podcast cross-link (zh/en/ja), re-sync EmDash rows so CMS cards get the CDN cover (20260929), commit+push"
     onEnter:
       - kind: shell
         options:
           command: >-
-            kk stage daily-patch-surfdash "${vars.work_dir}" "${vars.run_date}" "${vars.surfdash_root}"
+            kk stage daily-patch-surfdash "${vars.work_dir}" "${vars.run_date}" "${vars.surfdash_root}" "${vars.plugins_path}"
 
   - id: publish-partial
     description: "Partial publish — at least one target published, at least one failed; echo warning, write a .spur/run marker, preserve result.json for --retry-from, continue to done (task 0118: surfdash failure = no-URL fallback notes; podcast failure = its own partial)"
@@ -604,6 +606,7 @@ transitions:
   - from: review
     to: failed
     description: "VoiceScript rejected -> abort"
+    terminalReason: failed-agent
     guard:
       kind: shell
       options:
@@ -684,9 +687,17 @@ transitions:
     guard:
       kind: always
 
+  - from: xhs-publish
+    to: wechat-publish
+    description: "XHS dispatched (or skipped: onError continue) -> chain the WeChat draft when wechat_enabled. 20260929 dogfood: publish-podcast's edges are first-match-wins, so the direct publish-podcast -> wechat-publish edge is unreachable whenever xhs_enabled=true — wechat must hang off the xhs path"
+    guard:
+      kind: shell
+      options:
+        command: 'test "${vars.wechat_enabled}" = "true"'
+
   - from: xhs-publish
     to: publish
-    description: "XHS dispatched (or skipped: onError continue) -> classify the surfdash/podcast outcomes; XHS contributes no target to the classification"
+    description: "XHS dispatched (or skipped: onError continue) -> classify the surfdash/podcast outcomes; XHS contributes no target to the classification (direct edge when wechat_enabled is false)"
     guard:
       kind: always
 
@@ -735,6 +746,7 @@ transitions:
   - from: publish
     to: failed
     description: "Publish none (all targets failed or fan-out crash) -> pipeline failed"
+    terminalReason: failed-check
     guard:
       kind: shell
       options: