@gobing-ai/knowledge-kit 0.0.34 → 0.0.35

package/.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
     },
     "plugins": [
         {
-            "version": "0.0.34",
+            "version": "0.0.35",
             "name": "kk",
             "source": "./plugins/kk",
             "description": "knowledge-kit — a set of toolkit for knowledge management"

package/dist/index.js

@@ -43509,7 +43509,7 @@ function probeVerdict(exitCode, stdout) {
   }
   const fields = typeof parsed === "object" && parsed !== null ? parsed : undefined;
   const status = typeof fields?.status === "string" ? fields.status : undefined;
-  if ((status === "verified" || status === "refused") && (exitCode === 0 || exitCode === 1)) {
+  if (status === "verified" && exitCode === 0 || status === "refused" && exitCode === 1) {
     return {
       status,
       ...typeof fields?.profileDir === "string" ? { profileDir: fields.profileDir } : {},

package/package.json

@@ -1,6 +1,6 @@
 {
     "name": "@gobing-ai/knowledge-kit",
-    "version": "0.0.34",
+    "version": "0.0.35",
     "description": "An ingest → create → publish content pipeline CLI (Bun).",
     "repository": {
         "type": "git",

package/plugins/generations/daily-article-gen/dist/index.js

@@ -22062,6 +22062,10 @@ function tagsFor(docs) {
 var LEADING_H1_RE = /^#(?!#)\s*[^\n]+\n+/.source;
 var NOTE_EXCLUDED_RE = /\u672A\u5199\u5165[\uFF1A:](.*)/g;
 var HEX_ID_RE = /(?:[a-zA-Z0-9_-]+:)?[0-9a-f]{16,64}/g;
+var BOUNDARY_NOTE_RE = /^>[^\n]*(?:\u9009\u9898\u8FB9\u754C|\u672A\u5199\u5165)[^\n]*\n?/gm;
+function stripBoundaryNote(body) {
+  return body.replace(BOUNDARY_NOTE_RE, "");
+}
 function excludedIdsFromNote(body) {
   const ids = new Set;
   if (!body)
@@ -22080,18 +22084,27 @@ function docsToContent(docs, date6, bodyOverride) {
   const title = episodeTitle(date6);
   let body = renderDailyArticle(docs, date6);
   if (bodyOverride) {
-    const stripped = stripScaffold(bodyOverride.trim()).trim().replace(new RegExp(LEADING_H1_RE), "");
+    const cleaned = stripBoundaryNote(stripScaffold(bodyOverride.trim()).trim().replace(new RegExp(LEADING_H1_RE), "")).trim();
     body = `# ${title}
 
-${stripped}
+${cleaned}
 `;
+  } else {
+    body = stripBoundaryNote(body);
   }
+  const excludedIds = [...excludedIdsFromNote(bodyOverride)];
   return ContentSchema.parse({
     title,
     body,
     format: "markdown",
     references: docs.filter((d) => d.sourceUri).map((d) => ({ url: d.sourceUri, title: d.title, cite: d.id })),
-    metadata: { generator: "kk:daily-article", date: date6, doc_count: docs.length, tags: tagsFor(docs) }
+    metadata: {
+      generator: "kk:daily-article",
+      date: date6,
+      doc_count: docs.length,
+      tags: tagsFor(docs),
+      ...excludedIds.length > 0 ? { excludedIds } : {}
+    }
   });
 }
 async function processGeneratorIO(inputPath, outputPath) {
@@ -22140,6 +22153,7 @@ if (import.meta.main) {
 }
 export {
   stripScaffold,
+  stripBoundaryNote,
   renderDailyArticle,
   processGeneratorIO,
   main2 as main,

package/plugins/generations/daily-article-gen/src/index.ts

@@ -141,6 +141,20 @@ const LEADING_H1_RE = /^#(?!#)\s*[^\n]+\n+/.source;
 const NOTE_EXCLUDED_RE = /未写入[：:](.*)/g;
 const HEX_ID_RE = /(?:[a-zA-Z0-9_-]+:)?[0-9a-f]{16,64}/g;
 
+/**
+ * The 选题边界 note is pipeline bookkeeping: it names dropped doc ids and the rule that dropped them.
+ * It is the machine-readable exclusion signal (`excludedIdsFromNote`), so it stays in the agent's body
+ * file — but it must never reach a published body, where it printed an internal doc id and pipeline
+ * vocabulary to readers (operator report 2026-10-10). Matches only blockquote lines, which the article
+ * body reserves for exactly this note, so ordinary prose is untouched.
+ */
+const BOUNDARY_NOTE_RE = /^>[^\n]*(?:选题边界|未写入)[^\n]*\n?/gm;
+
+/** Strip the internal boundary note from a body destined for a customer-visible surface. */
+export function stripBoundaryNote(body: string): string {
+    return body.replace(BOUNDARY_NOTE_RE, '');
+}
+
 /** Parse the boundary-note ids out of an article body so references/metadata never cite excluded docs (judge finding 2026-10-02). */
 export function excludedIdsFromNote(body: string | undefined): Set<string> {
     const ids = new Set<string>();
@@ -162,15 +176,30 @@ export function docsToContent(docs: Doc[], date: string, bodyOverride?: string):
     if (bodyOverride) {
         // The canonical episode title is computed from the lead item; a leading H1 the agent may
         // still have written is replaced (not duplicated) so body and metadata agree on every path.
-        const stripped = stripScaffold(bodyOverride.trim()).trim().replace(new RegExp(LEADING_H1_RE), '');
-        body = `# ${title}\n\n${stripped}\n`;
+        // The boundary note is stripped here, not dropped upstream: the agent's body file keeps it as
+        // the exclusion signal, this Content is what readers see.
+        const cleaned = stripBoundaryNote(
+            stripScaffold(bodyOverride.trim()).trim().replace(new RegExp(LEADING_H1_RE), ''),
+        ).trim();
+        body = `# ${title}\n\n${cleaned}\n`;
+    } else {
+        body = stripBoundaryNote(body);
     }
+    // The ids the note named become metadata, so consumers that must filter excluded docs (show-notes
+    // plan copies) read a field instead of re-parsing a note that no longer exists.
+    const excludedIds = [...excludedIdsFromNote(bodyOverride)];
     return ContentSchema.parse({
         title,
         body,
         format: 'markdown',
         references: docs.filter((d) => d.sourceUri).map((d) => ({ url: d.sourceUri, title: d.title, cite: d.id })),
-        metadata: { generator: 'kk:daily-article', date, doc_count: docs.length, tags: tagsFor(docs) },
+        metadata: {
+            generator: 'kk:daily-article',
+            date,
+            doc_count: docs.length,
+            tags: tagsFor(docs),
+            ...(excludedIds.length > 0 ? { excludedIds } : {}),
+        },
     });
 }
 

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

@@ -22116,6 +22116,7 @@ function buildNewsVoiceScript(docs, options) {
   const defaultEffectPreset = options?.effectPreset;
   const broadcastDate = options?.date ?? new Date;
   const dateStr = formatBroadcastDate(broadcastDate, language);
+  const episodeDate = `${broadcastDate.getFullYear()}-${String(broadcastDate.getMonth() + 1).padStart(2, "0")}-${String(broadcastDate.getDate()).padStart(2, "0")}`;
   const showName = "\u54B8\u8BDD\u54B8\u8BF4\u4E4B\u6BCF\u65E5AI\u4E71\u7096";
   let hostName = options?.speakers?.host?.name || voiceProfile.replace(/[-_]?(news|voice|bot|profile)$/i, "") || voiceProfile;
   if (hostName.length > 0) {
@@ -22157,7 +22158,7 @@ function buildNewsVoiceScript(docs, options) {
       segments: segments2
     });
   }
-  const title = options?.title || (isZh ? `\u6BCF\u65E5AI\u4E71\u7096\uFF08\u5171${docs.length}\u6761\uFF09` : `\u6BCF\u65E5AI\u4E71\u7096 (${docs.length} stories)`);
+  const title = options?.title || `\u6BCF\u65E5AI\u4E71\u7096 \xB7 ${episodeDate}`;
   const segments = [];
   const scriptPlan = isZh ? options?.scriptPlan : undefined;
   const planItems = new Map((scriptPlan?.items ?? []).map((item) => [item.id, item]));

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

@@ -313,6 +313,11 @@ export function buildNewsVoiceScript(docs: Doc[], options?: ScriptBuilderOptions
     const defaultEffectPreset = options?.effectPreset;
     const broadcastDate = options?.date ?? new Date();
     const dateStr = formatBroadcastDate(broadcastDate, language);
+    // Canonical episode identifier: date-only, matching daily-article-gen's `episodeTitle`. One
+    // title has to cover the post body, ID3, RSS and the show notes, so the voicescript default
+    // must not invent its own form (it used to be `共N条`, which disagreed with the article title
+    // on every path where the workflow passes no explicit title).
+    const episodeDate = `${broadcastDate.getFullYear()}-${String(broadcastDate.getMonth() + 1).padStart(2, '0')}-${String(broadcastDate.getDate()).padStart(2, '0')}`;
     // Column of 咸话咸说. The show is not the speaker; the host is.
     const showName = '咸话咸说之每日AI乱炖';
     let hostName =
@@ -365,7 +370,7 @@ export function buildNewsVoiceScript(docs: Doc[], options?: ScriptBuilderOptions
         });
     }
 
-    const title = options?.title || (isZh ? `每日AI乱炖（共${docs.length}条）` : `每日AI乱炖 (${docs.length} stories)`);
+    const title = options?.title || `每日AI乱炖 · ${episodeDate}`;
     const segments: VoiceScriptSegment[] = [];
 
     // F613: the plan owns the words on the zh path; en keeps templates (design: substitution is

package/plugins/generations/gemini-tts/dist/index.js

@@ -23041,7 +23041,8 @@ async function processGeneratorIO(inputPath, outputPath, clientOverride, transco
       let loudnessDip = false;
       let lastError;
       let attemptBody = generateBody;
-      for (let verifyAttempt = 0;verifyAttempt <= 2 && wavBytes === undefined; verifyAttempt++) {
+      const maxVerifyAttempts = Number.parseInt(process.env.VOICEBOX_MAX_VERIFY_ATTEMPTS ?? "2", 10) || 0;
+      for (let verifyAttempt = 0;verifyAttempt <= maxVerifyAttempts && wavBytes === undefined; verifyAttempt++) {
         if (verifyAttempt > 0) {
           verifyRetries = verifyAttempt;
           const chunkChars = Math.max(VOICEBOX_CHUNK_CHARS_MIN, Math.floor(attemptBody.max_chunk_chars / 2));
@@ -23079,7 +23080,7 @@ async function processGeneratorIO(inputPath, outputPath, clientOverride, transco
           try {
             const trans = await client.transcribe(attemptBytes, language);
             const fidelity = transcriptionFidelity(segment.text, trans.text);
-            if (fidelity < TRANSCRIPTION_FIDELITY_FLOOR && verifyAttempt < 2) {
+            if (fidelity < TRANSCRIPTION_FIDELITY_FLOOR && verifyAttempt < maxVerifyAttempts) {
               echoError(`voice-gen: transcription fidelity ${fidelity.toFixed(2)} below ${TRANSCRIPTION_FIDELITY_FLOOR} (looped/garbage audio): "${trans.text.slice(0, 80)}..."`);
               continue;
             }
@@ -23088,7 +23089,7 @@ async function processGeneratorIO(inputPath, outputPath, clientOverride, transco
         }
         const dip = detectLoudnessDip(attemptBytes);
         loudnessDip = dip.dip;
-        if (dip.dip && verifyAttempt < 2) {
+        if (dip.dip && verifyAttempt < maxVerifyAttempts) {
           echoError(`voice-gen: loudness dip detected at ${dip.dipAtSec ?? "?"}s \u2014 regenerating segment`);
           continue;
         }

package/plugins/generations/omni-voice-gen/plugin.json

@@ -1,6 +1,6 @@
 {
     "name": "omni-voice-gen",
     "kind": "generator",
-    "entry": "./dist/index.js",
+    "entry": "./bin/omni-voice-gen",
     "version": "1.0.0"
 }

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

@@ -23038,7 +23038,8 @@ async function processGeneratorIO(inputPath, outputPath, clientOverride, transco
       let loudnessDip = false;
       let lastError;
       let attemptBody = generateBody;
-      for (let verifyAttempt = 0;verifyAttempt <= 2 && wavBytes === undefined; verifyAttempt++) {
+      const maxVerifyAttempts = Number.parseInt(process.env.VOICEBOX_MAX_VERIFY_ATTEMPTS ?? "2", 10) || 0;
+      for (let verifyAttempt = 0;verifyAttempt <= maxVerifyAttempts && wavBytes === undefined; verifyAttempt++) {
         if (verifyAttempt > 0) {
           verifyRetries = verifyAttempt;
           const chunkChars = Math.max(VOICEBOX_CHUNK_CHARS_MIN, Math.floor(attemptBody.max_chunk_chars / 2));
@@ -23076,7 +23077,7 @@ async function processGeneratorIO(inputPath, outputPath, clientOverride, transco
           try {
             const trans = await client.transcribe(attemptBytes, language);
             const fidelity = transcriptionFidelity(segment.text, trans.text);
-            if (fidelity < TRANSCRIPTION_FIDELITY_FLOOR && verifyAttempt < 2) {
+            if (fidelity < TRANSCRIPTION_FIDELITY_FLOOR && verifyAttempt < maxVerifyAttempts) {
               echoError(`voice-gen: transcription fidelity ${fidelity.toFixed(2)} below ${TRANSCRIPTION_FIDELITY_FLOOR} (looped/garbage audio): "${trans.text.slice(0, 80)}..."`);
               continue;
             }
@@ -23085,7 +23086,7 @@ async function processGeneratorIO(inputPath, outputPath, clientOverride, transco
         }
         const dip = detectLoudnessDip(attemptBytes);
         loudnessDip = dip.dip;
-        if (dip.dip && verifyAttempt < 2) {
+        if (dip.dip && verifyAttempt < maxVerifyAttempts) {
           echoError(`voice-gen: loudness dip detected at ${dip.dipAtSec ?? "?"}s \u2014 regenerating segment`);
           continue;
         }

package/plugins/generations/voice-gen/src/index.ts

@@ -212,7 +212,13 @@ export async function processGeneratorIO(
             let loudnessDip = false;
             let lastError: unknown;
             let attemptBody = generateBody;
-            for (let verifyAttempt = 0; verifyAttempt <= 2 && wavBytes === undefined; verifyAttempt++) {
+            // Verify-retry budget. Hosted TTS is billed per request against a daily quota
+            // (Gemini Tier 1: 100/day), and each retry is a full re-synthesis — a 41-segment
+            // episode with the default 2 retries asks for up to 123 requests and dies partway
+            // through the render with no audio (daily-generate failures of 2026-10-06/07/09/10).
+            // Lower it for a quota-bound driver; 0 keeps the dip/fidelity result for QC instead.
+            const maxVerifyAttempts = Number.parseInt(process.env.VOICEBOX_MAX_VERIFY_ATTEMPTS ?? '2', 10) || 0;
+            for (let verifyAttempt = 0; verifyAttempt <= maxVerifyAttempts && wavBytes === undefined; verifyAttempt++) {
                 if (verifyAttempt > 0) {
                     verifyRetries = verifyAttempt;
                     const chunkChars = Math.max(VOICEBOX_CHUNK_CHARS_MIN, Math.floor(attemptBody.max_chunk_chars / 2));
@@ -262,7 +268,7 @@ export async function processGeneratorIO(
                     try {
                         const trans = await client.transcribe(attemptBytes, language);
                         const fidelity = transcriptionFidelity(segment.text, trans.text);
-                        if (fidelity < TRANSCRIPTION_FIDELITY_FLOOR && verifyAttempt < 2) {
+                        if (fidelity < TRANSCRIPTION_FIDELITY_FLOOR && verifyAttempt < maxVerifyAttempts) {
                             echoError(
                                 `voice-gen: transcription fidelity ${fidelity.toFixed(2)} below ${TRANSCRIPTION_FIDELITY_FLOOR} (looped/garbage audio): "${trans.text.slice(0, 80)}..."`,
                             );
@@ -278,7 +284,7 @@ export async function processGeneratorIO(
                 // (dogfood 2026-09-08 A/B: same text regenerated clean, so retry usually clears).
                 const dip = detectLoudnessDip(attemptBytes);
                 loudnessDip = dip.dip;
-                if (dip.dip && verifyAttempt < 2) {
+                if (dip.dip && verifyAttempt < maxVerifyAttempts) {
                     echoError(`voice-gen: loudness dip detected at ${dip.dipAtSec ?? '?'}s — regenerating segment`);
                     continue;
                 }

package/plugins/kk/plugin.json

@@ -1,5 +1,5 @@
 {
     "name": "kk",
-    "version": "0.0.34",
+    "version": "0.0.35",
     "description": "knowledge-kit Claude Code plugin: skills, commands, subagents, hooks, and rules"
 }