@gobing-ai/knowledge-kit 0.0.19 → 0.0.21

package/.claude-plugin/marketplace.json

@@ -6,7 +6,7 @@
     },
     "plugins": [
         {
-            "version": "0.0.19",
+            "version": "0.0.21",
             "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": "0.0.19",
+    "version": "0.0.21",
     "description": "An ingest → create → publish content pipeline CLI (Bun).",
     "repository": {
         "type": "git",

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

@@ -22010,6 +22010,13 @@ function stripLeadingTitleEcho(body, title) {
     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) {
+  const kept = body.split(`
+`).filter((line) => !TAG_METADATA_LINE_RE.test(line)).join(`
+`);
+  return kept === body ? body : kept.trim();
+}
 function splitFactText(text, maxChars = MAX_FACT_SEGMENT_CHARS) {
   if (text.length <= maxChars)
     return [text];
@@ -22091,7 +22098,7 @@ function buildNewsVoiceScript(docs, options) {
   docs.forEach((doc2, idx) => {
     const isLast = idx === docs.length - 1 && docs.length > 1;
     const cleanTitle = (doc2.title ?? "").trim();
-    const cleanBody = stripLeadingTitleEcho((doc2.body ?? "").trim(), cleanTitle);
+    const cleanBody = stripLeadingTitleEcho(stripTagMetadataLines((doc2.body ?? "").trim()), cleanTitle);
     const lead = isZh ? idx === 0 ? `\u9996\u5148\u6765\u804A\u804A\u5927\u5BB6\u975E\u5E38\u5173\u6CE8\u7684\u3010${cleanTitle}\u3011\u3002` : isLast ? `\u6700\u540E\uFF0C\u6765\u770B\u770B\u4ECA\u5929\u7684\u6700\u540E\u4E00\u6761\u8D44\u8BAF\uFF0C\u3010${cleanTitle}\u3011\u3002` : `\u63A5\u7740\u6211\u4EEC\u628A\u76EE\u5149\u8F6C\u5411\u53E6\u4E00\u6761\u91CD\u8981\u8FDB\u5C55\uFF0C\u3010${cleanTitle}\u3011\u3002` : idx === 0 ? `First up today, let's look at ${cleanTitle}.` : isLast ? `And finally today, let's wrap up with ${cleanTitle}.` : `Next, turning our attention to ${cleanTitle}.`;
     const normalizedLead = normalizeBroadcastText(lead, isZh);
     const normalizedBody = cleanBody === "" ? "" : normalizeBroadcastText(cleanBody, isZh);
@@ -22341,6 +22348,7 @@ if (import.meta.main) {
 }
 export {
   synthesizePersonalUnderstanding,
+  stripTagMetadataLines,
   stripLeadingTitleEcho,
   splitFactText,
   processGeneratorIO,

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

@@ -245,6 +245,24 @@ export function stripLeadingTitleEcho(body: string, title: string): string {
     return body.slice(end).replace(/^[，。、：:；;!！?？…\s「」『』【】]+/, '');
 }
 
+/**
+ * Drop `**标签**: …` metadata lines from a plan doc body (0139 R3).
+ *
+ * `plan-translate` keeps the research note's trailing tag line in the body it hands over; spoken
+ * aloud it reads as a `标签，…` tail the script validator rejects. The line is article metadata,
+ * not copy — remove whole lines that open with the 标签 label + separator; anything else is left
+ * for the validator to fail loud on.
+ */
+const TAG_METADATA_LINE_RE = /^\s*\**\s*标签\s*\**\s*[：:，,]/;
+
+export function stripTagMetadataLines(body: string): string {
+    const kept = body
+        .split('\n')
+        .filter((line) => !TAG_METADATA_LINE_RE.test(line))
+        .join('\n');
+    return kept === body ? body : kept.trim();
+}
+
 /**
  * Split a fact text into chunks of at most `maxChars` (task 0139 R4).
  *
@@ -360,7 +378,7 @@ export function buildNewsVoiceScript(docs: Doc[], options?: ScriptBuilderOptions
         const cleanTitle = (doc.title ?? '').trim();
         // R1: drop the title restatement plan-translate prepends to the body, so the lead's
         // 【title】 is the only place the title is spoken for this item.
-        const cleanBody = stripLeadingTitleEcho((doc.body ?? '').trim(), cleanTitle);
+        const cleanBody = stripLeadingTitleEcho(stripTagMetadataLines((doc.body ?? '').trim()), cleanTitle);
 
         // Conversational lead
         const lead = isZh

package/plugins/generations/omni-voice-gen/src/omni_voice_gen/pipeline.py

@@ -133,8 +133,8 @@ def _transcribe(backend: Backend, wav: bytes, language: str) -> str | None:
         return None
 
 
-def _render_verified(backend: Backend, spec: SegmentSpec) -> tuple[bytes, float, int, str | None, bool]:
-    """Spec §5 verify-retry -> (wav, duration, verify_retries, transcription, loudness_dip)."""
+def _render_verified(backend: Backend, spec: SegmentSpec) -> tuple[bytes, float, int, str | None, bool, float | None]:
+    """Spec §5 verify-retry -> (wav, duration, verify_retries, transcription, loudness_dip, dip_at)."""
     for attempt in range(MAX_VERIFY_RETRIES + 1):
         final = attempt == MAX_VERIFY_RETRIES  # the final attempt is always accepted
         if attempt:
@@ -154,7 +154,7 @@ def _render_verified(backend: Backend, spec: SegmentSpec) -> tuple[bytes, float,
         if dip and not final:
             _log(f"loudness dip detected at {dip_at}s — regenerating segment")
             continue
-        return wav, duration, attempt, transcription, dip
+        return wav, duration, attempt, transcription, dip, dip_at
     raise AssertionError("unreachable: the final verify attempt always returns")  # pragma: no cover
 
 
@@ -212,7 +212,7 @@ def _render(
             "seed": None if segment.seed is None else int(segment.seed),
             "speed": None if segment.speed is None else float(segment.speed),
         }
-        wav, duration, retries, transcription, dip = _render_verified(backend, spec)
+        wav, duration, retries, transcription, dip, dip_at = _render_verified(backend, spec)
 
         gap_ms = segment.gap_ms or 0
         wavs.append(wav)
@@ -233,7 +233,15 @@ def _render(
                 }
             )
         )
-        audit_meta.append({"duration": duration, "transcription": transcription, "loudness_dip": dip})
+        audit_meta.append(
+            {
+                "duration": duration,
+                "transcription": transcription,
+                "loudness_dip": dip,
+                "dip_at": dip_at,
+                "verify_retries": retries,
+            }
+        )
         retry_note = f" ({retries} verify {'retry' if retries == 1 else 'retries'})" if retries else ""
         _log(f"segment {index + 1}/{total} done in {monotonic() - segment_started:.1f}s{retry_note}")
 

package/plugins/generations/omni-voice-gen/src/omni_voice_gen/qc.py

@@ -48,6 +48,7 @@ class SegmentQualityAudit:
     repetition_detected: bool
     duration_anomaly: bool
     fidelity_score: float | None
+    verify_retries: int = 0
     issues: list[str] = field(default_factory=list)
 
 
@@ -158,21 +159,6 @@ def audit_voice_segments(
             if fidelity < TRANSCRIPTION_FIDELITY_FLOOR:
                 issues.append(f"transcription_fidelity_low:{fidelity:.2f}")
 
-        # Repetition
-        repetition = detect_repetitions(text) or (
-            detect_repetitions(transcription) if transcription is not None else False
-        )
-        if repetition:
-            issues.append("repetition_detected")
-
-        # Loudness dip: reuse the verify-phase result when present; only run the windowed
-        # scan when the verify phase did not (meta.loudness_dip absent).
-        dip, dip_at = (
-            (bool(meta["loudness_dip"]), None) if "loudness_dip" in meta else detect_loudness_dip(wav)
-        )
-        if dip:
-            issues.append(f"loudness_dip:{f'{dip_at}s' if dip_at is not None else 'unknown'}")
-
         # Duration anomaly
         ratio_anomaly = duration > MAX_DURATION_RATIO * max_sec if max_sec > 0 else False
         abs_anomaly = duration > max_sec + MAX_ABS_DIFF_SEC if max_sec > 0 else False
@@ -187,6 +173,27 @@ def audit_voice_segments(
                 f"duration_anomaly: expected {min_sec:.2f}s-{max_sec:.2f}s, got {duration:.2f}s"
             )
 
+        # Repetition — transcription channel only, corroborated by duration anomaly (voice-gen
+        # qc.ts parity): the battery is an ASR/TTS-artifact detector, not a source-text lint.
+        # Source strings like version numbers ("1.1.1.") are legitimate copy and speak fine
+        # (dogfood 2026-09-18: a version string in the source tripped the word-run regex and
+        # condemned a 70-minute render that no regeneration could fix). A real TTS loop always
+        # inflates duration; a Whisper phantom loop over trailing silence does not.
+        repetition = detect_repetitions(transcription) if transcription is not None else False
+        if repetition and duration_anomaly:
+            # Bare token: pipeline.py's R5a gate matches issues by exact membership.
+            issues.append("repetition_detected")
+
+        # Loudness dip: reuse the verify-phase result when present; only run the windowed
+        # scan when the verify phase did not (meta.loudness_dip absent).
+        dip, dip_at = (
+            (bool(meta["loudness_dip"]), meta.get("dip_at"))
+            if "loudness_dip" in meta
+            else detect_loudness_dip(wav)
+        )
+        if dip:
+            issues.append(f"loudness_dip:{f'{dip_at}s' if dip_at is not None else 'unknown'}")
+
         audits.append(
             SegmentQualityAudit(
                 segment_index=index,
@@ -198,6 +205,7 @@ def audit_voice_segments(
                 repetition_detected=repetition,
                 duration_anomaly=duration_anomaly,
                 fidelity_score=fidelity,
+                verify_retries=int(meta.get("verify_retries", 0) or 0),
                 issues=issues,
             )
         )

package/plugins/ingestions/aihot-ingest/dist/index.js

@@ -22086,7 +22086,7 @@ function mapAihotItemsToDocs(items) {
 
 // ../../plugins/ingestions/aihot-ingest/src/rss.ts
 import { createHash as createHash2 } from "crypto";
-var AIHOT_FEED_URL_DEFAULT = "https://techcrunch.com/category/artificial-intelligence/feed/";
+var AIHOT_FEED_URL_DEFAULT = "https://arstechnica.com/ai/feed/";
 function stripMarkup(raw) {
   return raw.replace(/<!\[CDATA\[([\s\S]*?)\]\]>/g, "$1").replace(/<[^>]+>/g, " ").replace(/&nbsp;/g, " ").replace(/&amp;/g, "&").replace(/&lt;/g, "<").replace(/&gt;/g, ">").replace(/&quot;/g, '"').replace(/&#39;|&apos;/g, "'").replace(/\s+/g, " ").trim();
 }

package/plugins/ingestions/aihot-ingest/plugin.json

@@ -3,5 +3,5 @@
     "kind": "ingestion",
     "entry": "./dist/index.js",
     "version": "1.0.0",
-    "description": "Ingests fresh AI news from an RSS feed (default TechCrunch AI) into Doc[]; legacy aihot.virxact.com API behind AIHOT_SOURCE=api"
+    "description": "Ingests fresh AI news from an RSS feed (default Ars Technica AI, full text) into Doc[]; legacy aihot.virxact.com API behind AIHOT_SOURCE=api"
 }

package/plugins/ingestions/aihot-ingest/src/rss.ts

@@ -15,8 +15,18 @@ import type { Doc } from '@gobing-ai/kk-core';
  * still applies to feed items.
  */
 
-/** Default fresh AI-news feed (RSS 2.0, high volume, reachable without auth). */
-export const AIHOT_FEED_URL_DEFAULT = 'https://techcrunch.com/category/artificial-intelligence/feed/';
+/** Default fresh AI-news feed (RSS 2.0, full text, reachable without auth).
+ *
+ * Why Ars Technica (2026-09-20, dogfood): the previous default (TechCrunch AI
+ * category feed) stopped shipping `content:encoded` — items now carry only
+ * 86–236-char `description` summaries. Every such body lands below the planner's
+ * STUB_BODY_FLOOR (300 chars), scores quality 1, and is auto-rejected at
+ * `qc_min_quality=2`, so the aihot leg contributed ~0 usable items (the
+ * 2026-09-20 episode collapsed to 3 items / 5m33s). The Ars Technica AI feed
+ * ships full `content:encoded` (≈900–1800 chars) on every item. The parser
+ * already prefers `content:encoded`, so only the URL changes.
+ */
+export const AIHOT_FEED_URL_DEFAULT = 'https://arstechnica.com/ai/feed/';
 
 export type RssFetchLike = (url: string | URL | Request, init?: RequestInit) => Promise<Response>;
 

package/plugins/kk/commands/workflow-run.md

@@ -35,7 +35,10 @@ Parse per the frozen rule:
   `--fixture` → force `fixture: "true"`.
 - Profile `kk-itc`: `--dir <path>` (default `$works_dir/kk-itc/<kebab>`), `--playbook generic|english|wechat` (default `generic`),
   `--research` (`true`|`false`, default `false`), `--judge` (`true`|`false`, default `false`),
-  `--outline <a|b|c>` (default empty), `--writer itc-generating|topic` (default `itc-generating`).
+  `--outline <a|b|c>` (default empty), `--writer itc-generating|topic` (default `itc-generating`),
+  `--auto`, `--unslop true|false` (default `false`), `--max-revisions <N>` (default `2`),
+  `--cover`, `--inline-images`, `--targets <a,b>` (default empty), `--live`,
+  `--source-locale en|zh|ja` (default `en`; `--playbook wechat` without it implies `zh`).
   `--fixture` passed with `kk-itc` or `kk-solo-podcast` → exit 1 (`fixture is storm-only`).
 - Profile `kk-solo-podcast`: `--dir <path>` (default `$works_dir/kk-solo-podcast/<date>`), `--outline <a|b|c>`
   (default empty), `--duration <min>` (default `8`), `--language <code>` (default `en`),
@@ -48,7 +51,9 @@ Parse per the frozen rule:
 
 Resolve these before step 2: `NAME`, `TOPIC`, `INPUT_FILE` (empty in sentence mode), `FIXTURE`
 (`true`|`false`), `FORCE` (`true`|`false`), `DIR`, `PLAYBOOK`, `RESEARCH`, `JUDGE`, `OUTLINE`,
-`WRITER`, `DURATION` (`8`), `LANGUAGE` (`en`), `SCRIPT_APPROVED` (`true`|`false`),
+`WRITER`, `AUTO` (`true`|`false`), `UNSLOP` (`true`|`false`), `MAX_REVISIONS` (`2`), `COVER`
+(`true`|`false`), `INLINE_IMAGES` (`true`|`false`), `TARGETS`, `LIVE` (`true`|`false`),
+`SOURCE_LOCALE` (`en`), `DURATION` (`8`), `LANGUAGE` (`en`), `SCRIPT_APPROVED` (`true`|`false`),
 `VOICE_PROFILE` (from `VOICEBOX_DEFAULT_PROFILE`, default empty).
 
 ## 2. Resolve the config
@@ -185,6 +190,14 @@ mkdir -p "$work_dir"
 `work_dir="${DIR:-$works_dir/kk-itc/$kebab}"`. Empty derived `kebab` → exit 1. An explicit
 `--dir` still wins, and the default never writes into the caller's cwd.
 
+Binding (0153): `--auto` binds `auto=true` and defaults `research`, `judge` and `unslop` to
+`true` unless the operator passed them explicitly. `--live` is never implied — `publish_live`
+stays `false` unless `--live` is passed explicitly (ADR-022). `--playbook wechat` without
+`--source-locale` binds `source_locale=zh`; an explicit `--source-locale` always wins.
+`--targets <a,b>` binds `publish_targets`; `--cover` / `--inline-images` bind
+`cover_enabled` / `inline_images` (`false` unless passed); `--max-revisions` binds
+`max_revisions` (default `2`).
+
 ### Profile `kk-solo-podcast`
 
 `date` = today's date in `YYYY-MM-DD` (`$(date +%Y-%m-%d)`).
@@ -218,13 +231,14 @@ spur workflow run "$dest" --vars \
 
 ### Profile `kk-itc`
 
-Bind vars: `topic`, `input_file`, `work_dir`, `writer`, `playbook`, `research`, `judge`, `outline`,
-`force`, `rubric` (`tech-accuracy`), `agent`. The judge verdict lands at `$work_dir/.itc-verdict.json`
-(fixed path inside the workflow).
+Bind vars: `topic`, `input_file`, `work_dir`, `writer`, `playbook`, `research`, `judge`, `auto`,
+`unslop`, `max_revisions`, `outline`, `force`, `rubric` (`tech-accuracy`), `agent`,
+`cover_enabled`, `inline_images`, `publish_targets`, `publish_live`, `source_locale`. The judge
+verdict lands at `$work_dir/.itc-verdict.json` (fixed path inside the workflow).
 
 ```bash
 spur workflow run "$dest" --vars \
-  "{\"topic\":\"$TOPIC\",\"input_file\":\"$INPUT_FILE\",\"work_dir\":\"$work_dir\",\"writer\":\"$WRITER\",\"playbook\":\"$PLAYBOOK\",\"research\":\"$RESEARCH\",\"judge\":\"$JUDGE\",\"outline\":\"$OUTLINE\",\"force\":\"$FORCE\",\"rubric\":\"tech-accuracy\",\"agent\":\"$AGENT\"}"
+  "{\"topic\":\"$TOPIC\",\"input_file\":\"$INPUT_FILE\",\"work_dir\":\"$work_dir\",\"writer\":\"$WRITER\",\"playbook\":\"$PLAYBOOK\",\"research\":\"$RESEARCH\",\"judge\":\"$JUDGE\",\"auto\":\"$AUTO\",\"unslop\":\"$UNSLOP\",\"max_revisions\":\"$MAX_REVISIONS\",\"outline\":\"$OUTLINE\",\"force\":\"$FORCE\",\"rubric\":\"tech-accuracy\",\"agent\":\"$AGENT\",\"cover_enabled\":\"$COVER\",\"inline_images\":\"$INLINE_IMAGES\",\"publish_targets\":\"$TARGETS\",\"publish_live\":\"$LIVE\",\"source_locale\":\"$SOURCE_LOCALE\"}"
 ```
 
 ### Profile `kk-solo-podcast`

package/plugins/kk/plugin.json

@@ -1,5 +1,5 @@
 {
     "name": "kk",
-    "version": "0.0.19",
+    "version": "0.0.21",
     "description": "knowledge-kit Claude Code plugin: skills, commands, subagents, hooks, and rules"
 }