akari-video 0.1.62 → 0.1.63

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akari-video",
-  "version": "0.1.62",
+  "version": "0.1.63",
   "description": "AKARI Video launcher CLI — start an AI-edited video project from any directory: scaffold, connection check, then hand over to Claude Code (or opencode). AKARI Video を opencode や Claude Code で、どのディレクトリからでも始めるための `akari` ランチャー CLI。接続確認（doctor）→ 未セットアップならプロジェクト雛形を作成 → AI エージェントを起動する。外部 npm 依存ゼロ（Node.js 組み込みモジュールのみ）。",
   "type": "module",
   "bin": {

package/vendor/docs/contract-2026-08-02-preview-parity.md

@@ -12,6 +12,7 @@
 | 2026-08-28 | v2 | `packages/frame-engine` の意味論へ統合し、検収をゴールデンフレームへ一本化。出口を OSR と GPU 直結の 2 本に固定し、互換経路を退役節へ移動 |
 | 2026-08-31 | v2.1 | §5.2 に断片 CSS の `vw` / `vh` 系単位の出力サイズ基準化（`viewport-units.js`。プレビューがウィンドウ幅基準で解いていた実機報告の修正）を追記 |
 | 2026-09-02 | v2.2 | §2.8 に字幕時計の規約（active cue の判定は両プレビューとも出力秒。共有カーネル `caption-clock`）を追記。Web UI を shell に揃えた 4 点（字幕時計・字幕フォント名・`slot-params.js` の差し込み・最下段 cut の track 規則）の記録 |
+| 2026-09-10 | v2.3 | §5.8 に shell 合成面の内部解像度（auto / 手動倍率・停止時等倍）と構図一致の許容差を追加 |
 
 ## 1. 役割分担
 
@@ -346,3 +347,38 @@ frame-engine が MP4 を全体ストリームとして読むか、`ftyp` / `moov
 Range として読むかは、完成画の意味論に影響しない。どちらの読み込み経路も同じ presentation 時刻の
 VideoFrame を §4.1 の評価点へ供給し、`elst.media_time`、B フレームの並べ替え、メディア終端を含めて
 golden の `diff 0` を満たさなければならない。ソース取得方法の変更をパリティ差の許容理由にしてはならない。
+
+### 5.8 プレビュー内部解像度（render scale）
+
+shell の frame-engine 合成面は、辺あたり倍率 `s ∈ {1, 0.5, 0.25}` を使える。
+設定 `akari.preview.renderScale` の語彙は `'auto' | '1' | '0.5' | '0.25'`、既定は `'auto'`。
+環境変数 `AKARI_FRAME_ENGINE_RENDER_SCALE` は設定より優先し、不正値は `auto` とする。
+設定項目のみとし、専用ボタンやポップアップは設けない。
+
+`auto` は canvas の `getBoundingClientRect()` に `devicePixelRatio` を掛けた表示寸法に対し、
+`output.width × s ≥ 表示幅` かつ `output.height × s ≥ 表示高` を満たす最小の倍率を選ぶ。
+上限は `1`、表示寸法が 0 の場合も `1` とする。canvas の `ResizeObserver` と
+`matchMedia('(resolution: <dpr>dppx)')` の change で再評価し、100 ms デバウンスする。
+内部寸法は各辺を `roundEven(output寸法 × s)`（最寄りの偶数へ丸め、中間は上へ）とし、下限は 2 px。
+
+`auto` では再生停止・スクラブ終了から 250 ms 後に `s = 1` で 1 コマ描き直し、
+再生・スクラブの再開で表示寸法に応じた倍率へ戻す。手動倍率は停止中も固定し、停止による追加描画はしない。
+倍率変更は scheduler・評価 plan・compose が共有する `renderOutput` の width / height を in-place で更新する。
+scheduler を作り直して warmup を捨てず、停止中の寸法更新は 1 コマの再描画へ反映する。
+developerMode の計測パネルに倍率・内部寸法・等倍寸法・モードを 1 行表示する。
+
+エンジンのレイヤー / layer-style cut の幾何は出力 px 基準（box = crop × 素材 px × scale、
+出力中心 + (x, y)）なので、shell は評価 plan の px 基準の transform（x / y、および px 基準の scale）を
+`s` で射影してから合成する。fit 基準の通常 cut の scale と正規化座標は不変とし、入力 plan を変更せず、
+`plan.output` は共有参照を保つ。エンジン・ゴールデン検収・書き出しは不変。
+
+一致条件は**構図一致 = 位置・大きさ・時刻・ツマミ値の一致**であり、内部画素数は異なってよい。
+[vgpu レイヤー契約 §4](./contract-2026-09-06-vgpu-layer-v0.md#4-api解像度ツマミ) の例外を
+shell の合成面全体へ拡張する。等倍の完成画を縮小した画像と縮小倍率で描いた画像を比較し、
+`packages/frame-engine/src/metrics/frame-diff.ts` の MAD で **≤ 2.0 / 255** を要求する。
+粒状ノイズ FX と dissolve 区間はこの比較から除外する。DOM 層（字幕 / overlay / 3D / ペン）、
+`summary.output`、`frameScale`、当たり判定、インスペクター、書き込み経路の規約は不変。
+
+**書き出しとゴールデン検収は常に等倍で、本設定の対象外**。GPU / OSR の page-runtime、
+書き出し receipt、§4.1・§4.2 のゴールデンと許容差、`packages/frame-engine` とその生成バンドルは変更しない。
+Web UI の合成面、デコード解像度、vgpu 固有の `previewScale`、3D の `PREVIEW_3D_MAX_RENDER_SIZE` も対象外とする。

package/vendor/packages/akari-launcher/package.json

@@ -1,6 +1,6 @@
 {
   "name": "akari-video",
-  "version": "0.1.62",
+  "version": "0.1.63",
   "description": "AKARI Video launcher CLI — start an AI-edited video project from any directory: scaffold, connection check, then hand over to Claude Code (or opencode). AKARI Video を opencode や Claude Code で、どのディレクトリからでも始めるための `akari` ランチャー CLI。接続確認（doctor）→ 未セットアップならプロジェクト雛形を作成 → AI エージェントを起動する。外部 npm 依存ゼロ（Node.js 組み込みモジュールのみ）。 [akari-video npm vendor: bin/akari.mjs is reference-only. These CLI entrypoints are not included in the akari-video npm package. Use `akari doctor --json` and run the path reported in `render_cut.path`. Full installations provide it in a monorepo checkout, ~/.akari/app, /Applications/AKARI Video.app/Contents/Resources/packages, or %LOCALAPPDATA%\\Programs\\@akari-videoshell\\resources\\packages.]",
   "type": "module",
   "files": [

package/vendor/packages/edit-lint/README.md

@@ -42,15 +42,26 @@ item. Each source is probed once, and media finding paths include the source ID.
   warnings unless `--silence-error-seconds` promotes long intervals to errors.
 - `media.volume`: reports mean and maximum volume. Findings are warnings unless
   `--max-volume-error-db` promotes an over-limit maximum to an error.
+- `media.source-range`: errors when a media item asks for a span the source does not have. Every
+  lane is covered (visual cuts, `bgm` / `narration` / `sfx` items, nested bag items) plus the v1
+  `audio.*` declarations, and the container duration is used so sources without an audio stream are
+  checked too. `source.in` at or beyond the duration and an effective `source.out` past the duration
+  are both errors; sources whose container duration is unavailable are skipped with a reason.
 - `media.audio-shorter-than-out`: warns for each visual media item whose effective source `out`
-  exceeds that source's audio-stream duration. The message includes the missing duration.
+  exceeds that source's audio-stream duration while staying inside the container duration (the
+  source plays but its audio ends early). The message includes the missing duration. Items that run
+  past the container duration are reported by `media.source-range` instead.
 - `media.caption-silence-coverage`: warns when more than 30% of caption display time overlaps a
   silence interval of at least one second. `--caption-silence-warn-percent` changes the threshold.
 - `audio.narration.trim`: warns when a narration media item's `in` is at or beyond its audio-stream
-  duration. Invalid or reversed narration `in`/`out` values are errors even without `--media`.
+  duration but still inside the container duration. Invalid or reversed narration `in`/`out` values
+  are errors even without `--media`; an `in` past the container duration is a `media.source-range`
+  error.
 
 Silence and volume checks are skipped for a source without an audio stream or when ffprobe cannot
 inspect it. Duration-dependent checks are skipped when the audio-stream duration is unavailable.
+One ffprobe invocation per source path reads both the container duration and the first audio
+stream's duration.
 Caption/silence coverage runs only when the captions can be associated with exactly one referenced
 visual source: either every caption declares the same `src`, or the timeline references a single
 visual source and captions do not contradict it. Missing, empty, or multi-source/ambiguous captions

package/vendor/packages/edit-lint/src/edit-lint.mjs

@@ -380,6 +380,7 @@ export async function lintProject(input, options = {}) {
     validateIntake(intakeState.value, findings, paths);
   }
   await lintDecisionLogPredict({ projectRoot: paths.projectRoot, intake: intakeState.value, edit: rawEdit, findings });
+  runSourceVfrChecks(rawEdit, edit, findings, paths);
 
   if (options.media) {
     runReferencedMediaChecks(
@@ -5638,10 +5639,27 @@ function runReferencedMediaChecks(rawEdit, projectedEdit, findings, skipped, pat
   const visualSourceIds = new Set();
   const visualCuts = [];
   const narrationItems = [];
+  // media.source-range はレーンを問わず（袋・グループの入れ子も含めて）全 media item を見る。
+  // 既存 3 検査の入力（referencedSourceIds / visualCuts / narrationItems）は変えない。
+  const rangeItems = [];
+  const rangeSourceIds = new Set();
 
   if (rawEdit?.version === 2 && Array.isArray(rawEdit.tracks)) {
     for (const [trackIndex, track] of rawEdit.tracks.entries()) {
       if (!isRecord(track) || !Array.isArray(track.items)) continue;
+      const collectRange = (items, pathPrefix) => {
+        for (const [itemIndex, item] of items.entries()) {
+          if (!isRecord(item)) continue;
+          const itemPath = `${pathPrefix}[${itemIndex}]`;
+          if (isRecord(item.source) && item.source.kind === "media"
+            && isNonEmptyString(item.source.src)) {
+            rangeItems.push({ item, sourceId: item.source.src, itemPath });
+            rangeSourceIds.add(item.source.src);
+          }
+          if (Array.isArray(item.items)) collectRange(item.items, `${itemPath}.items`);
+        }
+      };
+      collectRange(track.items, `edit.json#tracks[${trackIndex}].items`);
       for (const [itemIndex, item] of track.items.entries()) {
         if (!isRecord(item) || !isRecord(item.source) || item.source.kind !== "media"
           || !isNonEmptyString(item.source.src)) continue;
@@ -5661,11 +5679,11 @@ function runReferencedMediaChecks(rawEdit, projectedEdit, findings, skipped, pat
       if (!isRecord(cut) || !isNonEmptyString(cut.src)) continue;
       referencedSourceIds.add(cut.src);
       visualSourceIds.add(cut.src);
-      visualCuts.push({
-        item: { id: cut.id ?? `cut-${index + 1}`, source: cut },
-        sourceId: cut.src,
-        itemPath: `edit.json#cuts[${index}]`,
-      });
+      const itemPath = `edit.json#cuts[${index}]`;
+      const projectedItem = { id: cut.id ?? `cut-${index + 1}`, source: cut };
+      visualCuts.push({ item: projectedItem, sourceId: cut.src, itemPath });
+      rangeItems.push({ item: projectedItem, sourceId: cut.src, itemPath });
+      rangeSourceIds.add(cut.src);
     }
   }
 
@@ -5706,13 +5724,24 @@ function runReferencedMediaChecks(rawEdit, projectedEdit, findings, skipped, pat
     );
   }
 
+  // range 検査だけが参照する source（入れ子・音声レーン）も同じ path キャッシュで probe する。
+  // 既存検査（silence / volume）の対象集合は referencedSourceIds のままで変えない。
+  for (const sourceId of rangeSourceIds) {
+    if (probeBySourceId.has(sourceId)) continue;
+    const source = sourcesById.get(sourceId);
+    if (!source || !isNonEmptyString(source.path)) continue;
+    probeBySourceId.set(sourceId, probeForPath(resolveReference(paths.editPath, source.path, paths)));
+  }
+
+  const fps = projectedEdit?.output?.fps;
   validateVisualAudioDuration(
     visualCuts,
     probeBySourceId,
     findings,
     skipped,
-    projectedEdit?.output?.fps,
+    fps,
   );
+  validateMediaSourceRange(rangeItems, probeBySourceId, sourcesById, findings, skipped, fps);
   validateNarrationMediaStart(narrationItems, probeBySourceId, findings, skipped);
 
   if (Array.isArray(rawEdit?.audio?.narration)) {
@@ -5730,6 +5759,115 @@ function runReferencedMediaChecks(rawEdit, projectedEdit, findings, skipped, pat
       );
     }
   }
+
+  // v1 形の audio 宣言（path 直指定）も同じ実尺で見る。bgm は schema に out が無いので in だけ。
+  for (const key of ["narration", "sfx"]) {
+    const entries = rawEdit?.audio?.[key];
+    if (!Array.isArray(entries)) continue;
+    for (const [index, item] of entries.entries()) {
+      if (!isRecord(item) || !isNonEmptyString(item.path)) continue;
+      const probe = probeForPath(resolveReference(paths.editPath, item.path, paths));
+      if (!isPositiveNumber(probe?.containerDuration)) continue;
+      addSourceRangeFindings({
+        label: isNonEmptyString(item.id) ? item.id : `${key}-${index + 1}`,
+        duration: probe.containerDuration,
+        inSeconds: isFiniteNumber(item.in) ? item.in : 0,
+        outSeconds: isFiniteNumber(item.out) ? item.out : null,
+        pathPrefix: `edit.json#audio.${key}[${index}]`,
+        pathSuffix: "",
+        findings,
+      });
+    }
+  }
+  const declaredBgm = rawEdit?.audio?.bgm;
+  if (isRecord(declaredBgm) && isNonEmptyString(declaredBgm.path)) {
+    const probe = probeForPath(resolveReference(paths.editPath, declaredBgm.path, paths));
+    if (isPositiveNumber(probe?.containerDuration)) {
+      addSourceRangeFindings({
+        label: "bgm",
+        duration: probe.containerDuration,
+        inSeconds: isFiniteNumber(declaredBgm.in) ? declaredBgm.in : 0,
+        outSeconds: isFiniteNumber(declaredBgm.out) ? declaredBgm.out : null,
+        pathPrefix: "edit.json#audio.bgm",
+        pathSuffix: "",
+        findings,
+      });
+    }
+  }
+}
+
+function runSourceVfrChecks(rawEdit, projectedEdit, findings, paths) {
+  const declaredSources = Array.isArray(rawEdit?.sources)
+    ? rawEdit.sources
+    : Array.isArray(projectedEdit?.sources) ? projectedEdit.sources : [];
+  const sourcesById = new Map(declaredSources
+    .filter((source) => isRecord(source) && isNonEmptyString(source.id))
+    .map((source) => [source.id, source]));
+  const referencedSourceIds = new Set();
+  if (rawEdit?.version === 2 && Array.isArray(rawEdit.tracks)) {
+    for (const track of rawEdit.tracks) {
+      if (!isRecord(track) || !Array.isArray(track.items)) continue;
+      for (const item of track.items) {
+        if (isRecord(item) && isRecord(item.source) && item.source.kind === "media"
+          && isNonEmptyString(item.source.src)) {
+          referencedSourceIds.add(item.source.src);
+        }
+      }
+    }
+  } else if (Array.isArray(projectedEdit?.cuts)) {
+    for (const cut of projectedEdit.cuts) {
+      if (isRecord(cut) && isNonEmptyString(cut.src)) referencedSourceIds.add(cut.src);
+    }
+  }
+  for (const sourceId of referencedSourceIds) {
+    const source = sourcesById.get(sourceId);
+    if (source && isNonEmptyString(source.path)) {
+      addSourceVfrFinding(source, declaredSources.indexOf(source), findings, paths);
+    }
+  }
+}
+
+function addSourceVfrFinding(source, sourceIndex, findings, paths) {
+  const analysisPath = sourceAnalysisPath(paths.projectRoot, source.path);
+  if (analysisPath === null) return;
+  let timing;
+  try {
+    timing = JSON.parse(readFileSync(analysisPath, "utf8"))?.probe?.video?.frame_timing;
+  } catch {
+    return;
+  }
+  if (!isRecord(timing) || timing.mode !== "vfr") return;
+  const irregularDeltas = Number.isFinite(timing.irregular_deltas) ? timing.irregular_deltas : 0;
+  const maxDeviationMs = Number.isFinite(timing.max_deviation_ms) ? timing.max_deviation_ms : 0;
+  const cumulativeDriftMs = Number.isFinite(timing.cumulative_drift_ms)
+    ? timing.cumulative_drift_ms
+    : 0;
+  const nominalFrameMs = Number.isFinite(timing.nominal_frame_ms) ? timing.nominal_frame_ms : 0;
+  let message = `この素材は可変フレームレートです（ぶれ ${formatNumber(irregularDeltas)} 回・最大 ${formatNumber(maxDeviationMs)} ms）。最近傍で写像しています`;
+  if (nominalFrameMs > 0 && Math.abs(cumulativeDriftMs) > nominalFrameMs / 2) {
+    message += "。固定フレームレートに変換すると音ズレを防げます（任意）";
+  }
+  addFinding(findings, {
+    severity: "warning",
+    check: "source.vfr",
+    message,
+    path: `edit.json#sources[${sourceIndex}].path`,
+  });
+}
+
+function sourceAnalysisPath(projectRoot, sourcePath) {
+  if (!isNonEmptyString(sourcePath) || isAbsolute(sourcePath)) return null;
+  const resolvedSource = resolve(projectRoot, sourcePath);
+  const projectRelative = relative(projectRoot, resolvedSource);
+  if (projectRelative.startsWith("..") || isAbsolute(projectRelative)) return null;
+  const posixRelative = projectRelative.split("\\").join("/");
+  return join(
+    projectRoot,
+    ".akari",
+    "sidecars",
+    `${posixRelative}.analysis`,
+    "analysis.json",
+  );
 }
 
 function bindCaptionsToVisualSource(captionsRoot, visualSourceIds) {
@@ -5769,6 +5907,17 @@ function bindCaptionsToVisualSource(captionsRoot, visualSourceIds) {
   };
 }
 
+/** item が要求する素材終端（秒）。out が無ければフレーム尺と speed から導く。 */
+function effectiveSourceOut(item, fps) {
+  const source = item?.source;
+  const inSeconds = isFiniteNumber(source?.in) ? source.in : 0;
+  if (isFiniteNumber(source?.out)) return source.out;
+  if (Number.isInteger(item?.duration) && isPositiveNumber(fps)) {
+    return inSeconds + (item.duration / fps) * (isPositiveNumber(source?.speed) ? source.speed : 1);
+  }
+  return null;
+}
+
 function validateVisualAudioDuration(visualCuts, probeBySourceId, findings, skipped, fps) {
   const unavailableSourceIds = new Set();
   for (const { item, sourceId, itemPath } of visualCuts) {
@@ -5777,14 +5926,12 @@ function validateVisualAudioDuration(visualCuts, probeBySourceId, findings, skip
       unavailableSourceIds.add(sourceId);
       continue;
     }
-    const source = item.source;
-    const inSeconds = isFiniteNumber(source?.in) ? source.in : 0;
-    const effectiveOut = isFiniteNumber(source?.out)
-      ? source.out
-      : Number.isInteger(item.duration) && isPositiveNumber(fps)
-        ? inSeconds + (item.duration / fps) * (isPositiveNumber(source?.speed) ? source.speed : 1)
-        : null;
+    const effectiveOut = effectiveSourceOut(item, fps);
     if (!isFiniteNumber(effectiveOut) || effectiveOut <= probe.duration + EPSILON) continue;
+    // 素材そのものの実尺を超えているぶんは media.source-range が error で出す。
+    // この検査が見るのは「コンテナ内で音声ストリームだけが先に終わる」素材。
+    if (isPositiveNumber(probe.containerDuration)
+      && effectiveOut > probe.containerDuration + EPSILON) continue;
     const itemId = isNonEmptyString(item.id) ? item.id : "media item";
     addFinding(findings, {
       severity: "warning",
@@ -5802,6 +5949,61 @@ function validateVisualAudioDuration(visualCuts, probeBySourceId, findings, skip
   }
 }
 
+/**
+ * 実在しない区間の要求を素材のコンテナ実尺で弾く（issue #68）。
+ * media.audio-shorter-than-out は音声ストリーム基準のため、音声を持たない映像と
+ * audio レーンの item（bgm / narration / sfx）が無検査のまま PASS していた。
+ * レーンも入れ子も問わず、in / out を実尺と突き合わせて error にする。
+ */
+function validateMediaSourceRange(rangeItems, probeBySourceId, sourcesById, findings, skipped, fps) {
+  const unavailableSourceIds = new Set();
+  for (const { item, sourceId, itemPath } of rangeItems) {
+    const probe = probeBySourceId.get(sourceId);
+    if (!probe || !isPositiveNumber(probe.containerDuration)) {
+      if (sourcesById.has(sourceId)) unavailableSourceIds.add(sourceId);
+      continue;
+    }
+    addSourceRangeFindings({
+      label: isNonEmptyString(item.id) ? item.id : "media item",
+      duration: probe.containerDuration,
+      inSeconds: isFiniteNumber(item.source?.in) ? item.source.in : 0,
+      outSeconds: effectiveSourceOut(item, fps),
+      pathPrefix: `${itemPath}.source`,
+      pathSuffix: `[src=${sourceId}]`,
+      findings,
+    });
+  }
+  for (const sourceId of unavailableSourceIds) {
+    addSkipped(
+      skipped,
+      "media.source-range",
+      `source ${sourceId}: container duration is unavailable`,
+    );
+  }
+}
+
+function addSourceRangeFindings(
+  { label, duration, inSeconds, outSeconds, pathPrefix, pathSuffix, findings },
+) {
+  if (isFiniteNumber(inSeconds) && inSeconds >= duration - EPSILON) {
+    addFinding(findings, {
+      severity: "error",
+      check: "media.source-range",
+      message: `${label}: in=${inSeconds.toFixed(3)}s は素材の実尺 ${duration.toFixed(3)}s 以上です（その先に素材がありません）`,
+      path: `${pathPrefix}.in${pathSuffix}`,
+    });
+    return;
+  }
+  if (isFiniteNumber(outSeconds) && outSeconds > duration + EPSILON) {
+    addFinding(findings, {
+      severity: "error",
+      check: "media.source-range",
+      message: `${label}: out=${outSeconds.toFixed(3)}s が素材の実尺 ${duration.toFixed(3)}s を ${(outSeconds - duration).toFixed(3)}s 超えています（存在しない区間の要求）`,
+      path: `${pathPrefix}.out${pathSuffix}`,
+    });
+  }
+}
+
 function validateNarrationMediaStart(narrationItems, probeBySourceId, findings, skipped) {
   const unavailableSourceIds = new Set();
   for (const { item, sourceId, itemPath } of narrationItems) {
@@ -5831,6 +6033,9 @@ function validateNarrationMediaStart(narrationItems, probeBySourceId, findings,
 function addNarrationStartWarning(itemId, inSeconds, probe, path, findings) {
   if (probe?.hasAudio !== true || !isPositiveNumber(probe.duration)
     || inSeconds < probe.duration - EPSILON) return;
+  // 素材の実尺そのものを超えているぶんは media.source-range が error で出す。
+  if (isPositiveNumber(probe.containerDuration)
+    && inSeconds >= probe.containerDuration - EPSILON) return;
   addFinding(findings, {
     severity: "warning",
     check: "audio.narration.trim",
@@ -5978,6 +6183,7 @@ function probeMediaAudio(sourcePath, configuredCommand) {
   } catch (error) {
     return {
       hasAudio: null,
+      containerDuration: null,
       reason: `audio stream detection unavailable: ${messageOf(error)}`,
     };
   }
@@ -5989,7 +6195,7 @@ function probeMediaAudio(sourcePath, configuredCommand) {
       "-select_streams",
       "a:0",
       "-show_entries",
-      "stream=index,duration",
+      "format=duration:stream=index,duration",
       "-of",
       "json",
       sourcePath,
@@ -5999,6 +6205,7 @@ function probeMediaAudio(sourcePath, configuredCommand) {
   if (result.error) {
     return {
       hasAudio: null,
+      containerDuration: null,
       reason: `audio stream detection unavailable: ${messageOf(result.error)}`,
     };
   }
@@ -6006,6 +6213,7 @@ function probeMediaAudio(sourcePath, configuredCommand) {
     const detail = String(result.stderr || result.stdout || "").trim().split("\n").at(-1);
     return {
       hasAudio: null,
+      containerDuration: null,
       reason: `audio stream detection unavailable: ${detail || `ffprobe exited with status ${result.status}`}`,
     };
   }
@@ -6016,18 +6224,26 @@ function probeMediaAudio(sourcePath, configuredCommand) {
     return {
       hasAudio: null,
       duration: null,
+      containerDuration: null,
       reason: `audio stream detection unavailable: invalid ffprobe JSON (${messageOf(error)})`,
     };
   }
+  // コンテナ実尺は音声ストリームの有無に関わらず取れる（音声なし映像でも format.duration は出る）。
+  // media.source-range はこちらを基準にする。
+  const containerDurationValue = Number(parsed?.format?.duration);
+  const container = isPositiveNumber(containerDurationValue)
+    ? { containerDuration: containerDurationValue }
+    : { containerDuration: null };
   const stream = Array.isArray(parsed?.streams) ? parsed.streams[0] : undefined;
   if (!stream) {
-    return { hasAudio: false, reason: "source has no audio stream" };
+    return { hasAudio: false, reason: "source has no audio stream", ...container };
   }
   const duration = Number(stream.duration);
   return {
     hasAudio: true,
     duration: isPositiveNumber(duration) ? duration : null,
     reason: isPositiveNumber(duration) ? null : "audio stream duration is unavailable",
+    ...container,
   };
 }
 

package/vendor/skills/analyze-footage/bin/transcribe-sa.mjs

@@ -14,6 +14,15 @@ function printJson(value) {
   process.stdout.write(`${JSON.stringify(value)}\n`);
 }
 
+function writeJson(outputPath, value) {
+  const text = `${JSON.stringify(value)}\n`;
+  if (outputPath) {
+    fs.writeFileSync(outputPath, text, "utf8");
+    return;
+  }
+  process.stdout.write(text);
+}
+
 function summarize(value, fallback) {
   const text = String(value ?? "").replace(/\s+/g, " ").trim();
   return text ? text.slice(0, 500) : fallback;
@@ -27,6 +36,48 @@ function spawn(command, args) {
   }
 }
 
+function runHelper(helperBin, inputPath) {
+  const temporaryDirectory = fs.mkdtempSync(path.join(os.tmpdir(), "akari-speech-analyzer-output-"));
+  const helperOutputPath = path.join(temporaryDirectory, "stdout.json");
+  const helperFd = fs.openSync(helperOutputPath, "w");
+  let helperResult = {};
+  let helperError = null;
+  try {
+    helperResult = spawnSync(helperBin, [inputPath], { stdio: ["ignore", helperFd, "pipe"], encoding: "utf8" });
+  } catch (error) {
+    helperError = error;
+  } finally {
+    try {
+      fs.closeSync(helperFd);
+    } catch {
+      // noop
+    }
+  }
+
+  let stdoutText = "";
+  try {
+    if (!helperError) {
+      stdoutText = fs.readFileSync(helperOutputPath, "utf8");
+    }
+  } finally {
+    try {
+      fs.rmSync(temporaryDirectory, { recursive: true, force: true });
+    } catch {
+      // noop
+    }
+  }
+
+  if (helperError) {
+    return { error: helperError };
+  }
+  return {
+    status: helperResult.status,
+    stderr: helperResult.stderr,
+    stdoutText,
+    error: helperResult.error,
+  };
+}
+
 function checkAvailability() {
   if (os.platform() !== "darwin") {
     return { available: false, reason: `macOS ではありません（${os.platform()}）` };
@@ -59,7 +110,7 @@ function checkAvailability() {
 }
 
 function parseArguments(argv) {
-  const result = { check: false, input: null, helperBin: defaultHelperBin };
+  const result = { check: false, input: null, helperBin: defaultHelperBin, output: null };
   for (let index = 0; index < argv.length; index += 1) {
     const argument = argv[index];
     if (argument === "--check") {
@@ -70,6 +121,11 @@ function parseArguments(argv) {
       if (argument === "--input") result.input = path.resolve(value);
       else result.helperBin = path.resolve(value);
       index += 1;
+    } else if (argument === "--output") {
+      const value = argv[index + 1];
+      if (!value || value.startsWith("--")) throw new Error(`${argument} の値がありません`);
+      result.output = path.resolve(value);
+      index += 1;
     } else {
       throw new Error(`不明な引数です: ${argument}`);
     }
@@ -163,13 +219,15 @@ function main() {
     return;
   }
 
-  const helperResult = spawn(options.helperBin, [options.input]);
+  const helperResult = runHelper(options.helperBin, options.input);
+
   let helperJson = null;
   try {
-    helperJson = JSON.parse(String(helperResult.stdout ?? ""));
+    helperJson = JSON.parse(helperResult.stdoutText || "");
   } catch {
     // 非 JSON の stdout は採用しない。
   }
+
   if (helperResult.error || helperResult.status !== 0 || helperJson === null) {
     const detail = typeof helperJson?.error === "string"
       ? helperJson.error
@@ -183,7 +241,7 @@ function main() {
   if (normalized.droppedSegments > 0 || normalized.droppedWords > 0) {
     process.stderr.write(`SpeechAnalyzer の無効時刻を除外: segments=${normalized.droppedSegments}, words=${normalized.droppedWords}\n`);
   }
-  printJson({ backend: "speechanalyzer", available: true, segments: normalized.segments });
+  writeJson(options.output, { backend: "speechanalyzer", available: true, segments: normalized.segments });
 }
 
 main();