dsh-bili-watch 0.2.27 → 0.2.29

package/index.js

@@ -30,7 +30,9 @@ function cleanKey(s) {
 
 /** 按视频分目录：CACHE/<BV>/<cmd>.txt —— 同一个视频各命令的输出收在一处 */
 function videoDir(key) {
-  const base = cleanKey(key)
+  // T34-P0-1：模型可能把整条链接当 key 传进来，先取 BV 再建目录，别在缓存根下建乱码目录
+  const bvHit = /(BV[0-9A-Za-z]{10})/.exec(String(key || ''))
+  const base = cleanKey(bvHit ? bvHit[1] : key)
   mkdirSync(CACHE, { recursive: true })
   // 同一个 BV 已经有目录（可能被 CLI 改成「BV_标题」）就用它，别建第二个兄弟目录（2026-10-26 审查 T01/T02）
   try {
@@ -433,7 +435,11 @@ function pickFolderDialog() {
  * 缓存自动清理：先按天数删过期文件，再按总量从最旧删到阈值内。
  * 返回被删统计，供面板显示。
  */
-function autoCleanCache() {
+let lastCleanAt = 0
+
+function autoCleanCache(opts = {}) {
+  // T31-P2-6：面板每 8 秒轮询 /status，不许每次都全量 walk 缓存目录
+  if (!opts.force && Date.now() - lastCleanAt < 10 * 60 * 1000) return { removed: 0, bytes: 0, skipped: true }
   if (!existsSync(CACHE)) return { removed: 0, bytes: 0 }
   const { days, maxMb } = cacheCfg()
   const cutoff = Date.now() - days * 86400000
@@ -466,6 +472,7 @@ function autoCleanCache() {
       if (total <= limit) break
     }
   }
+  lastCleanAt = Date.now()
   return { removed, bytes }
 }
 
@@ -682,7 +689,13 @@ function parseCommandLine(raw) {
   let cred = ''
   let badOut = false
   while (tokens.length) {
-    const t = tokens.shift()
+    let t = tokens.shift()
+    // T15-4：支持 --option=value（原来整串被当位置参数静默吞掉）
+    if (t.startsWith('--') && t.includes('=')) {
+      const eq = t.indexOf('=')
+      tokens.unshift(t.slice(eq + 1))
+      t = t.slice(0, eq)
+    }
     if (t === '--step' && tokens.length) {
       step = tokens.shift()
       extra.push('--step', step)
@@ -719,10 +732,10 @@ function parseCommandLine(raw) {
       positionals.push(t)
     }
   }
-  return { command, bv: positionals[0] || '', extra, step, cred, badOut }
+  return { command, bv: positionals[0] || '', positionals, extra, step, cred, badOut }
 }
 
-async function lookup({ command, bv, extra = [], badOut = false }) {
+async function lookup({ command, bv, positionals = [], extra = [], badOut = false }) {
   if (badOut) return { ok: false, text: '--out 只接受工作区内的相对路径（不能用绝对路径、盘符、~ 或 ..）' }
   if (!command || !COMMANDS.includes(command)) {
     return { ok: false, text: `用法：/bili <${COMMANDS.join('|')}> <BV> [--step 2] [--cred 凭证路径]` }
@@ -737,7 +750,9 @@ async function lookup({ command, bv, extra = [], badOut = false }) {
         ? ['login', ...extra]
         : command === 'hot'
           ? ['hot', ...extra]
-          : [command, bv, ...extra]
+          : command === 'comments' && String(positionals[1] || '').trim()
+            ? [command, bv, String(positionals[1]).trim(), ...extra]  // T34-P0-3：评论条数是第二个位置参数，别静默丢掉
+            : [command, bv, ...extra]
   const r = await runLook(args)
   // login 只生成二维码，真正的登录成功由面板 poll 写；命令层不记 lastRun，避免「生成成功」被读成「已登录」
   if (command !== 'login') writeLastRun(command, r.code)
@@ -756,7 +771,7 @@ async function lookup({ command, bv, extra = [], badOut = false }) {
 function apply(ctx) {
   // 启动时按「天数 + 总量阈值」清一次缓存，不点按钮也不会无限涨
   try {
-    const r = autoCleanCache()
+    const r = autoCleanCache({ force: true })
     if (r.removed) ctx.logger?.info?.(`[dsh-bili-watch] 自动清理缓存 ${r.removed} 个文件 / ${(r.bytes / 1048576).toFixed(1)} MB`)
   } catch {}
 
@@ -769,8 +784,8 @@ function apply(ctx) {
       hint: '<search 关键词|hot|peek|view|sub|comments|danmaku|asr|frames|sheet|substrip|ocr|text|login|clean> [BV] [--p N] [--step 2] [--days 3] [--order pubdate|click] [--cred 凭证路径] [--fast] [--scene] [--grid 3x3] [--format srt|json|md|txt] [--out 文件（需配合 --format）]',
     },
     handler: async (invocation) => {
-      const { command, bv, extra, badOut } = parseCommandLine(invocation?.rawInput)
-      const r = await lookup({ command, bv, extra, badOut })
+      const { command, bv, positionals, extra, badOut } = parseCommandLine(invocation?.rawInput)
+      const r = await lookup({ command, bv, positionals, extra, badOut })
       return { kind: r.ok ? 'success' : 'error', text: trimForModel(r.text, command, bv) }
     },
   })
@@ -780,6 +795,7 @@ function apply(ctx) {
   try {
     if (createUserMessage && uiCfg().autoHint !== false) {
       const hinted = new Map() // message.id → true，同一条消息只提示一次
+      const HINT_CAP = 500 // T31-P2-5：常驻 Map 不能无限涨，满了整体清空（只影响重复提示）
       const hint = (msg) => {
         const { ok, text } = textOfUserMessage(msg)
         if (!ok) return null
@@ -806,6 +822,7 @@ function apply(ctx) {
             const extra = hint(msg)
             if (!extra) continue
             hinted.set(msg.id, true)
+            if (hinted.size > HINT_CAP) hinted.clear()
             fresh.push(extra)
           }
           if (!fresh.length) return decision

package/package.json

@@ -1,6 +1,6 @@
 {
     "name":  "dsh-bili-watch",
-    "version":  "0.2.27",
+    "version":  "0.2.29",
     "private":  false,
     "description":  "把 B 站视频读成文本给 DSH agent 用：元数据、评论、弹幕、字幕、本地转写、抽帧 OCR；支持 b23.tv 短链与 av 号，抽帧缓存目录可换。只读命令零依赖。",
     "type":  "module",

package/tools/look.mjs

@@ -579,7 +579,9 @@ async function cmdPeek(bv, opts = {}) {
 async function grabFrames(bv, { step = 0, outDir, ffmpeg, fresh = false, p: page = 1, scene = false } = {}) {
   const v = await viewOf(bv, page)
   const useStep = step > 0 ? step : autoStep(v.dur)
-  const dirTag = scene ? `scene-${useStep}s` : `${useStep}s`
+  // T34-P0-2：时长拿不到（0/异常）时别用 2 秒间隔——3 小时视频会抽几千帧
+  const safeStep = (v.dur > 0) ? safeStep : Math.max(safeStep, 10)
+  const dirTag = scene ? `scene-${safeStep}s` : `${safeStep}s`
   const dir = outDir || join(CACHE, 'frames', `${bv}-p${v.p}-${dirTag}`)
   // --out 指向用户目录时绝不整目录递归删：只覆盖本工具产物（下面会重下 mp4、清 jpg）
   // --fresh 的清理**必须等拿到锁之后**：锁外删会把别人刚写的 .lock 一起删掉，两个进程都以为持锁（2026-10-26 审查 T24）
@@ -588,13 +590,13 @@ async function grabFrames(bv, { step = 0, outDir, ffmpeg, fresh = false, p: page
   const jpgs = () => readdirSync(dir).filter((f) => f.endsWith('.jpg'))
   // 只有 manifest 说这批帧抽完了才算缓存有效；残帧（中途被打断）会重抽
   const m0 = readManifest(dir)
-  if (!fresh && m0?.ok && m0.step === useStep && m0.scene === scene && jpgs().length > 0) return { dir, mp4, cached: true, frames: jpgs().length }
+  if (!fresh && m0?.ok && m0.step === safeStep && m0.scene === scene && jpgs().length > 0) return { dir, mp4, cached: true, frames: jpgs().length }
   const lock = tryLock(dir)
   if (!lock) {
     for (let i = 0; i < 60; i++) {
       await sleep(1000)
       const m = readManifest(dir)
-      if (m?.ok && m.step === useStep && m.scene === scene && jpgs().length > 0) return { dir, mp4, cached: true, frames: jpgs().length }
+      if (m?.ok && m.step === safeStep && m.scene === scene && jpgs().length > 0) return { dir, mp4, cached: true, frames: jpgs().length }
     }
     throw new Error('同目录有另一个实例在抽帧，等不到结果')
   }
@@ -619,7 +621,7 @@ async function grabFrames(bv, { step = 0, outDir, ffmpeg, fresh = false, p: page
     let sceneTh = 0
     if (scene) {
       // 等间隔模式会抽多少帧，scene 模式就挑多少个变化点
-      const targetN = Math.max(3, Math.ceil(v.dur / useStep))
+      const targetN = Math.max(3, Math.ceil(v.dur / safeStep))
       for (const th of [0.3, 0.15, 0.08]) {
         const pts = sceneTimes(mp4, th, bin)
         if (pts.length >= targetN) { sceneTh = th; frameTimes = pickEvenly(pts, targetN); break }
@@ -638,12 +640,12 @@ async function grabFrames(bv, { step = 0, outDir, ffmpeg, fresh = false, p: page
     } else {
       // 默认等间隔
       const pattern = join(dir, `${bv}-%02d.jpg`)
-      execFileSync(bin, ['-y', '-threads', '2', '-i', mp4, '-vf', `fps=1/${useStep}`, '-q:v', '2', pattern], { stdio: 'ignore', timeout: 300000 })
+      execFileSync(bin, ['-y', '-threads', '2', '-i', mp4, '-vf', `fps=1/${safeStep}`, '-q:v', '2', pattern], { stdio: 'ignore', timeout: 300000 })
     }
 
     const n = jpgs().length
     if (!n) throw new Error('抽帧没产出图片（可能源流异常）')
-    writeManifest(dir, { ok: true, bv, p: v.p, step: useStep, scene, sceneTh: scene ? sceneTh : 0, frames: n })
+    writeManifest(dir, { ok: true, bv, p: v.p, step: safeStep, scene, sceneTh: scene ? sceneTh : 0, frames: n })
     return { dir, mp4, frames: n, sceneTh: scene ? sceneTh : 0 }
   } finally {
     unlock(lock)
@@ -1070,7 +1072,7 @@ async function cmdComments(bv, ps = 10, opts = {}) {
   if (format === 'json') {
     const payload = { ok: true, source: 'comments', bv, title: v.title, count: items.length, items }
     const s = JSON.stringify(payload, null, 2)
-    if (out) { writeFileSync(out, s + '\n', 'utf8'); console.error(`✓ 已写入 ${out}（${items.length} 条）`) }
+    if (out) { try { mkdirSync(dirname(out), { recursive: true }) } catch {}; writeFileSync(out, s + '\n', 'utf8'); console.error(`✓ 已写入 ${out}（${items.length} 条）`) }
     else process.stdout.write(s + '\n')
     return
   }
@@ -1133,7 +1135,7 @@ async function cmdDanmaku(bv, p = 1, opts = {}) {
     const items = all.map((x) => ({ time: x.t, mode: x.mode, size: x.size, color: x.color, text: x.text }))
     const payload = { ok: true, source: 'danmaku', bv, cid: v.cid, count: items.length, items }
     const s = JSON.stringify(payload, null, 2)
-    if (out) { writeFileSync(out, s + '\n', 'utf8'); console.error(`✓ 已写入 ${out}（${items.length} 条）`) }
+    if (out) { try { mkdirSync(dirname(out), { recursive: true }) } catch {}; writeFileSync(out, s + '\n', 'utf8'); console.error(`✓ 已写入 ${out}（${items.length} 条）`) }
     else process.stdout.write(s + '\n')
     return
   }
@@ -1376,6 +1378,18 @@ async function cmdSheet(bv, opts = {}) {
   mkdirSync(dirname(out), { recursive: true })
   execFileSync(bin, ['-y', '-threads', '2', '-i', join(tmp, '%02d.jpg'), '-vf', `scale=640:-1,tile=${cols}x${rows}`, '-frames:v', '1', out], { stdio: 'ignore', timeout: 300000 })
   console.log(`拼图: ${out}（${(statSync(out).size / 1024).toFixed(0)} KiB，共 ${picked.length} 帧）`)
+  // T42：九宫格没有时间轴，模型看图不知道「第几格是第几秒」；写一份 manifest 并打一行
+  const stepUsed = Number((/-([0-9]+)s$/.exec(r.dir) || [])[1] || opts.step || 0)
+  const cellSec = (f, i) => {
+    const m = /-([0-9]+)\.jpg$/.exec(f)
+    const n = m ? Number(m[1]) : i + 1
+    return stepUsed > 0 ? Math.max(0, (n - 1) * stepUsed) : null
+  }
+  const cells = picked.map((f, i) => { const sec = cellSec(f, i); return { cell: i + 1, file: basename(f), sec, mmss: sec == null ? '?' : fmtSec(sec) } })
+  try {
+    writeFileSync(out.replace(/\.jpg$/, '.manifest.json'), JSON.stringify({ bv, grid, step: stepUsed || null, cells }, null, 2) + '\n', 'utf8')
+  } catch {}
+  console.log('每格时间：' + cells.map((x) => `#${x.cell}=${x.mmss}`).join('  '))
   console.log('读这张图就能过一遍画面；要精确读字用 ocr <BV>。')
 }
 
@@ -1781,12 +1795,12 @@ async function cmdText(bv, credPath, opts = {}) {
   say(`[track] 字幕: miss（${s.why}）`)
 
   // ② 本地 ASR 转写（无字幕时最靠谱，零 token；--fast 跳过，直接走画面）
-  if (!opts.fast && asrReady() && v.dur > 1800) {
-    say(`[track] 转写: skip（时长 ${fmtSec(v.dur)} 超过 30 分钟，转写会超时）`)
+  if (!opts.fast && asrReady() && v.dur > 900) {
+    say(`[track] 转写: skip（时长 ${fmtSec(v.dur)} 超过 15 分钟，转写会超时）`)
   }
   if (!opts.fast && asrReady() && v.dur <= 1800) {
-    say(`[track] 转写: 开始（faster-whisper，最多等 3 分钟）…`)
-    const a = await transcribe(bv, { timeoutMs: 180_000, p: v.p })
+    say(`[track] 转写: 开始（faster-whisper，最多等 4 分钟）…`)
+    const a = await transcribe(bv, { timeoutMs: 240_000, p: v.p })
     if (a.ok) {
       // 口播类正常 4–6 字/秒；低于 2 字/秒多半是背景音乐、外语歌或纯画面，
       // 这时候转写出来的「字」只是把杂音听成了字，继续往下走 OCR 更靠谱。
@@ -2140,7 +2154,7 @@ async function transcribeLocal(wav, tag, { model = 'small', timeoutMs = 900_000
       }
     } catch {}
     if (!text && existsSync(cacheTxt)) text = readFileSync(cacheTxt, 'utf8').trim()
-    if (!text) return { ok: false, why: `转写零字（stderr 末尾：${String(r.stderr || '').slice(-160)}）` }
+    if (!text) return { ok: false, why: `转写零字（stderr 末尾：${String(r.stderr || '').split(String.fromCharCode(10)).filter(Boolean).slice(-6).join(' | ').slice(-1200)}）` }
     return { ok: true, text, chars: text.length, verify }
   } catch (e) {
     return { ok: false, why: `转写失败：${String(e?.stderr ?? e?.message ?? e).slice(-200)}` }