weflow-cli 1.17.0 → 1.19.1

package/scripts/create_reading_notes.py

@@ -480,21 +480,21 @@ def main():
     parser.add_argument('--date', help='日期 YYYY-MM-DD（--refresh-queries 时不需要）')
     parser.add_argument('--source', default=DEFAULT_SOURCE, help='文章目录')
     parser.add_argument('--vault', default=DEFAULT_VAULT, help='Vault 目录')
-    parser.add_argument('--clean-summaries', action='store_true',
-                        help='只把已有笔记摘要段里的微信界面残留洗掉（本地，不调模型）')
+    parser.add_argument('--clean-summaries', action='store_true',
+                        help='只把已有笔记摘要段里的微信界面残留洗掉（本地，不调模型）')
     parser.add_argument('--refresh-queries', action='store_true',
                         help='只做这件事：把已有笔记里的 dataview 查询重写成当前模板（本地，不调模型）')
     args = parser.parse_args()
 
-    if args.clean_summaries:
-        # 本地清洗：动的是本地就能算出来的东西，不必把两万多篇重问一遍
-        result = clean_summaries(args.vault)
-        print('清洗摘要：%d 篇；本来就干净、没动的 %d 篇；读不出来的 %d 篇'
-              % (len(result['cleaned']), len(result['untouched']), len(result['skipped'])))
-        for name in result['skipped'][:5]:
-            print('   跳过:', name[:70])
-        return
-
+    if args.clean_summaries:
+        # 本地清洗：动的是本地就能算出来的东西，不必把两万多篇重问一遍
+        result = clean_summaries(args.vault)
+        print('清洗摘要：%d 篇；本来就干净、没动的 %d 篇；读不出来的 %d 篇'
+              % (len(result['cleaned']), len(result['untouched']), len(result['skipped'])))
+        for name in result['skipped'][:5]:
+            print('   跳过:', name[:70])
+        return
+
     if args.refresh_queries:
         result = refresh_dataview_blocks(args.vault)
         print('重写 %d 篇；没有查询块、没动的 %d 篇；读不出来的 %d 篇'

package/scripts/daily_mail.py

@@ -32,7 +32,16 @@ sys.path.insert(0, str(SCRIPT_DIR))
 from _utils import get_api_key, load_config  # noqa: E402
 
 ROOT = SCRIPT_DIR.parent
-DEFAULT_TOPICS = ('AI', '学术')
+#: **政治不进默认**：2026-10-26 实测，含政治题材的邮件被 QQ SMTP 直接拒收
+#: （`550 The mail may contain inappropriate words or content.`），去掉后同一封信
+#: 立刻发出。政治类仍然在本地阅读器里（`daily-server`），只是不走邮件。
+DEFAULT_TOPICS = ('AI', '学术', '新闻', '文学', '投资')
+#: 只有这两类生成 AI 综述 + 聚类。其余题材**照常发出**（标题 + 一句摘要），但不额外
+#: 花钱：六个主题全做综述和聚类是 12 次调用，而新闻/文学/投资/政治通常只需扫一眼标题。
+DIGEST_TOPICS = ('AI', '学术')
+#: 非重点主题最多列几条。六个主题全量渲染是 10 万字符（2026-10-26 实测 356 篇），
+#: 多数邮件客户端会截断；只留最前面这些，其余让读的人在本地阅读器里翻。
+OTHER_TOPIC_LIMIT = 12
 SMTP_CONFIG = pathlib.Path(os.path.expanduser('~/.weflow-issue-watch.json'))
 
 
@@ -232,7 +241,16 @@ def render(date: str, groups, digests, bodies=None, clusters=None) -> tuple[str,
     total = sum(len(v) for v in groups.values())
     # 主题配色：AI 靛蓝、学术 青绿。用 table + inline style —— 邮件客户端会剥离
     # `<style>` 与 class，class 选择器在 QQ / Gmail / 企业微信里表现不一致。
-    accents = {'AI': ('#4f46e5', '#eef2ff'), '学术': ('#0d9488', '#ecfdf5')}
+    # 六个主题各一套配色。同一封信里出现六种强调色不算花——它们是**分区标记**，
+    # 换主题时颜色跟着换，扫读时不用先找回标题才知道换段了。
+    accents = {
+        'AI': ('#4f46e5', '#eef2ff'),
+        '学术': ('#0d9488', '#ecfdf5'),
+        '新闻': ('#2563eb', '#eff6ff'),
+        '文学': ('#db2777', '#fdf2f8'),
+        '投资': ('#ca8a04', '#fefce8'),
+        '政治': ('#dc2626', '#fef2f2'),
+    }
 
     text_lines = [f'公众号日报 {date}', f'共 {total} 篇', '']
     blocks: list[str] = []
@@ -298,7 +316,12 @@ def render(date: str, groups, digests, bodies=None, clusters=None) -> tuple[str,
                 )
                 text_lines.append(f'-- {cname}（{len(citems)}）--')
 
-            for a in citems:
+            # **非重点主题走紧凑模式**：一行标题、不带摘要，且只列前 N 条。
+            # 六个主题全量渲染是 10 万字符（2026-10-26 实测 356 篇），多数邮件客户端
+            # 会截断，人也不会读完——而新闻/文学/投资/政治本来就是"扫一眼标题"的量。
+            compact = topic not in DIGEST_TOPICS
+            shown = citems[:OTHER_TOPIC_LIMIT] if compact else citems
+            for a in shown:
                 i = items.index(a) + 1
                 title = a.get('title') or ''
                 source = a.get('source') or ''
@@ -311,6 +334,18 @@ def render(date: str, groups, digests, bodies=None, clusters=None) -> tuple[str,
                 title_html = (f'<a href="{html.escape(url)}" style="color:#111827;text-decoration:none;">'
                               f'{html.escape(title)}</a>' if url else html.escape(title))
 
+                if compact:
+                    # 紧凑模式下不按相关度分档：它们整体就是"浏览用"的流水。
+                    summary = ''
+                    text_lines.append(f'{num}. {title} — {meta}')
+                    blocks.append(
+                        '<tr><td style="padding:6px 28px 0;">'
+                        f'<span style="font-size:11px;color:#c0c7d0;">{num}</span> '
+                        f'<span style="font-size:13px;color:#374151;">{title_html}</span>'
+                        f'<span style="font-size:11px;color:#b3bac4;"> · {html.escape(source)}</span>'
+                        '</td></tr>'
+                    )
+                    continue
                 # **颜色必须自己说明自己。** 上一版把序号徽章染成主题色来表示"高相关"，
                 # 但没有任何图例——用户看到截图的第一反应是"这边为什么有的是绿的"（2026-10-20）。
                 # 现在徽章统一成灰色，相关度改成 meta 行里一个**写着"高"的小色块**：
@@ -440,8 +475,14 @@ def main():
     # `--no-digest` 只关"每主题一段综述"；聚类是版式结构，仍要做（两件事分开开关，
     # 否则想看新版式的人一加 `--no-digest` 就退回一根长肠子）。
     api_key = get_api_key(cfg)
-    digests = {} if args.no_digest else {t: topic_digest(t, items, api_key) for t, items in groups.items() if items}
-    clusters = {t: cluster_items(t, items, api_key, want=5) for t, items in groups.items() if len(items) >= 8}
+    digests = {} if args.no_digest else {
+        t: topic_digest(t, items, api_key)
+        for t, items in groups.items() if items and t in DIGEST_TOPICS}
+    # 聚类只给重点主题做。非重点主题分 4-6 组没有收益（新闻/文学本来就是几条标题的
+    # 流水），却要为每个主题多付一次调用。
+    clusters = {t: cluster_items(t, items, api_key, want=5)
+                for t, items in groups.items()
+                if len(items) >= 8 and t in DIGEST_TOPICS}
     for t, cs in clusters.items():
         print(f'  {t} 分了 {len(cs)} 组: ' + ' / '.join(f'{n}({len(v)})' for n, v in cs))
 
@@ -450,7 +491,12 @@ def main():
     print(f'正文含 GitHub 链接: {with_repo} 篇')
 
     text_body, html_body, total = render(args.date, groups, digests, bodies, clusters)
+    # **`--label` 支持 ASCII 代号**，中文在这里映射。原因：调用方 `daily_task.ps1` 是
+    # 无 BOM 的 UTF-8，Windows PowerShell 5.1 会按 GBK 解码它，直接传中文会变成
+    # `鎴嚦涓崍` 那种乱码进到邮件主题（2026-10-26 实际发生）。
+    LABELS = {'noon': '截至中午'}
     label = (args.label or '').strip()
+    label = LABELS.get(label, label)
     suffix = f'（{label}）' if label else ''
     subject = f'公众号日报 {args.date}{suffix}：{" / ".join(t for t in topics if groups.get(t))} {total} 篇'
     if label:

package/scripts/daily_task.ps1

@@ -1,213 +1,216 @@
-# Runs a daily report pass, with everything written to logs/daily-<date>.log.
-#
-# Two modes, because a WeChat official-account day only fills up gradually:
-#
-#   daily_task.ps1 noon   -> today so far (run at 13:00; half a day of push messages)
-#   daily_task.ps1        -> yesterday, complete (run at 00:05; same as before)
-#
-# The second pass is incremental by design: `.run_state.json` makes biz_daily skip the
-# articles the noon pass already handled, so 00:05 only fetches/summarises what arrived
-# in the afternoon, while README/index.html are rebuilt from every md on disk.
-#
-# Why this wrapper exists instead of putting the command straight into the scheduled task:
-#
-#  1. The task must not depend on a build artifact. It calls the python layer
-#     (scripts/pipeline.py) directly, which is the same source the repo builds from.
-#  2. A task whose output goes nowhere answers "it failed" and nothing else. One log
-#     file per run, so "why was yesterday short?" has an answer.
-#  3. Redirecting stdout to a FILE changes python's stdout encoding (it stops being the
-#     console and becomes the locale codepage). A title containing a character outside
-#     that codepage would then raise UnicodeEncodeError *after* the day was already
-#     written, i.e. the run looks like a failure while the artifacts are fine. Hence
-#     PYTHONIOENCODING/PYTHONUTF8 below.
-#
-# Extra arguments are appended, so `daily_task.ps1 --help` is a safe smoke test (it
-# prints the pipeline's help into that day's log and exits without touching anything).
-$ErrorActionPreference = 'Stop'
-
-$root = Split-Path -Parent $PSScriptRoot
-$logDir = Join-Path $root 'logs'
-if (-not (Test-Path $logDir)) { New-Item -ItemType Directory -Path $logDir | Out-Null }
-
-# Mode is the first positional word; anything else is forwarded to pipeline.py.
-$mode = 'day'
-$rest = @($args)
-if ($args.Count -gt 0 -and ($args[0] -eq 'noon' -or $args[0] -eq 'day')) {
-    $mode = $args[0]
-    $rest = if ($args.Count -gt 1) { @($args[1..($args.Count - 1)]) } else { @() }
-}
-
-$today = (Get-Date).ToString('yyyy-MM-dd')
-$yesterday = (Get-Date).AddDays(-1).ToString('yyyy-MM-dd')
-$targetDay = if ($mode -eq 'noon') { $today } else { $yesterday }
-
-$stamp = Get-Date -Format 'yyyy-MM-dd'
-$suffix = if ($mode -eq 'noon') { '-noon' } else { '' }
-$out = Join-Path $logDir ('daily-{0}{1}.log' -f $stamp, $suffix)
-$err = Join-Path $logDir ('daily-{0}{1}.err.log' -f $stamp, $suffix)
-
-$env:PYTHONIOENCODING = 'utf-8'
-$env:PYTHONUTF8 = '1'
-
-$python = (Get-Command python.exe).Source
-$script = Join-Path $PSScriptRoot 'pipeline.py'
-$modeArgs = if ($mode -eq 'noon') { @('--date', $today) } else { @('--yesterday') }
-$allArgs = @($script) + $modeArgs + @('--prefetch-bodies') + $rest
-
-# Idempotent "make sure the target day is as complete as it can be right now", not
-# "run at 08:20".
-#
-# Why: this machine slept from 2026-10-17 02:09 to 2026-10-18 15:26 and BOTH 08:20
-# triggers were counted as missed while StartWhenAvailable=On did not start the task
-# on wake (NumberOfMissedRuns=2, no run). A schedule that only fires if the machine
-# happens to be awake at one instant is not a schedule. With the guard below the task
-# can be triggered cheaply and often (schedule + at logon): if the target is already
-# done this exits in milliseconds, otherwise it produces/extends the day.
-#
-# The day-mode guard deliberately does NOT trust index.html alone: the noon pass also
-# writes one, so "index.html exists" would make the 00:05 pass skip the afternoon and
-# leave the day half-done behind a complete-looking page. `.articles.json`'s
-# generated_at tells us when the last pass actually ran; anything during that day
-# (before 23:00) is a mid-day artifact and must be extended.
-#
-# Note the day pass runs at 00:05 **the next day**, so a finished day is one whose
-# generated_at is later than the day itself - not one stamped at 23:xx. Testing
-# `hour -ge 23` alone would have made every 00:05 run redo the whole day (observed
-# while verifying this guard: generated_at 00:20 was judged incomplete).
-function Test-DayComplete([string]$dir, [string]$targetDay) {
-    $idx = Join-Path $dir 'index.html'
-    if (-not (Test-Path $idx)) { return $false }
-    $art = Join-Path $dir '.articles.json'
-    if (-not (Test-Path $art)) { return $true }   # old artifact without json: trust it
-    try {
-        $j = Get-Content $art -Raw -Encoding UTF8 | ConvertFrom-Json
-        $gen = [datetime]::ParseExact([string]$j.generated_at, 'yyyy-MM-dd HH:mm:ss', $null)
-        $genDay = $gen.ToString('yyyy-MM-dd')
-        # finished = generated on a later day (the usual 00:05 pass), or late the same night
-        return (($genDay -gt $targetDay) -or (($genDay -eq $targetDay) -and ($gen.Hour -ge 23)))
-    } catch { return $true }
-}
-
-# Skipped when arguments are passed, so `daily_task.ps1 --help` still works as a
-# smoke test with no side effects.
-if ($rest.Count -eq 0) {
-    $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
-    $done = if ($mode -eq 'noon') { Test-Path (Join-Path $dayDir '.noon_done') } else { Test-DayComplete $dayDir $targetDay }
-    if ($done) {
-        $note = '===== {0}  [{1}] {2} already complete - skip =====' -f `
-            (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay
-        Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value $note -Encoding UTF8
-        exit 0
-    }
-}
-
-# Start-Process OVERWRITES its redirect targets, so a same-day rerun would erase the log
-# of the run that just failed - exactly the one worth keeping. Rotate any existing pair to
-# daily-<date>.run<N>[.err].log first, so every attempt stays on disk. N is the first free
-# index, which also works when several reruns happen in one day.
-if (Test-Path $out) {
-    $n = 1
-    while (Test-Path (Join-Path $logDir ('daily-{0}{1}.run{2}.log' -f $stamp, $suffix, $n))) { $n++ }
-    Move-Item -Path $out -Destination (Join-Path $logDir ('daily-{0}{1}.run{2}.log' -f $stamp, $suffix, $n))
-    if (Test-Path $err) {
-        Move-Item -Path $err -Destination (Join-Path $logDir ('daily-{0}{1}.run{2}.err.log' -f $stamp, $suffix, $n))
-    }
-}
-
-# Start-Process writes the raw bytes through: no PowerShell error-record wrapping, and
-# no shell quoting to get wrong.
-$proc = Start-Process -FilePath $python -ArgumentList $allArgs -NoNewWindow -Wait -PassThru `
-    -RedirectStandardOutput $out -RedirectStandardError $err
-
-# Fallback for the HTML stage.
-#
-# `pipeline.py` fetches, summarises, writes the day, then renders `index.html`. That last step
-# has hung twice now (2026-10-18: 11 hours with no output; 2026-10-23: the run died after the
-# README and never produced the page). Losing the page also loses the mail, because mailing
-# happens only on success - so a single slow step silently costs the whole morning report.
-#
-# The fetch and summaries are already on disk at that point (`.articles.json`), so rendering is
-# retried on its own. `$proc` is replaced by the retry's exit code, which is what lets the mail
-# and index steps below run on a day where only the renderer failed.
-if ($proc.ExitCode -ne 0) {
-    $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
-    $art = Join-Path $dayDir '.articles.json'
-    $idx = Join-Path $dayDir 'index.html'
-    $htmlScript = Join-Path $PSScriptRoot 'generate_html.py'
-    # Only retry when the data is there and the page is missing or older than the data:
-    # a page newer than `.articles.json` means rendering already happened this run.
-    $needsRender = (Test-Path $art) -and (Test-Path $htmlScript)
-    if ($needsRender -and (Test-Path $idx)) {
-        $needsRender = (Get-Item $art).LastWriteTime -gt (Get-Item $idx).LastWriteTime
-    }
-    if ($needsRender) {
-        Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
-            ('===== {0}  [{1}] {2} pipeline exit {3}, retrying HTML only =====' -f `
-                (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay, $proc.ExitCode) -Encoding UTF8
-        $retryProc = Start-Process -FilePath $python `
-            -ArgumentList @($htmlScript, '--date', $targetDay) `
-            -NoNewWindow -Wait -PassThru `
-            -RedirectStandardOutput (Join-Path $logDir ('html-retry-{0}{1}.log' -f $stamp, $suffix)) `
-            -RedirectStandardError (Join-Path $logDir ('html-retry-{0}{1}.err.log' -f $stamp, $suffix))
-        if ($retryProc.ExitCode -eq 0) {
-            $proc = $retryProc
-        }
-    }
-}
-
-if ($proc.ExitCode -eq 0 -and $rest.Count -eq 0) {
-    # Marker for the noon pass only; the day pass is judged by generated_at above.
-    if ($mode -eq 'noon') {
-        $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
-        if (Test-Path $dayDir) {
-            Set-Content -Path (Join-Path $dayDir '.noon_done') -Value (Get-Date -Format 'yyyy-MM-dd HH:mm:ss') -Encoding UTF8
-        }
-    }
-
-    # Retrieval indexes lag behind unless something refreshes them: on 2026-10-22 the
-    # semantic index was still at 10-15 (7 days stale) and the phrase index at 10-17, so
-    # `search` simply could not find anything newer. Both run here, after the day is on
-    # disk, in their own logs -- a failure is recorded and never fails the day.
-    $searchScript = Join-Path $PSScriptRoot 'semantic_search.py'
-    if ($mode -eq 'day' -and (Test-Path $searchScript)) {
-        $idxOut = Join-Path $logDir ('index-{0}.log' -f $stamp)
-        $idxErr = Join-Path $logDir ('index-{0}.err.log' -f $stamp)
-        # `update` only embeds what is not in the index yet (4,061 new records on the first
-        # run after a week). `phrase-build` is local FTS, no API, so it rebuilds in full.
-        $idxProc = Start-Process -FilePath $python `
-            -ArgumentList @($searchScript, 'update', '--article-days', '30') `
-            -NoNewWindow -Wait -PassThru -RedirectStandardOutput $idxOut -RedirectStandardError $idxErr
-        $phraseProc = Start-Process -FilePath $python `
-            -ArgumentList @($searchScript, 'phrase-build') `
-            -NoNewWindow -Wait -PassThru `
-            -RedirectStandardOutput (Join-Path $logDir ('phrase-{0}.log' -f $stamp)) `
-            -RedirectStandardError (Join-Path $logDir ('phrase-{0}.err.log' -f $stamp))
-        if ($idxProc.ExitCode -ne 0 -or $phraseProc.ExitCode -ne 0) {
-            Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
-                ('===== {0}  index update failed (update={1}, phrase={2}), see {3} =====' -f `
-                    (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $idxProc.ExitCode, $phraseProc.ExitCode, $idxErr) -Encoding UTF8
-        }
-    }
-
-    # **Both passes mail now.** The noon pass used to be silent, which read as "it never
-    # ran" - the only way to tell it had done anything was to open the log. The two emails
-    # are clearly different: the noon one is labelled 截至中午 and says the complete version
-    # follows next morning. Mail is still its own step with its own log pair, so a mail
-    # failure never fails the day nor blocks the noon marker.
-    $mailScript = Join-Path $PSScriptRoot 'daily_mail.py'
-    $mailOut = Join-Path $logDir ('daily-mail-{0}{1}.log' -f $stamp, $suffix)
-    $mailErr = Join-Path $logDir ('daily-mail-{0}{1}.err.log' -f $stamp, $suffix)
-    if (Test-Path $mailScript) {
-        $mailArgs = @($mailScript, '--date', $targetDay)
-        if ($mode -eq 'noon') { $mailArgs += @('--label', '截至中午') }
-        $mailProc = Start-Process -FilePath $python -ArgumentList $mailArgs `
-            -NoNewWindow -Wait -PassThru -RedirectStandardOutput $mailOut -RedirectStandardError $mailErr
-        if ($mailProc.ExitCode -ne 0) {
-            Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
-                ('===== {0}  [{1}] {2} mail failed (exit {3}), see {4} =====' -f `
-                    (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay, $mailProc.ExitCode, $mailErr) -Encoding UTF8
-        }
-    }
-}
-
-exit $proc.ExitCode
+# Runs a daily report pass, with everything written to logs/daily-<date>.log.
+#
+# Two modes, because a WeChat official-account day only fills up gradually:
+#
+#   daily_task.ps1 noon   -> today so far (run at 13:00; half a day of push messages)
+#   daily_task.ps1        -> yesterday, complete (run at 00:05; same as before)
+#
+# The second pass is incremental by design: `.run_state.json` makes biz_daily skip the
+# articles the noon pass already handled, so 00:05 only fetches/summarises what arrived
+# in the afternoon, while README/index.html are rebuilt from every md on disk.
+#
+# Why this wrapper exists instead of putting the command straight into the scheduled task:
+#
+#  1. The task must not depend on a build artifact. It calls the python layer
+#     (scripts/pipeline.py) directly, which is the same source the repo builds from.
+#  2. A task whose output goes nowhere answers "it failed" and nothing else. One log
+#     file per run, so "why was yesterday short?" has an answer.
+#  3. Redirecting stdout to a FILE changes python's stdout encoding (it stops being the
+#     console and becomes the locale codepage). A title containing a character outside
+#     that codepage would then raise UnicodeEncodeError *after* the day was already
+#     written, i.e. the run looks like a failure while the artifacts are fine. Hence
+#     PYTHONIOENCODING/PYTHONUTF8 below.
+#
+# Extra arguments are appended, so `daily_task.ps1 --help` is a safe smoke test (it
+# prints the pipeline's help into that day's log and exits without touching anything).
+$ErrorActionPreference = 'Stop'
+
+$root = Split-Path -Parent $PSScriptRoot
+$logDir = Join-Path $root 'logs'
+if (-not (Test-Path $logDir)) { New-Item -ItemType Directory -Path $logDir | Out-Null }
+
+# Mode is the first positional word; anything else is forwarded to pipeline.py.
+$mode = 'day'
+$rest = @($args)
+if ($args.Count -gt 0 -and ($args[0] -eq 'noon' -or $args[0] -eq 'day')) {
+    $mode = $args[0]
+    $rest = if ($args.Count -gt 1) { @($args[1..($args.Count - 1)]) } else { @() }
+}
+
+$today = (Get-Date).ToString('yyyy-MM-dd')
+$yesterday = (Get-Date).AddDays(-1).ToString('yyyy-MM-dd')
+$targetDay = if ($mode -eq 'noon') { $today } else { $yesterday }
+
+$stamp = Get-Date -Format 'yyyy-MM-dd'
+$suffix = if ($mode -eq 'noon') { '-noon' } else { '' }
+$out = Join-Path $logDir ('daily-{0}{1}.log' -f $stamp, $suffix)
+$err = Join-Path $logDir ('daily-{0}{1}.err.log' -f $stamp, $suffix)
+
+$env:PYTHONIOENCODING = 'utf-8'
+$env:PYTHONUTF8 = '1'
+
+$python = (Get-Command python.exe).Source
+$script = Join-Path $PSScriptRoot 'pipeline.py'
+$modeArgs = if ($mode -eq 'noon') { @('--date', $today) } else { @('--yesterday') }
+$allArgs = @($script) + $modeArgs + @('--prefetch-bodies') + $rest
+
+# Idempotent "make sure the target day is as complete as it can be right now", not
+# "run at 08:20".
+#
+# Why: this machine slept from 2026-10-17 02:09 to 2026-10-18 15:26 and BOTH 08:20
+# triggers were counted as missed while StartWhenAvailable=On did not start the task
+# on wake (NumberOfMissedRuns=2, no run). A schedule that only fires if the machine
+# happens to be awake at one instant is not a schedule. With the guard below the task
+# can be triggered cheaply and often (schedule + at logon): if the target is already
+# done this exits in milliseconds, otherwise it produces/extends the day.
+#
+# The day-mode guard deliberately does NOT trust index.html alone: the noon pass also
+# writes one, so "index.html exists" would make the 00:05 pass skip the afternoon and
+# leave the day half-done behind a complete-looking page. `.articles.json`'s
+# generated_at tells us when the last pass actually ran; anything during that day
+# (before 23:00) is a mid-day artifact and must be extended.
+#
+# Note the day pass runs at 00:05 **the next day**, so a finished day is one whose
+# generated_at is later than the day itself - not one stamped at 23:xx. Testing
+# `hour -ge 23` alone would have made every 00:05 run redo the whole day (observed
+# while verifying this guard: generated_at 00:20 was judged incomplete).
+function Test-DayComplete([string]$dir, [string]$targetDay) {
+    $idx = Join-Path $dir 'index.html'
+    if (-not (Test-Path $idx)) { return $false }
+    $art = Join-Path $dir '.articles.json'
+    if (-not (Test-Path $art)) { return $true }   # old artifact without json: trust it
+    try {
+        $j = Get-Content $art -Raw -Encoding UTF8 | ConvertFrom-Json
+        $gen = [datetime]::ParseExact([string]$j.generated_at, 'yyyy-MM-dd HH:mm:ss', $null)
+        $genDay = $gen.ToString('yyyy-MM-dd')
+        # finished = generated on a later day (the usual 00:05 pass), or late the same night
+        return (($genDay -gt $targetDay) -or (($genDay -eq $targetDay) -and ($gen.Hour -ge 23)))
+    } catch { return $true }
+}
+
+# Skipped when arguments are passed, so `daily_task.ps1 --help` still works as a
+# smoke test with no side effects.
+if ($rest.Count -eq 0) {
+    $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
+    $done = if ($mode -eq 'noon') { Test-Path (Join-Path $dayDir '.noon_done') } else { Test-DayComplete $dayDir $targetDay }
+    if ($done) {
+        $note = '===== {0}  [{1}] {2} already complete - skip =====' -f `
+            (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay
+        Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value $note -Encoding UTF8
+        exit 0
+    }
+}
+
+# Start-Process OVERWRITES its redirect targets, so a same-day rerun would erase the log
+# of the run that just failed - exactly the one worth keeping. Rotate any existing pair to
+# daily-<date>.run<N>[.err].log first, so every attempt stays on disk. N is the first free
+# index, which also works when several reruns happen in one day.
+if (Test-Path $out) {
+    $n = 1
+    while (Test-Path (Join-Path $logDir ('daily-{0}{1}.run{2}.log' -f $stamp, $suffix, $n))) { $n++ }
+    Move-Item -Path $out -Destination (Join-Path $logDir ('daily-{0}{1}.run{2}.log' -f $stamp, $suffix, $n))
+    if (Test-Path $err) {
+        Move-Item -Path $err -Destination (Join-Path $logDir ('daily-{0}{1}.run{2}.err.log' -f $stamp, $suffix, $n))
+    }
+}
+
+# Start-Process writes the raw bytes through: no PowerShell error-record wrapping, and
+# no shell quoting to get wrong.
+$proc = Start-Process -FilePath $python -ArgumentList $allArgs -NoNewWindow -Wait -PassThru `
+    -RedirectStandardOutput $out -RedirectStandardError $err
+
+# Fallback for the HTML stage.
+#
+# `pipeline.py` fetches, summarises, writes the day, then renders `index.html`. That last step
+# has hung twice now (2026-10-18: 11 hours with no output; 2026-10-23: the run died after the
+# README and never produced the page). Losing the page also loses the mail, because mailing
+# happens only on success - so a single slow step silently costs the whole morning report.
+#
+# The fetch and summaries are already on disk at that point (`.articles.json`), so rendering is
+# retried on its own. `$proc` is replaced by the retry's exit code, which is what lets the mail
+# and index steps below run on a day where only the renderer failed.
+if ($proc.ExitCode -ne 0) {
+    $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
+    $art = Join-Path $dayDir '.articles.json'
+    $idx = Join-Path $dayDir 'index.html'
+    $htmlScript = Join-Path $PSScriptRoot 'generate_html.py'
+    # Only retry when the data is there and the page is missing or older than the data:
+    # a page newer than `.articles.json` means rendering already happened this run.
+    $needsRender = (Test-Path $art) -and (Test-Path $htmlScript)
+    if ($needsRender -and (Test-Path $idx)) {
+        $needsRender = (Get-Item $art).LastWriteTime -gt (Get-Item $idx).LastWriteTime
+    }
+    if ($needsRender) {
+        Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
+            ('===== {0}  [{1}] {2} pipeline exit {3}, retrying HTML only =====' -f `
+                (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay, $proc.ExitCode) -Encoding UTF8
+        $retryProc = Start-Process -FilePath $python `
+            -ArgumentList @($htmlScript, '--date', $targetDay) `
+            -NoNewWindow -Wait -PassThru `
+            -RedirectStandardOutput (Join-Path $logDir ('html-retry-{0}{1}.log' -f $stamp, $suffix)) `
+            -RedirectStandardError (Join-Path $logDir ('html-retry-{0}{1}.err.log' -f $stamp, $suffix))
+        if ($retryProc.ExitCode -eq 0) {
+            $proc = $retryProc
+        }
+    }
+}
+
+if ($proc.ExitCode -eq 0 -and $rest.Count -eq 0) {
+    # Marker for the noon pass only; the day pass is judged by generated_at above.
+    if ($mode -eq 'noon') {
+        $dayDir = Join-Path $root ('output\biz-daily\' + $targetDay)
+        if (Test-Path $dayDir) {
+            Set-Content -Path (Join-Path $dayDir '.noon_done') -Value (Get-Date -Format 'yyyy-MM-dd HH:mm:ss') -Encoding UTF8
+        }
+    }
+
+    # Retrieval indexes lag behind unless something refreshes them: on 2026-10-22 the
+    # semantic index was still at 10-15 (7 days stale) and the phrase index at 10-17, so
+    # `search` simply could not find anything newer. Both run here, after the day is on
+    # disk, in their own logs -- a failure is recorded and never fails the day.
+    $searchScript = Join-Path $PSScriptRoot 'semantic_search.py'
+    if ($mode -eq 'day' -and (Test-Path $searchScript)) {
+        $idxOut = Join-Path $logDir ('index-{0}.log' -f $stamp)
+        $idxErr = Join-Path $logDir ('index-{0}.err.log' -f $stamp)
+        # `update` only embeds what is not in the index yet (4,061 new records on the first
+        # run after a week). `phrase-build` is local FTS, no API, so it rebuilds in full.
+        $idxProc = Start-Process -FilePath $python `
+            -ArgumentList @($searchScript, 'update', '--article-days', '30') `
+            -NoNewWindow -Wait -PassThru -RedirectStandardOutput $idxOut -RedirectStandardError $idxErr
+        $phraseProc = Start-Process -FilePath $python `
+            -ArgumentList @($searchScript, 'phrase-build') `
+            -NoNewWindow -Wait -PassThru `
+            -RedirectStandardOutput (Join-Path $logDir ('phrase-{0}.log' -f $stamp)) `
+            -RedirectStandardError (Join-Path $logDir ('phrase-{0}.err.log' -f $stamp))
+        if ($idxProc.ExitCode -ne 0 -or $phraseProc.ExitCode -ne 0) {
+            Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
+                ('===== {0}  index update failed (update={1}, phrase={2}), see {3} =====' -f `
+                    (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $idxProc.ExitCode, $phraseProc.ExitCode, $idxErr) -Encoding UTF8
+        }
+    }
+
+    # **Both passes mail now.** The noon pass used to be silent, which read as "it never
+    # ran" - the only way to tell it had done anything was to open the log. The two emails
+    # are clearly different: the noon one is labelled 截至中午 and says the complete version
+    # follows next morning. Mail is still its own step with its own log pair, so a mail
+    # failure never fails the day nor blocks the noon marker.
+    $mailScript = Join-Path $PSScriptRoot 'daily_mail.py'
+    $mailOut = Join-Path $logDir ('daily-mail-{0}{1}.log' -f $stamp, $suffix)
+    $mailErr = Join-Path $logDir ('daily-mail-{0}{1}.err.log' -f $stamp, $suffix)
+    if (Test-Path $mailScript) {
+        $mailArgs = @($mailScript, '--date', $targetDay)
+        # **只传 ASCII**：这个脚本是无 BOM 的 UTF-8，而 Windows PowerShell 5.1 会把
+        # 无 BOM 文件按 ANSI（本机 GBK）解码——直接写中文标签会变成 `鎴嚦涓崍`
+        # 这种乱码进到邮件主题里（2026-10-26 实测）。中文由 Python 侧映射。
+        if ($mode -eq 'noon') { $mailArgs += @('--label', 'noon') }
+        $mailProc = Start-Process -FilePath $python -ArgumentList $mailArgs `
+            -NoNewWindow -Wait -PassThru -RedirectStandardOutput $mailOut -RedirectStandardError $mailErr
+        if ($mailProc.ExitCode -ne 0) {
+            Add-Content -Path (Join-Path $logDir 'daily-task.log') -Value `
+                ('===== {0}  [{1}] {2} mail failed (exit {3}), see {4} =====' -f `
+                    (Get-Date -Format 'yyyy-MM-dd HH:mm:ss'), $mode, $targetDay, $mailProc.ExitCode, $mailErr) -Encoding UTF8
+        }
+    }
+}
+
+exit $proc.ExitCode

package/scripts/extract_todos.py

@@ -23,6 +23,7 @@ from pathlib import Path
 
 sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
 from _utils import load_config, decrypt_lock, call_deepseek, get_api_key
+from nt_common import decode_content
 
 TODOS_FILE = os.path.join(os.path.expanduser('~'), '.weflow-cli', 'todos.json')
 TZ = timezone(timedelta(hours=8))
@@ -108,7 +109,8 @@ def collect_recent_messages(conn, start_ts, end_ts, name_map, limit_per_session=
             ''', (start_ts, end_ts, limit_per_session))
 
             for ts, sender_id, content in c.fetchall():
-                if not content or not isinstance(content, str) or len(content) < 5:
+                content = decode_content(content)   # 压缩正文别再整条丢掉（issue #12）
+                if len(content) < 5:
                     continue
                 # 关键词过滤
                 content_lower = content.lower()