weflow-cli 1.11.1 → 1.13.0

package/scripts/export_chat_html.py

@@ -111,10 +111,13 @@ _FILE_ASSET_SUBDIR = 'files'
 # 文档那半的计数（图片那半的计数在写盘时现算）。放在模块级是为了不改 `write_assets` 的返回形状 ——
 # 它已经有两个调用方（markdown 导出与 `--media`），与 COVER_STATE/VOICE_STATE 是同一套写法。
 #
-# `exported` 与 `written` 是两件事：`written` 只数**这次新拷**的，`exported` 数"这条消息的文档确实在
-# assets 里"。分开是因为第二次跑同一个目录时新拷贝是 0 —— 只报 `written` 会打印成"0 document(s)"，
-# 而实际上 67 份都在，那是给自己制造假象（2026-10-13 真的这么打印过一次）。
-FILE_ASSET_STATE = {'exported': 0, 'written': 0, 'unavailable': 0}
+# **三个词各管一件事，别混**（媒体状态统一后，报告/清单/JSON 三处都按这套说法）：
+#   `written`     = 现在躺在导出目录里的份数（"可用数"，同一目录跑第二遍时它不为 0）
+#   `copied`      = **这一轮**新拷进去的份数（第二遍跑就是 0）
+#   `unavailable` = 没能落盘的份数（本地没有那份文件 / 太大 / 拷贝失败）
+# 分开是因为只报 `copied` 会把第二遍打印成"0 document(s)"，而实际上 67 份都在 —— 那是给自己制造
+# 假象（2026-10-13 真这么打印过一次）。
+FILE_ASSET_STATE = {'written': 0, 'copied': 0, 'unavailable': 0}
 
 # Off by default: full-resolution originals are 5-24MB / 3000-5700px each, and
 # downscaling ~500 of them costs minutes (PIL decode dominates). The cache's
@@ -242,17 +245,28 @@ def fetch_messages_from_shards(db_path, key_hex, salt_hex, talker, date='', pass
     # 原先的 `if not shards: shards = [db_path]` 是给"可能返回空"的那一份实现擦屁股的。
     shards = discover_message_shards(db_path)
     messages = []
+    failed = []
     for shard in shards:
         shard_conn = None
         try:
             shard_key, shard_salt = derive_database_key(shard, key_hex, salt_hex, passphrase)
             shard_conn, _ = connect(shard, shard_key, shard_salt)
             messages.extend(fetch_messages(shard_conn, talker, date))
-        except Exception:
+        except Exception as exc:
+            # **少读一个分片 = 导出里少一段聊天，而用户看到的是"导出成功"**。
+            # 必须留下分片名与原因：只有知道是哪个分片，才分得清"这个分片本来就空"、
+            # "密钥/权限不对"还是"文件损坏"。原先直接 continue，等于静默丢数据。
+            failed.append((os.path.basename(shard), '%s: %s' % (type(exc).__name__, exc)))
             continue
         finally:
             if shard_conn is not None:
                 shard_conn.close()
+    if failed:
+        print('[WARN] %d/%d 个消息分片读不了，导出内容会缺这几段（不是"没有消息"）：'
+              % (len(failed), len(shards)), file=sys.stderr)
+        for name, why in failed:
+            print('  - %s  (%s)' % (name, why), file=sys.stderr)
+        sys.stderr.flush()
     messages.sort(key=lambda row: (int(row[5] or 0), int(row[0] or 0)))
     return messages
 
@@ -2145,10 +2159,15 @@ def build_media_record(local_type, image_b64, image_map, local_id, create_time,
     if file_info is not None:
         # 文档本体（appmsg type 6）走的是另一条路：没有缩略图、也不在图片索引里，能不能导出只看
         # 本地 `msg/file/` 有没有那份文件。不并入图片那条链，否则报告会把"文件没下载"写成"没有可靠标识"。
+        #
+        # 状态词是 `found` 而**不是** `exported`：这份报告是在**写 assets 之前**生成的，它答的是
+        # "能不能在源数据里定位到"（`found` / `missing`），答不了"有没有写进导出目录"。
+        # 后者由清单的 `documents.written` 回答（`write_assets` 写的）。两件事用两个词，
+        # 免得"这份文件已导出"这种话在拷贝失败时变成假的。
         record['kind'] = 'file'
         record['bytes'] = file_info.get('bytes') or None
         if file_hit:
-            record['status'] = 'exported'
+            record['status'] = 'found'
             record['mediaKey'] = 'file'
         else:
             record['status'] = 'missing'
@@ -2609,15 +2628,15 @@ def write_assets(out_dir, file_prefix, formatted, assets_dir=''):
             'bytes': size,
         }
 
-    # `exported` 数的是**合并后清单里真实存在的文档条目**（第二遍跑同一目录时它不为 0，而新拷贝是 0）；
+    # `written` 数的是**合并后清单里真实存在的文档条目**（第二遍跑同一目录时它不为 0，而新拷贝是 0）；
     # `copied` 才是"这次新拷了几个"。两个都写出来，免得读清单的人把后者当前者。
-    file_exported = sum(1 for record in manifest.values() if record.get('kind') == 'file')
-    FILE_ASSET_STATE['exported'] = file_exported
-    FILE_ASSET_STATE['written'] = file_written
+    file_written_total = sum(1 for record in manifest.values() if record.get('kind') == 'file')
+    FILE_ASSET_STATE['written'] = file_written_total
+    FILE_ASSET_STATE['copied'] = file_written
     FILE_ASSET_STATE['unavailable'] = file_unavailable
     with open(manifest_path, 'w', encoding='utf-8') as fh:
         json.dump({'schema': 'weflow-media-manifest/v2', 'messages': manifest,
-                   'documents': {'exported': file_exported, 'copied': file_written,
+                   'documents': {'written': file_written_total, 'copied': file_written,
                                  'unavailable': file_unavailable}},
                   fh, ensure_ascii=False, indent=2)
     return os.path.basename(os.path.normpath(assets)), manifest, written + file_written
@@ -2641,6 +2660,35 @@ def _markdown_link_target(rel):
     return ''.join(out)
 
 
+def media_json_summary(media_items, media_report_path=None, manifest=None, asset_new=0, assets_dir=''):
+    """命令行最后那行 JSON 里的媒体摘要 —— **三种格式共用同一份词汇与同一份计数逻辑**。
+
+    每个词只回答一个问题，互不覆盖：
+
+    * `reportStatus`：按**能不能在源数据里定位到**分类（图片是 `embedded`/`remote-fetched`/`missing`，
+      文档是 `found`/`missing`）。这是"这份媒体在源数据里有没有"。
+    * `assets.written / copied`：**有没有写进导出目录**（written 是现在那里的总数，copied 是这一轮新写的）。
+      与上一条是两回事：源数据里找得到的，可能因为太大或拷贝失败而没写进去。
+    * `documents.unavailable`：没能落盘的份数（没找到 / 太大 / 拷贝失败）。
+
+    三处原本各拼一份 JSON（还各写了不同的键名），同一个事实就有三种说法；这里收成一份，
+    顺带把**报告文件名**一起给出来（此前只有 HTML 那条给）。不含路径、正文或密钥。
+    """
+    counts = {}
+    for item in media_items or []:
+        status = item.get('status') or 'unknown'
+        counts[status] = counts.get(status, 0) + 1
+    return {
+        'report': os.path.basename(media_report_path) if media_report_path else None,
+        'reportItems': len(media_items or []),
+        'reportStatus': counts,
+        'assets': {'items': len(manifest or {}), 'new': asset_new, 'dir': assets_dir or None},
+        'documents': {'written': FILE_ASSET_STATE['written'],
+                      'copied': FILE_ASSET_STATE['copied'],
+                      'unavailable': FILE_ASSET_STATE['unavailable']},
+    }
+
+
 def write_markdown(out_dir, file_prefix, formatted, assets_dir=''):
     """markdown 正文 + 媒体落盘（落盘那半在 `write_assets`，两者共用同一份清单）。"""
     assets_name, manifest, written = write_assets(out_dir, file_prefix, formatted, assets_dir)
@@ -2917,35 +2965,39 @@ def main():
     if args.assets_only:
         # 只抽媒体：其它格式的 `--media` 会先跑这一步，再各自消费清单。
         assets_name, manifest, asset_new = write_assets(args.out, file_prefix, formatted, args.assets_dir)
-        print(f"Assets: {len(manifest)} media item(s), {FILE_ASSET_STATE['exported']} document(s) "
-              f"({FILE_ASSET_STATE['unavailable']} unavailable, {FILE_ASSET_STATE['written']} new) "
+        # 这一步也写媒体报告：三种形态的调用方应当拿到同一份东西（状态计数 + 报告文件名），
+        # 少写这一份的话，`--media` 的调用者只能看到文档计数、看不到"哪些媒体没嵌上、为什么"。
+        media_report_path = None
+        if media_items:
+            media_report_path = os.path.join(args.out, f'{file_prefix}_media.json')
+            with open(media_report_path, 'w', encoding='utf-8') as handle:
+                json.dump(build_media_report(media_items, display_name, args.talker, len(formatted)),
+                          handle, ensure_ascii=False, indent=2)
+        print(f"Assets: {len(manifest)} media item(s), {FILE_ASSET_STATE['written']} document(s) "
+              f"({FILE_ASSET_STATE['unavailable']} unavailable, {FILE_ASSET_STATE['copied']} new) "
               f"({asset_new} new) -> {assets_name}")
         print(json.dumps({"success": True, "total": len(formatted),
                           "files": [os.path.join(args.out, assets_name)], "format": "assets",
-                          "assets": len(manifest), "assetsNew": asset_new,
-                          "documents": FILE_ASSET_STATE['exported'],
-                          "documentsNew": FILE_ASSET_STATE['written'],
-                          "documentsUnavailable": FILE_ASSET_STATE['unavailable'],
-                          "assetsDir": assets_name}, ensure_ascii=False))
+                          "media": media_json_summary(media_items, media_report_path, manifest,
+                                                      asset_new, assets_name)},
+                         ensure_ascii=False))
         conn.close()
         return
     if args.format == 'md':
         md_path, asset_total, asset_new, assets_name = write_markdown(args.out, file_prefix, formatted, args.assets_dir)
+        media_report_path = None      # 没有媒体时下面那段不跑，但 JSON 里仍要如实给出 None
         if media_items:
             media_report_path = os.path.join(args.out, f'{file_prefix}_media.json')
             with open(media_report_path, 'w', encoding='utf-8') as handle:
                 json.dump(build_media_report(media_items, display_name, args.talker, len(formatted)),
                           handle, ensure_ascii=False, indent=2)
         print(f"Markdown: {len(formatted)} message(s), {asset_total} media item(s), "
-              f"{FILE_ASSET_STATE['exported']} document(s) ({FILE_ASSET_STATE['unavailable']} unavailable, "
-              f"{FILE_ASSET_STATE['written']} new) ({asset_new} new) -> {os.path.basename(md_path)}")
+              f"{FILE_ASSET_STATE['written']} document(s) ({FILE_ASSET_STATE['unavailable']} unavailable, "
+              f"{FILE_ASSET_STATE['copied']} new) ({asset_new} new) -> {os.path.basename(md_path)}")
         print(json.dumps({"success": True, "total": len(formatted), "parts": 1,
                           "files": [md_path], "format": "md",
-                          "assets": asset_total, "assetsNew": asset_new,
-                          "documents": FILE_ASSET_STATE['exported'],
-                          "documentsNew": FILE_ASSET_STATE['written'],
-                          "documentsUnavailable": FILE_ASSET_STATE['unavailable'],
-                          "assetsDir": assets_name},
+                          "media": media_json_summary(media_items, media_report_path, manifest=None,
+                                                      asset_new=asset_new, assets_dir=assets_name)},
                          ensure_ascii=False))
         conn.close()
         return
@@ -2995,17 +3047,13 @@ def main():
     # Print JSON summary for CLI integration.
     # The four original keys are untouched; `media` is purely additive and the
     # one existing consumer (src/services/exportService.ts) ignores it.
+    # 三种格式（html/md/assets）用**同一个** media_json_summary，词汇与计数不再各说各话。
     print(json.dumps({
         "success": True,
         "total": total,
         "parts": len(html_files),
         "files": html_files,
-        "media": ({
-            "schema": "weflow-media-report/v1",
-            "records": len(media_items),
-            "status": media_report['counts'],
-            "report": os.path.basename(media_report_path) if media_report_path else None,
-        } if media_report else None),
+        "media": media_json_summary(media_items, media_report_path),
     }))
 
 

package/scripts/install_tasks.cmd

@@ -0,0 +1,32 @@
+@echo off
+chcp 65001 >nul
+setlocal
+rem One-click installer for the two daily-report scheduled tasks.
+rem
+rem Why this .cmd exists: agent sandboxes and restricted shells refuse Task Scheduler writes
+rem as a class (schtasks /create, the Schedule.Service COM API, and even the in-memory
+rem New-ScheduledTask* builders all return Access is denied while reads keep working).
+rem A process started from your own desktop is outside that boundary, so this asks for
+rem elevation itself and runs install_tasks.ps1 there.
+
+set "PS1=%~dp0install_tasks.ps1"
+if not exist "%PS1%" (
+  echo [ERROR] install_tasks.ps1 not found next to this file.
+  pause
+  exit /b 1
+)
+
+echo Requesting administrator privileges...
+powershell -NoProfile -ExecutionPolicy Bypass -Command "try { Start-Process -FilePath 'powershell.exe' -ArgumentList @('-NoProfile','-ExecutionPolicy','Bypass','-NoExit','-File','\"%PS1%\"') -Verb RunAs -ErrorAction Stop; exit 0 } catch { Write-Host ('Could not elevate: ' + $_.Exception.Message) -ForegroundColor Red; exit 1 }"
+if errorlevel 1 (
+  echo.
+  echo If the elevation prompt was cancelled: right-click this file and pick "Run as administrator".
+  pause
+  exit /b 1
+)
+
+echo.
+echo A new window opened and is registering the tasks. It stays open so you can read the result.
+echo To remove them later:  install_tasks.ps1 -Uninstall
+echo.
+pause

package/scripts/install_tasks.ps1

@@ -0,0 +1,91 @@
+# Register / update the two scheduled tasks for the official-account daily report.
+# Idempotent: running it again overwrites the definitions with the same values.
+#
+# Why this cannot always be done for you: writing to Task Scheduler is refused in a
+# restricted shell / agent sandbox, and refused as a *class* - measured on 2026-10-19 with
+# an elevated Administrator token, all three write paths said "Access is denied" while
+# reads kept working:
+#     schtasks /create ...                      -> Access is denied
+#     (New-Object -ComObject Schedule.Service)   -> RegisterTaskDefinition 0x80070005
+#     New-ScheduledTaskAction / -SettingsSet ... -> 拒绝访问  (even the in-memory builders)
+# So "I can see the tasks but cannot create them" is not an illusion, and it is not fixed
+# by "run as administrator". The one reliable route is a terminal the sandbox does not
+# supervise: double-click scripts\install_tasks.cmd (it asks for elevation itself), or run
+# this file from your own PowerShell window.
+#
+# Usage:
+#   powershell -NoProfile -ExecutionPolicy Bypass -File <repo>\scripts\install_tasks.ps1
+#   powershell -NoProfile -ExecutionPolicy Bypass -File <repo>\scripts\install_tasks.ps1 -Uninstall
+#
+# The two tasks:
+#   WeFlow Daily       00:05 + at logon   -> YESTERDAY (the last complete day)
+#   WeFlow Daily Noon  13:00              -> TODAY so far (half a day of pushes)
+#
+# Why 00:05 the next day instead of 24:00 tonight: it is the same instant, but "yesterday"
+# then lines up with the date directory the run writes into, and it does not collide with
+# other jobs on the hour. Official accounts keep pushing until 24:00, so this is the only
+# moment that has the complete day.
+#
+# This file is intentionally ASCII-only: Windows PowerShell 5.1 reads a BOM-less file as
+# the ANSI codepage, so non-ASCII text here turns into mojibake *and* a parse error.
+param(
+    [switch]$Uninstall
+)
+
+$ErrorActionPreference = 'Stop'
+
+$repo = Split-Path -Parent $PSScriptRoot
+$script = Join-Path $repo 'scripts\daily_task.ps1'
+if (-not (Test-Path $script)) { throw "not found: $script (this script belongs in the repo's scripts\ folder)" }
+
+$isAdmin = (New-Object Security.Principal.WindowsPrincipal(
+    [Security.Principal.WindowsIdentity]::GetCurrent())
+).IsInRole([Security.Principal.WindowsBuiltInRole]::Administrator)
+if (-not $isAdmin) {
+    Write-Host 'Need elevation: reopen PowerShell as administrator, then run this again.' -ForegroundColor Yellow
+    exit 1
+}
+
+if ($Uninstall) {
+    foreach ($name in @('WeFlow Daily Noon', 'WeFlow Daily')) {
+        if (Get-ScheduledTask -TaskName $name -ErrorAction SilentlyContinue) {
+            Unregister-ScheduledTask -TaskName $name -Confirm:$false
+            Write-Host "removed: $name"
+        }
+    }
+    exit 0
+}
+
+$settings = New-ScheduledTaskSettingsSet -StartWhenAvailable -MultipleInstances IgnoreNew
+$psExe = 'powershell.exe'
+$common = '-NoProfile -ExecutionPolicy Bypass -File ' + $script
+
+# End of day: yesterday. The at-logon trigger is the catch-up: this machine slept for
+# 37 hours and StartWhenAvailable did NOT actually start the missed run (NumberOfMissedRuns=2,
+# no run record), so "it will catch up by itself" is not something to rely on.
+Register-ScheduledTask -TaskName 'WeFlow Daily' `
+    -Action (New-ScheduledTaskAction -Execute $psExe -Argument $common) `
+    -Trigger @((New-ScheduledTaskTrigger -Daily -At '00:05'),
+               (New-ScheduledTaskTrigger -AtLogOn)) `
+    -Settings $settings `
+    -Description 'WeChat daily report for yesterday (00:05 + at logon, idempotent)' -Force | Out-Null
+
+# Noon: today so far. Deliberately NO at-logon trigger: a morning logon would produce a
+# very thin "noon" report and write the .noon_done marker, which would then make the real
+# 13:00 run skip itself (the guard keys on that marker).
+Register-ScheduledTask -TaskName 'WeFlow Daily Noon' `
+    -Action (New-ScheduledTaskAction -Execute $psExe -Argument ($common + ' noon')) `
+    -Trigger (New-ScheduledTaskTrigger -Daily -At '13:00') `
+    -Settings $settings `
+    -Description 'WeChat daily report for today so far (13:00)' -Force | Out-Null
+
+foreach ($name in @('WeFlow Daily', 'WeFlow Daily Noon')) {
+    $t = Get-ScheduledTask -TaskName $name
+    $i = Get-ScheduledTaskInfo -TaskName $name
+    Write-Host ("{0,-20} {1,-8} next {2}" -f $t.TaskName, $t.State, $i.NextRunTime) -ForegroundColor Green
+}
+Write-Host ''
+Write-Host 'Run one now instead of waiting:'
+Write-Host '  Start-ScheduledTask -TaskName "WeFlow Daily Noon"   # today so far'
+Write-Host '  Start-ScheduledTask -TaskName "WeFlow Daily"        # yesterday, complete'
+Write-Host 'Logs: <repo>\logs\daily-<date>[-noon].log'