weflow-cli 1.13.1 → 1.14.1

package/CHANGELOG.md

@@ -17,6 +17,44 @@ purpose** - a hand-written "latest is X" is exactly the claim that goes stale he
 
 All notable user-facing changes are recorded here. This project follows [Semantic Versioning](https://semver.org/).
 
+## 1.14.1
+
+- **The daily run now keeps the retrieval indexes fresh.** On 2026-10-22 the semantic index was
+  still at 10-15 and the phrase index at 10-17, so `search` could not find anything newer than a
+  week ago - nothing had ever refreshed them. After the day's report is written, the wrapper runs
+  `semantic_search.py update` (embeds only what is missing: 4,061 new records on that first run)
+  and `semantic_search.py phrase-build` (local FTS, no API cost). Each has its own log and neither
+  can fail the day.
+- **The noon pass mails too, and says it is half a day.** It used to generate silently, which read
+  as "it never ran" - the only evidence was a log file. The noon email is titled
+  「…（截至中午）」 and states that the complete version arrives the next morning, so the two daily
+  emails cannot be mistaken for duplicates.
+- **`phrase-build` no longer fails when `VACUUM` cannot run.** `VACUUM` needs a temp file, and in a
+  restricted environment it reports `unable to open database file` *after* the index has been
+  written successfully - that turned a good rebuild into exit code 1.
+
+## 1.14.0
+
+- **A daily report you can receive by mail, cut down to the topics you actually read.**
+  `daily --email` mails one report with two sections - AI and 学术 - dropping 新闻 / 政治 / 投资 /
+  文学, and each section opens with an AI-written digest over that section's titles. Every article
+  keeps the one-line analysis `biz_daily` already produced (no extra calls), its source and time,
+  its link, and any GitHub repository mentioned in the body (14 of the 110 articles on 2026-10-18).
+  Recipient comes from `--mail-to`, then config `dailyMailTo`, then the SMTP account itself.
+- **Each topic is clustered again before it is printed.** 68 AI articles in one list is a wall;
+  an AI pass groups them into 4-6 named sections (模型与产品 / 智能体与工程 / 科研工具 …), each with a
+  coloured dot. Tags were not usable: 68 articles carried 382 distinct tags, the most frequent
+  appearing 6 times. A degenerate answer (everything in one group) is retried with a stricter
+  instruction, because it silently collapses the layout back to a single wall - observed on
+  2026-10-21, where 91 articles ended up in one group.
+- **Uniform cards, and no colour without a label.** Every article is the same card; only the badge
+  colour and summary length differ, so a section never looks like two layouts mixed together.
+  Relevance is a small tag that *says* 高 instead of a colour the reader has to guess.
+- **The scheduled run mails the report once a day.** `scripts/daily_task.ps1` calls `daily_mail.py`
+  only after the day pass produced the report - the noon pass generates but does not mail, so the
+  inbox gets one complete day instead of a half day followed by a repeat. A mail failure is logged
+  separately and does not turn a successful day into a failed one.
+
 ## 1.13.1
 
 - **Fix: a data directory containing non-ASCII characters was decoded wrong, so `check --report`

package/bin/weflow-cli.ts

@@ -1169,13 +1169,13 @@ const configurableKeys = [
   'ntDbPath', 'ntKey', 'ntSalt', 'contactDbPath', 'contactKey', 'contactSalt',
   'vaultRepo', 'aiEngine', 'aiBaseUrl', 'aiModel', 'deepseekApiKey', 'typesafeApiKey',
   'dashscopeApiKey', 'favPassphrase',
-  'judgeProvider', 'openrouterApiKey', 'openrouterJudgeModel',
+  'judgeProvider', 'openrouterApiKey', 'openrouterJudgeModel', 'judgeLocalUrl',
   'wereadApiKey',
   'assistantPrivacy', 'assistantWhitelist', 'assistantGroupWhitelist',
   'assistantGroupRequireMention', 'assistantFastRoute', 'assistantPanelUser', 'quickReplyContacts',
   'skillDirs', 'skillDisabled',
   'dailySources', 'dailySourceCategories',
-  'dailyExcludeTopics', 'dailyAiEnabled',
+  'dailyExcludeTopics', 'dailyAiEnabled', 'dailyMailTo',
   'wechatMediaDownload',
   'emoticonSeed',
 ] as const
@@ -6359,12 +6359,15 @@ program
   .option('--no-summary', '只判断不生成：不调 LLM 写摘要/标签/简报（主题与相关度仍由 Jev 判断），无需 DeepSeek key')
   .option('--prefetch-bodies [n]', '抓取之前先并发把当天正文抓进缓存（n=并发数，默认 8）：抓取阶段于是全部命中缓存、不再逐篇 sleep 8-12 秒（450 篇约 75 分钟 → 秒级）。对上游的请求次数不变，只是压进更短的时间窗')
   .option('--json', '输出机器可读的最终结果；运行日志写入 stderr')
+  .option('--email', '生成后把日报按主题（AI / 文学 / 学术）发到邮箱（每个主题附一段 AI 综述，逐篇给要点）')
+  .option('--mail-to <address>', '邮件收件人；缺省用配置 dailyMailTo，再缺省发给自己')
   .action(async (opts) => {
     const { spawn } = await import('child_process')
     const { existsSync, statSync } = await import('fs')
     const pkgRoot = resolvePackageRoot()
     const pipeline = join(pkgRoot, 'scripts', 'pipeline.py')
     const bizDaily = join(pkgRoot, 'scripts', 'biz_daily.py')
+    const dailyMail = join(pkgRoot, 'scripts', 'daily_mail.py')
 
     const now = new Date()
     const localDate = `${now.getFullYear()}-${String(now.getMonth() + 1).padStart(2, '0')}-${String(now.getDate()).padStart(2, '0')}`
@@ -6489,6 +6492,24 @@ program
     const code = await runPipeline(date)
     if (code === 0) {
       const complete = isComplete(date)
+      // 邮件只依赖 `.articles.json`（逐篇摘要已经由 biz_daily 写好），所以放在生成之后、
+      // 与"是否完整"无关：不完整时也先把已有的发出去，用户不至于什么都收不到。
+      let mail: { ok: boolean; code: number } | null = null
+      if (opts.email) {
+        const mailArgs = [dailyMail, '--date', date]
+        if (opts.mailTo) mailArgs.push('--to', String(opts.mailTo))
+        dailyLog(chalk.cyan('\n📧 正在按主题整理并发邮件...'))
+        mail = await new Promise((resolve) => {
+          const child = spawn(getPythonCommand(), mailArgs, {
+            windowsHide: true,
+            stdio: 'inherit',
+            env: pythonProcessEnv(apiKey),
+          })
+          child.on('error', () => resolve({ ok: false, code: 1 }))
+          child.on('exit', (c) => resolve({ ok: (c || 0) === 0, code: c || 0 }))
+        })
+        if (mail && !mail.ok) dailyLog(chalk.yellow('  ! 邮件没发出去（日报本身已生成，见上面的报错）'))
+      }
       if (opts.json) {
         console.log(JSON.stringify({
           success: complete,
@@ -6497,11 +6518,13 @@ program
           aiEnabled: !noAi,
           output: `output/biz-daily/${date}`,
           artifacts: artifactStatus(date),
+          mail,
         }, null, 2))
         if (!complete) process.exit(1)
         return
       }
       console.log(chalk.green(`\n✓ 日报已生成: output/biz-daily/${date}/`))
+      if (mail) console.log(chalk.green(`✓ 邮件已发送${opts.mailTo ? ` → ${opts.mailTo}` : ''}`))
       console.log(chalk.gray(`  HTML 阅读器: weflow-cli daily-server --date ${date}`))
     } else {
       if (opts.json) console.log(JSON.stringify({ success: false, code: 'DAILY_PIPELINE_FAILED', date, aiEnabled: !noAi }))

package/dist/bin/weflow-cli.js

@@ -1141,13 +1141,13 @@ const configurableKeys = [
     'ntDbPath', 'ntKey', 'ntSalt', 'contactDbPath', 'contactKey', 'contactSalt',
     'vaultRepo', 'aiEngine', 'aiBaseUrl', 'aiModel', 'deepseekApiKey', 'typesafeApiKey',
     'dashscopeApiKey', 'favPassphrase',
-    'judgeProvider', 'openrouterApiKey', 'openrouterJudgeModel',
+    'judgeProvider', 'openrouterApiKey', 'openrouterJudgeModel', 'judgeLocalUrl',
     'wereadApiKey',
     'assistantPrivacy', 'assistantWhitelist', 'assistantGroupWhitelist',
     'assistantGroupRequireMention', 'assistantFastRoute', 'assistantPanelUser', 'quickReplyContacts',
     'skillDirs', 'skillDisabled',
     'dailySources', 'dailySourceCategories',
-    'dailyExcludeTopics', 'dailyAiEnabled',
+    'dailyExcludeTopics', 'dailyAiEnabled', 'dailyMailTo',
     'wechatMediaDownload',
     'emoticonSeed',
 ];
@@ -6627,12 +6627,15 @@ program
     .option('--no-summary', '只判断不生成：不调 LLM 写摘要/标签/简报（主题与相关度仍由 Jev 判断），无需 DeepSeek key')
     .option('--prefetch-bodies [n]', '抓取之前先并发把当天正文抓进缓存（n=并发数，默认 8）：抓取阶段于是全部命中缓存、不再逐篇 sleep 8-12 秒（450 篇约 75 分钟 → 秒级）。对上游的请求次数不变，只是压进更短的时间窗')
     .option('--json', '输出机器可读的最终结果；运行日志写入 stderr')
+    .option('--email', '生成后把日报按主题（AI / 文学 / 学术）发到邮箱（每个主题附一段 AI 综述，逐篇给要点）')
+    .option('--mail-to <address>', '邮件收件人；缺省用配置 dailyMailTo，再缺省发给自己')
     .action(async (opts) => {
     const { spawn } = await import('child_process');
     const { existsSync, statSync } = await import('fs');
     const pkgRoot = resolvePackageRoot();
     const pipeline = join(pkgRoot, 'scripts', 'pipeline.py');
     const bizDaily = join(pkgRoot, 'scripts', 'biz_daily.py');
+    const dailyMail = join(pkgRoot, 'scripts', 'daily_mail.py');
     const now = new Date();
     const localDate = `${now.getFullYear()}-${String(now.getMonth() + 1).padStart(2, '0')}-${String(now.getDate()).padStart(2, '0')}`;
     const isoDate = (when) => `${when.getFullYear()}-${String(when.getMonth() + 1).padStart(2, '0')}-${String(when.getDate()).padStart(2, '0')}`;
@@ -6761,6 +6764,26 @@ program
     const code = await runPipeline(date);
     if (code === 0) {
         const complete = isComplete(date);
+        // 邮件只依赖 `.articles.json`（逐篇摘要已经由 biz_daily 写好），所以放在生成之后、
+        // 与"是否完整"无关：不完整时也先把已有的发出去，用户不至于什么都收不到。
+        let mail = null;
+        if (opts.email) {
+            const mailArgs = [dailyMail, '--date', date];
+            if (opts.mailTo)
+                mailArgs.push('--to', String(opts.mailTo));
+            dailyLog(chalk.cyan('\n📧 正在按主题整理并发邮件...'));
+            mail = await new Promise((resolve) => {
+                const child = spawn(getPythonCommand(), mailArgs, {
+                    windowsHide: true,
+                    stdio: 'inherit',
+                    env: pythonProcessEnv(apiKey),
+                });
+                child.on('error', () => resolve({ ok: false, code: 1 }));
+                child.on('exit', (c) => resolve({ ok: (c || 0) === 0, code: c || 0 }));
+            });
+            if (mail && !mail.ok)
+                dailyLog(chalk.yellow('  ! 邮件没发出去（日报本身已生成，见上面的报错）'));
+        }
         if (opts.json) {
             console.log(JSON.stringify({
                 success: complete,
@@ -6769,12 +6792,15 @@ program
                 aiEnabled: !noAi,
                 output: `output/biz-daily/${date}`,
                 artifacts: artifactStatus(date),
+                mail,
             }, null, 2));
             if (!complete)
                 process.exit(1);
             return;
         }
         console.log(chalk.green(`\n✓ 日报已生成: output/biz-daily/${date}/`));
+        if (mail)
+            console.log(chalk.green(`✓ 邮件已发送${opts.mailTo ? ` → ${opts.mailTo}` : ''}`));
         console.log(chalk.gray(`  HTML 阅读器: weflow-cli daily-server --date ${date}`));
     }
     else {