fix(xiaoyuzhou): macOS BSD grep compat — replace grep -oP with perl -ne

[hymansun]

Summary

• Fixes transcribe_xiaoyuzhou.sh failing on every macOS install with grep: invalid option -- P
• Root cause: grep -oP (PCRE) is a GNU extension and is not available in BSD grep that ships with macOS
• Replaces the three affected sites with perl -ne — macOS ships perl by default, so no new dependency, and works identically on Linux

Closes #289

What changed

agent_reach/scripts/transcribe_xiaoyuzhou.sh, lines 38, 39, 114:

-AUDIO_URL=$(echo "$PAGE" | grep -oP 'https://media\.xyzcdn\.net/[^"]*\.(m4a|mp3)' | head -1)
-TITLE=$(echo "$PAGE" | grep -oP '"title":"[^"]*"' | head -1 | sed 's/"title":"//;s/"//')
+AUDIO_URL=$(echo "$PAGE" | perl -ne 'while (/(https:\/\/media\.xyzcdn\.net\/[^"]*\.(?:m4a|mp3))/g) { print "$1\n" }' | head -1)
+TITLE=$(echo "$PAGE" | perl -ne 'if (/"title":"([^"]*)"/) { print "$1\n"; last }' | head -1)

-WAIT_SEC=$(echo "$BODY" | grep -oP 'in \K[0-9]+m' | sed 's/m//' | head -1)
+WAIT_SEC=$(echo "$BODY" | perl -ne 'if (/in (\d+)m/) { print "$1\n"; exit }')

Net change: 3 insertions, 3 deletions. No behavior change on Linux; fixes the macOS path entirely.

Test plan

• Lint: shell still parses (bash -n agent_reach/scripts/transcribe_xiaoyuzhou.sh)

• End-to-end on macOS 14 (Apple Silicon, bash 3.2, BSD grep, ffmpeg 8.1, Python 3.14, agent-reach 1.4.0): full transcription of https://www.xiaoyuzhoufm.com/episode/69f2d432bb3ffa11e59cc5b0 (9:05, 8.4MB → 4MB mono mp3 → 1 chunk → Groq Whisper large-v3 → 2561 字 markdown). Sample run:

  📻 小宇宙播客转文字
  🔍 正在解析页面...
  📝 标题: 五一前夜，5A景区集体翻车背后
  🔗 音频: https://media.xyzcdn.net/.../xxx.m4a
  ⬇️  正在下载音频...
  📦 文件大小: 8.4M
  ⏱️  时长: 9分5秒
  🔄 正在转码...
  📦 转码后: 4MB
  📎 无需切片
  🎙️  正在转录 (Groq Whisper large-v3)...
     段 1/1... ✅ (2561 字)
  ✅ 完成！
  

• Linux smoke (any maintainer with a Linux box): same command should still work — perl regexes are equivalent to the prior PCRE patterns.

🤖 Generated with Claude Code