the reason is that we have only one “kind” of match, and marking the end of the position is always correct. with arbitrary lookarounds, instead of marking the match ended here, you would have to mark the match for lookaround #7 ended here, and the match for lookaround #3 ended 5 characters ago, and the match for lookaround #12 ended 100 characters ago. i had a lot of temptation to support arbitrary lookarounds, and it really does seem like an easy fix at first glance, but the complexities arise much later in the match loop, and this is a good example of how some theory just does not hold up when you implement it. as a side note, we can actually loosen this restriction on R3 (above) which is still work in progress, but not R1, because R1 is the one that determines the start position of the match.
Coding agents are insanely smart for some tasks but lack taste and good judgement in others. They are mortally terrified of errors, often duplicate code, leave dead code behind, or fail to reuse existing working patterns. My initial approach to solving this was an ever-growing CLAUDE.md which eventually got impractically long, and many of the entries didn’t always apply universally and felt like a waste of precious context window. So I created the dev guide (docs/dev_guide/). Agents read a summary on session start and can go deeper into any specific entry when prompted to do so. In my original project the dev guide grew organically, and I plan to extend the same concept to my new projects. Here’s an example of what a dev_guide might include:
，更多细节参见体育直播
Industry Studies
patgen tool, but
，推荐阅读safew官方版本下载获取更多信息
For inquiries related to this message please contact。服务器推荐是该领域的重要参考
Житель столицы заметил в городе необычный «подснежник» и снял его на видео. Внимание на ситуацию обратил Telegram-канал «Москва М125».