今日份和TRAE国内版斗智斗勇

TRAE国内版免费使用的初衷大概也包括让用户总结遇到的麻烦和解决办法吧我猜 #ai #大模型 我先用一阵，如果skill和rules确实好用，就分享出来，不好用就算了😂 #trae国内版 skill是让TRAE国内版去总结最近网上流行的先进AI的agent控制逻辑（非模型）， 然后让ChatGPT修正补充了一下做出来的，内容太多，等好用我再分享 rules先贴一下： Use an evidence-first workflow. Be careful and precise when reporting progress. Do not describe work as verified, completed, working, or build-passed unless supporting evidence is available. If a result is based only on code inspection or reasoning, label it clearly as “static only, not executed”. When uncertain, prefer explicit uncertainty over confident wording. Additional requirements: - Do not update STATUS.md, PLAN.md, or PROJECT_LOG.md before verification. - Record failures, blocked checks, and unresolved issues honestly. - Do not turn assumptions or expectations into facts. - Prefer narrow diffs over full rewrites. - Prefer manual confirmation for destructive or risky commands. - In every handoff, include: files changed, commands actually run, raw result summary, unverified items, and remaining uncertainty.