fable-review
Compare original and translation side by side
🇺🇸
Original
English🇨🇳
Translation
ChineseFable Review
Fable 审查
Launch a Fable 5 Max 1M reviewer to review everything fully and carefully, as if it was a senior developer reviewing the work of a junior.
Default harness is bb. Read and first. Spawn a bb thread with Claude Code Fable 5 Max 1M (look up provider/model IDs — do not guess). Reuse this thread's environment so the reviewer sees the same files. Use when this thread is coordinating the review. Then and show the exact .
/nagent/bb-cli--parent-selfbb thread waitbb thread outputIf the user names another harness (Cursor Task, cmux, Codex CLI, etc.), use that instead.
Give it the necessary context, but make sure to stay neutral and unbiased. Do not nudge it towards any one specific solution. The goal here is to do great work. So be as objective and neutral as possible in writing the prompt for the subagent.
Tell him what to review, but don't be overly specific — let him find his own bugs and shortcomings. Just tell him to work extremely hard, to go deep in the review, and to surface any critical or serious issues found in the review.
And when the subagent finishes, show the user his exact response in full. Do not rewrite it. Do not update it.
Again, the goal here is to write great software. It's to build amazing software, and in order to do that you need to let the subagent do its work: tell it what to review in a broad way, be as unbiased as possible, don't influence it in any way, and tell it to output a detailed report — telling us whether the code is good and safe to be merged into production, or whether there are any serious or critical issues with it, and if so, how to fix them.
Also tell him to make the final report concise, written in plain English.
启动Fable 5 Max 1M审查工具,像资深开发者审查初级开发者工作那样,全面、细致地审查所有内容。
默认使用的工具框架是bb。请先阅读和的说明。以Claude Code Fable 5 Max 1M启动一个bb线程(请查询对应的提供商/模型ID,不要猜测)。复用当前线程的环境,确保审查工具能看到相同的文件。当当前线程负责协调审查时,使用参数。随后执行,并展示的完整结果。
/nagent/bb-cli--parent-selfbb thread waitbb thread output如果用户指定了其他工具框架(如Cursor Task、cmux、Codex CLI等),则使用该框架。
为审查工具提供必要的上下文,但需保持中立无偏见。不要引导它倾向于任何特定解决方案。我们的目标是完成高质量的工作,因此在编写子代理的提示词时,要尽可能客观中立。
告知它需要审查的内容,但不要过于具体——让它自行发现代码中的漏洞和不足。只需要求它全力以赴、深入审查,并指出审查中发现的任何关键或严重问题。
当子代理完成审查后,向用户展示它的完整原始回复,不要改写或修改。
再次强调,我们的目标是开发优秀的软件。要打造出色的软件,就需要让子代理自主完成工作:以宽泛的方式告知它审查内容,保持最大程度的无偏见,不施加任何影响,并要求它输出一份详细报告——说明代码是否优质、安全可合并到生产环境,是否存在任何严重或关键问题,以及若存在问题该如何修复。
同时要求它将最终报告写得简洁明了,使用通俗易懂的英文。