refine autonomous_sop single-task enforcement + subagent input.txt length rule

This commit is contained in:
Liang Jiaqing
2026-03-13 09:51:14 +08:00
parent 3942fe2056
commit 029cd91c9a
2 changed files with 13 additions and 3 deletions

View File

@@ -6,6 +6,7 @@
- 启动:`python agentmain.py --task {task_name} [--llm_no N]`cwd=代码根其中agentmain.py位于代码根目录
- 流程:写 input.txt → 启动 → 轮询 output.txt → 读回复 → 写 reply.txt 继续 → 不写则5min自动退出
- input.txt原则写目标+约束可指定SOP名。禁写具体实现步骤——除非主agent已读过该SOP确认正确。凭印象猜的步骤会误导subagent
- **input.txt长度红线**input.txt必须精简。只写目标+约束+关键参数。大量数据给路径让subagent自己读禁止入input.txt
- output.txt首轮对话的流式输出持续append用mtime/size判断更新
- output1.txt, output2.txt...reply后各轮的流式输出递增编号同样持续append