Archived
feat(loop-engine): 配方执行器 —— project / prune / say,每一笔之前刷读数
三条硬规矩:每一笔之前刷读数(后一笔要数前一笔刚写下的行,不刷新就恒为 0); 值没变的写静默跳过(配方不是模型,原地打转不必被看见); 出错交人不重试不退回模型(配方抛错是包的 bug,退回模型就是让它去凑)。 主键算不出来的那一行先不写 —— 空主键写下去是一行永远对不上的垃圾。 Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_01LKMZDwMbyKzjw6Hembwvet
This commit is contained in:
@@ -139,6 +139,7 @@ function harness(
|
||||
} as never;
|
||||
},
|
||||
} as never,
|
||||
analyze: { run: async () => ({ object: {} }) } as never,
|
||||
tools: {
|
||||
invoke: async (input: { tool: string }) => {
|
||||
toolCalls.push(input.tool);
|
||||
|
||||
Reference in New Issue
Block a user