-
|
Even though coding-mode is a possibility I have seen agents using pctx in a step-by-step fashion with multiple tool invocation of very basic programs, without chaining work where it clearly would be possible. Is there a way to nudge the agent into fusing more operations and making full use of codingmode? Perhaps a concise AGENTS.md section explaining and encouraging use, or a skill file possibly. |
Beta Was this translation helpful? Give feedback.
Replies: 1 comment 1 reply
-
|
@jtkornel this is an interesting point. Could you confirm that this behavior still occurs when "output schemas" are defined on the tools? We usually find that the models do not try to batch operations if they cannot chain together actions with code. Overall, great suggestion though, to have a preferred AGENTS.md included |
Beta Was this translation helpful? Give feedback.
-
|
Not all the tools I configured provided output schemas (Output type:any for some of them), so that may explain it. After all with completely unstructured output it could make sense that the LLM gets into the loop. I can experiment with tools that have output schemas and see how that works. Thanks for the insight. |
Beta Was this translation helpful? Give feedback.
@jtkornel this is an interesting point. Could you confirm that this behavior still occurs when "output schemas" are defined on the tools? We usually find that the models do not try to batch operations if they cannot chain together actions with code.
Overall, great suggestion though, to have a preferred AGENTS.md included