4 ms·
Super cool for how compact yet still readable it is. Shouldn’t there be a tool description passed to the LLM though?
by JSR_FDED 3mo ago
Super cool for how compact yet still readable it is. Shouldn’t there be a tool description passed to the LLM though?
- tosh 3mo agoGreat question! I had a tool description earlier but 'sh' as tool name seems to be sufficient, the agent behaviour was the same. There might be performance gains if a description is added though, or worth trying different ways of telling the agent about what is available in the environment. That said, the newer models are fairly good at driving a harness to explore the environment.