when I use /opsx-propose in opencode against a local qwen3.6 llm (no idea if it does this on other llms or not, don't have any subs to test against) it always runs the create first, then an explore on the codebase (wasting a huge amount of time) and only after this does it make the call to 'instructions', which includes the project context.
That context supplies a great deal of the information that the system spends time and tokens exploring. If the context is injected earlier in the process it would save a lot of time I think?
I wonder if context should be returned from the create command, like a 'by the way, here's some info about the project you might find useful as you create this change'
Or am I misunderstanding the workflow?
Ta
Ben
when I use /opsx-propose in opencode against a local qwen3.6 llm (no idea if it does this on other llms or not, don't have any subs to test against) it always runs the create first, then an explore on the codebase (wasting a huge amount of time) and only after this does it make the call to 'instructions', which includes the project context.
That context supplies a great deal of the information that the system spends time and tokens exploring. If the context is injected earlier in the process it would save a lot of time I think?
I wonder if context should be returned from the create command, like a 'by the way, here's some info about the project you might find useful as you create this change'
Or am I misunderstanding the workflow?
Ta
Ben