Lets add some skills for creating plans. The current skill is "ok" but I suspect adding a sort of ralph/iterative approach would really help to improve quality. Maybe using some sort of staged approach similar to what HumanLayer are doing? Can subagents ask user's questions? Might be a way of getting fresh context.
Probably can't integrate directly into Yarr without falling foul of Anthropic's no automation ToS.
Lets add some skills for creating plans. The current skill is "ok" but I suspect adding a sort of ralph/iterative approach would really help to improve quality. Maybe using some sort of staged approach similar to what HumanLayer are doing? Can subagents ask user's questions? Might be a way of getting fresh context.
Probably can't integrate directly into Yarr without falling foul of Anthropic's no automation ToS.