Add auto-exp-designer skill and bump version to 0.16.0#48
Add auto-exp-designer skill and bump version to 0.16.0#48charliemcgrady wants to merge 1 commit into
Conversation
|
|
||
| Don't sit and poll in a loop, and don't do `sleep 60` between turns — the harness will block it and you'll lose the thread. Run one background loop that exits when the task is done. The harness will ping you when it finishes. | ||
|
|
||
| ```bash |
There was a problem hiding this comment.
I'd like to see this put into a script in an assets directory.
|
|
||
| The agent returns one artifact with three pieces: the structured design as JSON, a standalone HTML report, and a plain-language summary. The HTML is what the user actually wants to look at. Export it and open it in the browser: | ||
|
|
||
| ```bash |
| return get_api_config("auto-exp-designer")["base_url"] | ||
|
|
||
|
|
||
| auto_exp_designer = make_a2a_group( |
There was a problem hiding this comment.
There's a conversation going on in #50 (review) over whether every a2a proxy command should have a card/send-message/task interface. It feels like information hiding to me.
In @regan-huff's PR, I proposed rolling the submit/poll/post-process steps into a single command. Yes, the instructions here work, but at the cost of tokens and the risk of messing up the steps.
It would be super cool if make_a2a_group exposed the skills listed in the card. That would have to be done dynamically, but seems possible.
Intuitively, I think an agent will work better with a more explicit interface and fewer steps, but I admit we don't really know until we have some objective way of measuring.
|
See #52 for Ai2I's PF plugin, which uses A2A on the server side (also more complicated due to being streaming/interactive) |
rodneykinney
left a comment
There was a problem hiding this comment.
Enriching the CLI will be a separate effort
Summary
auto-exp-designerskill that drives the Auto Experiment Designer A2A agent throughasta-gateway.After merge
Tag the release from main: