shaun smith PRO

Hi --

1. The most reliable way to do this is by setting the skills directory - I'll do a few runs to confirm - the system prompt should specify the absolute path ({{availableSkills}} expands to this) .
2. Noted 🤗
3. gpt-5-nano is mapped to responses; I'll add the date alias but to use it directly specify responses.gpt-5-nano-2025-08-07.
4. I think this ships with that display off by default. You see so many as we run in parallel for performance. I'll make this switchable and ensure that it doesn't collide with the console output.
5. Different runs may have wildly different token usage depending on the trajectory - will take a look at making this easier to report.

Thanks for the feedback - we're planning a sweep of PR/Issues early next week, and considering incorporating a sandbox mechanism!

It can -- the only place where a model string is included is here: https://github.com/huggingface/upskill/blob/16a688fa32e03ee6859f01f3e13f23986e67006f/src/upskill/agent_cards/test_gen.md?plain=1#L5 where we use Opus' structured output to generate test cases. All else should be configurable from the --model switch.