We need a registry from prompts. These can be versioned files in Git.
The equivalent card for a prompt would also be a file on a URL. Rather than agent-card.json, it is SKILL.md.
Name and description can be extracted from the front-matter.
When invoking an agent, I want to know which prompt was used.
This means that A2A needs to return this as meta-data or extension.
Similarly, I want to tell the agent which prompt/s to use. This would be meta-data or extension too.
This would make it easy to eval or A/B test prompts.
We need a registry from prompts. These can be versioned files in Git.
The equivalent card for a prompt would also be a file on a URL. Rather than agent-card.json, it is SKILL.md.
Name and description can be extracted from the front-matter.
When invoking an agent, I want to know which prompt was used.
This means that A2A needs to return this as meta-data or extension.
Similarly, I want to tell the agent which prompt/s to use. This would be meta-data or extension too.
This would make it easy to eval or A/B test prompts.