A skills directory with agentic skills that adhere to the https://agentskills.io/home protocol and are installable with https://github.com/vercel-labs/skills.
Subfolders for individual skills, e.g. skills/kfp-generator, skills/kfp-debugger, etc.
Include example inputs and outputs, e.g. prompts and the pipelines they generate.
Ideally, the generator prompt will pull down a dataset, do some etl, train / test split, train a model, and eval a model, all in the context of a kfp pipeline. If you can do this with a handful of different models, it would make for a very skill and demo.
The codebase itself should be fed in as context. If the cwd is not kubeflow/pipelines, clone it locally and pull it in as context. The kubeflow/website documentation is likely helpful as well.
We should probably instruct it to put particular emphasis on the sample pipelines given their breadth and range.
https://github.com/kubeflow/pipelines-components would be another good data source to ingest once it matures.
Love this idea? Give it a 👍.
A skills directory with agentic skills that adhere to the https://agentskills.io/home protocol and are installable with https://github.com/vercel-labs/skills.
Subfolders for individual skills, e.g.
skills/kfp-generator,skills/kfp-debugger, etc.Include example inputs and outputs, e.g. prompts and the pipelines they generate.
Ideally, the generator prompt will pull down a dataset, do some etl, train / test split, train a model, and eval a model, all in the context of a kfp pipeline. If you can do this with a handful of different models, it would make for a very skill and demo.
The codebase itself should be fed in as context. If the cwd is not kubeflow/pipelines, clone it locally and pull it in as context. The kubeflow/website documentation is likely helpful as well.
We should probably instruct it to put particular emphasis on the sample pipelines given their breadth and range.
https://github.com/kubeflow/pipelines-components would be another good data source to ingest once it matures.
Love this idea? Give it a 👍.