Contributing¶
Thanks for helping. docs-agent follows the
Kubeflow contributor guide:
sign your commits for the DCO (git commit -s), open a pull request against
main, and a reviewer approves it with /lgtm and /approve.
Where to start¶
| You want to | Look at |
|---|---|
| Improve answers | Prompts in docs-agent-mcp/charts/docs-agent/files/, retrieval in docs-agent-mcp/mcp-server/ |
| Improve the chat widget | frontend/docs_scripts/chatbot.js, frontend/docs_styles/chatbot.css |
| Add or fix ingestion | docs-agent-mcp/pipelines/ |
| Deployment and charts | docs-agent-mcp/charts/, deploy/ |
| Evals | tests/eval/ |
| A first issue | Issues labelled good first issue, and issues filed from the nightly triage report |
Before you open a PR¶
pip install -r requirements-test.txt ruff==0.15.9
ruff check .
ruff format --check docs-agent-mcp/mcp-server tests
pytest -q
helm lint docs-agent-mcp/charts/docs-agent
CI runs the same checks, compiles the Kubeflow pipelines in their own environment, and renders every chart.
Opening issues¶
Use a title of the form <type>(<area>): <summary>, so the issue triage
workflow can attach source context and labels.
- Types:
bug,feat,chore,docs,test,ci,security - Areas:
mcp,pipelines,frontend,kagent,gateway,terraform,embeddings,ci,tests,docs,infra
Example: bug(mcp): search_kubeflow_docs returns Search failed on empty collection.
After your PR merges¶
Your change goes to staging automatically and is evaluated. See How a change reaches users for the full path and for how to find your eval run.
This documentation¶
The site is built with MkDocs Material
from docs/ and mkdocs.yml. Pages under Architecture, Operations and
Decisions include the READMEs that sit next to the code, so edit those files and
the site picks them up.