Derive foundational flow context category from flow reference
What does this MR do and why?
Each foundational flow picks its own name for the context category it sends to the AI Gateway. So the names do not match. Code Review sends code_review_context, and the other flows send agent_platform_<flow>_context. Someone writing a new flow config must look up the name, and nothing stops a new flow from picking yet another one.
With this MR, the flow resolvers return only the content. Rails names the category from the flow reference, always agent_platform_<flow>_context. Recommend Reviewers and Risk Classification already use that name, so nothing changes for them.
Code Review moves to agent_platform_code_review_context. Its flow configs still read code_review_context, so the flow also sends the same content under that old name with the new context_category_alias attribute. When the flow configs read the new name, we can remove the alias.
References
- Stacked on !263595 (merged)
Screenshots or screen recordings
Not applicable. There is no UI change.
How to set up and validate locally
Use the gitlab-duo/test project with Duo Agent Platform enabled for its group.
Code Review sends both categories
-
On a merge request in
gitlab-duo/test, request a review from the Duo Code Review bot. -
Make sure the bot posts its review, the same as on
master. -
Open the agent session for the review and go to its CI job. In a Rails console, look at the additional context the job got:
build = Ci::Build.find(<job_id>) context = Gitlab::Json.parse(build.variables.to_hash['DUO_WORKFLOW_ADDITIONAL_CONTEXT_CONTENT']) context.pluck('Category')The list has both
agent_platform_code_review_contextandcode_review_context, with the sameContent.
Recommend Reviewers keeps its category
- Turn on the Recommend Reviewers flow for the group, and add an approval rule with two or more eligible approvers to
gitlab-duo/test. - Create a merge request. Make sure the agent session finishes and recommends reviewers, the same as on
master. - Check the session's CI job the same way as above. The list has
agent_platform_recommend_reviewers_contextonly once.
MR acceptance checklist
Evaluate this MR against the MR acceptance checklist. It helps you analyze changes to reduce risks in quality, performance, reliability, security, and maintainability.