bug: review agents produce irrelevant findings for non-code content (blog posts, markdown plans) #391
Description
When a PR contains only blog posts (.mdx), content plans, and markdown files, the review agents still apply code-level review logic and produce findings that don't make sense for prose content.
Example: a PR adding a blog post about Amazon GuardDuty pricing got a "Not ready to merge" verdict with findings like "missing phased implementation checklist with exit criteria", "diagram feasibility gap", and "missing pre-publish QA tasks for MDX rendering". These are code review patterns applied to content that's just text and markdown.
The review agents should detect the content type of the changed files before reviewing. A PR that only touches .mdx, .md, .excalidraw, and image files isn't a code change and shouldn't be reviewed with the same lens as application code.
Possible approaches:
- Detect file types in the diff and adjust the review prompt accordingly
- Skip or soften design/architecture review for content-only PRs
- Add a content type classification step before the review agents run