Automate ADR-backed component review guidance#8174
Conversation
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: 68a3cde5-7a50-4d4e-97d9-194c8b112e88
This reverts commit c03b86a.
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: 68a3cde5-7a50-4d4e-97d9-194c8b112e88
|
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: 68a3cde5-7a50-4d4e-97d9-194c8b112e88
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: 68a3cde5-7a50-4d4e-97d9-194c8b112e88
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com> Copilot-Session: 68a3cde5-7a50-4d4e-97d9-194c8b112e88
51876a1 to
e9ca907
Compare
|
joshblack
left a comment
There was a problem hiding this comment.
I tried running this to see what kind of stuff it would output and I was surprised that it didn't seem to honor some of the ADRs 🤔
As an example, I tried to create an accordion component locally and was surprised that it didn't follow the file conventions ADR to include a test file.
Could you do like a before/after kind of approach for what this kind of stuff would help out with compared to baseline? In just a one-off test it didn't seem to provide a better outcome but I also get how unscientific that is lol.
Happy to keep testing if that would be better, just wanted to leave a note while looking at this today 👀
|
@joshblack What prompts were you using? I had it take a similar component, and same prompt and I could see the tool calls to the ADRs which made some changes to how it generated code. In the baseline example it used CSS modules, but didn't use shared mixins for focus (and did in the other), similarly my test-branch generated VRT and ART tests but the baseline one didn't. |
Closes #
Builds on #8173's small ADR-backed review trial by replacing its hand-authored instruction with a complete, validated policy generation system. Public ADRs and maintained contributor guidance produce path-scoped Copilot Code Review instructions; sanitized internal direction is limited to outcome-only rules.
Changelog
New
Changed
Removed
Rollout strategy
Testing & Reviewing
Review this after the base trial establishes that path-scoped ADR rules produce useful CCR feedback. Compare
contributor-docs/review-guidance/with the generated instructions and consolidated skill reference. The internal policy intentionally excludes private provenance, identifiers, rationale, links, and alternatives.Run
npm run test:review-guidanceandnpm run check:review-guidance.