Update on per-patch AI code reviews in CI

[Patrick Robb]

Hi all. I have enabled per-patchseries Anthropic AI code review reports
(sent to test-report mailing list, available on patchwork) in the UNH CI
system, so you will start seeing those come in for patches in the future
which meet these conditions:

1. Applied cleanly to DPDK
2. Build test succeeded
3. Ran dpdk-test fast tests, and <= 3 tests failed
4. Patchseries does not make any modifications to AGENTS.md (protection
against malicious prompt modification in CI)

The reports look like this:
https://mails.dpdk.org/archives/test-report/2026-March/973793.html

A PASS status means the AI code review was able to run and report without
issue. A WARN status means the AI code review could not run (because it
failed the gating conditions explained above).

Aaron Conole came over to UNH yesterday and we went over some additional
features for the automated reviews which the UNH team is working on adding
now. These include:

1. When a gating condition fails and we report a WARN with no report,
indicate in the email which gating condition prevented the report from
running.
2. Instead of emailing 1 AI review report for the entire series (sent to
the last patch in the series on patchwork), parse out the AI review into
smaller chunks (one per patch) and send individual AI review emails to each
patch, including just the comments for that patch.
3. Allow for maintainers to override a WARN AI code review (i.e. force it
to actually run the code review on a series which was skipped) by sending
an override string to the mailing list as a response to the respective
patch.
4. Migrate to using the upstream AGENTS.md (to be merged into DPDK in the
future). Right now we are using our own out of tree AGENTS.md (which Aaron
Conole worked on) until DPDK has its AGENTS.md merged.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mails.dpdk.org/archives/dev/attachments/20260327/162d3ed0/attachment.htm>