Cool stuff. Lots to review. I am about to start global Magpie renaming (not
skills yet), so I will take a close look. I also fixed the stalled actions.
We had some extra fields in Dependabot that caused it to fail and our
actions were silently failing because they were too old already ;). But I
will look at it right after.





On Mon, May 25, 2026 at 4:26 AM Justin Mclean <[email protected]>
wrote:

> Hi,
>
> I've opened a number of PRs related to Skills over the last week or so.
> Most are currently intentionally marked as Draft. The goal is to get
> broader review and feedback on the approach before they get merged.
>
> Several of these introduce new skills, mentoring workflows, or
> intervention patterns, and I'd prefer to get feedback on design, usability,
> and direction early rather than merge and iterate later.
>
> Current Skills PRs include:
>
> - #252: feat(mentoring): add pr-management-mentor intervention eval suite;
> mark Mentoring experimental
> - #251: feat(pairing-self-review): add pre-flight self-review skill and
> eval suite
> - #250: Add spec-driven build loop and product specs
> - #229: committer-onboarding — post-vote onboarding for committers and PMC
> members
> - #228: contributor-activity-sweep skill with eval suite
> - #227: contributor-nomination skill with eval suite
>
> #228 and #229 are related. #228 is intentionally lightweight and narrow in
> scope as an initial building block rather than attempting to solve
> contributor analysis in a single large skill. Feedback on whether that
> decomposition makes sense would be particularly useful.
>
> #250 is a little different from the others. It introduces a spec-driven
> workflow that can generate new skills from written specifications. Several
> of the PRs above were created using that approach, so it also serves to
> exercise and validate the workflow itself.
>
> A simple way to test it is:
>
> Start on the #250 branch and run:
>
> ./tools/spec-loop/look.sh 1
>
> This reviews the specs and existing PRs, decides what should be
> implemented next, creates a branch, and attempts to implement it. Expect it
> to take around 10–15 minutes.
>
> The output still needs review. You'll want to inspect what it generated
> and run tests as normal, but it should give a sense of how the workflow
> behaves in practice.
>
> Separately, there are also two smaller eval/test PRs:
>
> - #268: feat(eval): add write-skill eval suite for Step 5 security
> checklist
> - #267: feat(eval): add list-steward-skills eval suite (7 cases, 2 steps)
>
> These should be relatively straightforward to review and are probably a
> good place to start if you'd like to provide feedback without digging into
> something bigger.
>
> Feedback on approach, naming, scope, assumptions, or anything that feels
> off would be very helpful before these move out of Draft.
>
> Thanks,
> Justin

Reply via email to