Reuse your Claude skills in automated code reviews

We run an agent code reviewer on our GitLab MRs and GitHub PRs, we even built it ourselves to explore agent-driven tooling: @weareikko/code-review.

One context that the review bot was still missing when reviewing our changes was all the skills shipped with our internal Claude Code plugins: accessibility patterns, our JS/UI toolkits, internal conventions, etc.

This was already possible with the ability for the agent to fetch remote skills with git:

review:
  variables:
    CODE_REVIEW_SKILLS: code-review,git:https://$REVIEW_BOT:$GITLAB_BOT_API_TOKEN@$CI_GITLAB_HOST/ikko/ikko-tools.git#0.6.13/plugins/dev/skills/aria-apg
  script:
    - npx @weareikko/code-review

But adding each skill with their full Git repository path could quickly become verbose and hard to maintain.

Version 0.9.4 of the @weareikko/code-review package fixes that: you can now declare a Claude plugin marketplace once and then reference it to declare the skills you want the reviewer to have.

- CODE_REVIEW_SKILLS: 'code-review,git:https://…@…/ikko-tools.git#0.6.13/plugins/dev/skills/aria-apg'
+ CODE_REVIEW_MARKETPLACES: 'ikko=https://…@…/ikko-tools.git#0.6.13'
+ CODE_REVIEW_SKILLS: 'code-review,ikko:dev/aria-apg'

The path is resolved from the repo's .claude-plugin/marketplace.json, so you don't need to know its internal structure and adding more skills from the same marketplace no longer repeats the URL.

We haven't implemented support for other marketplaces yet, as we mainly use Claude internally, but do reach out if this is something you'd find useful.