kube-agents: Skill for Claude Code

Task

Get every pre-PR review pass into a context that did not write the change, then merge what comes back into the list that becomes the pull request's Self-Review section.

This skill is plumbing. It holds no review method — review-adversarial and review-docs-drift own that — and it does not state the requirement: "Pull Request Hygiene" in AGENTS.md does, and that list wins if this file disagrees with it.

.claude/commands/pr-preflight.md is the Claude Code wrapper. Any harness can follow this file directly.

Procedure

1. Fix the diff range

One range, shared by every pass. Fetch the base first — a base you have not refreshed silently widens the range with commits that already merged, and the passes then review other people's work as though it were yours:

# The remote pointing at gke-labs/kube-agents is not reliably called `upstream`, and
# on a clone of the upstream repository rather than a fork it is `origin`. Find it.
BASE_BRANCH=main                                  # or the branch you will target
BASE_REMOTE=$(git remote | while read -r r; do
  case "$(git remote get-url "$r")" in *gke-labs/kube-agents*) echo "$r"; break;; esac
done)
# HTTPS, not SSH. The fallback fires when no remote matches — a plain clone of a fork,
# which every open pull request here comes from — and that is where a sandbox or a CI
# runner has no key. The repository is public, so HTTPS needs no credential at all.
: "${BASE_REMOTE:=https://github.com/gke-labs/kube-agents.git}"

git fetch "$BASE_REMOTE" "+refs/heads/$BASE_BRANCH:refs/kube-agents-base/$BASE_BRANCH" || {
  echo "base fetch failed — stop here rather than reviewing against a stale base" >&2
  exit 1
}
BASE=refs/kube-agents-base/$BASE_BRANCH

git diff --stat "$BASE"...HEAD     # three-dot: against the merge base, not the base tip

Three-dot keeps base-branch drift out of everyone's scope. If the fetch fails, stop and say so: there is no safe fallback, because the wrong base fails in the direction that hides findings.