integrate-prs

Prompt file imported from lksnext-ai-lab/ai-core-tools (.github/prompts/integrate-prs.prompt.md). Fill in {{pr}} before use. Copyright stays with the author.

Audit the open pull requests for integration into develop{{pr}}.

This runs as a three-stage, gated pipeline:

1. Triage (@pr-triager, read-only) — audit every open PR and emit the prioritized PR Health Report.
2. Verify (@pr-verifier) — for each integrable candidate, confirm it does what its goal/linked issue asks and does not regress develop, then emit a PASS/RISK/FAIL verdict.
3. Integrate (@pr-integrator) — update from develop and squash-merge the PASS PRs behind confirmation gates.

Stage 1 — Triage (you, @pr-triager)

1. Fetch the latest refs (git fetch origin --prune) — read-only.
2. List every open PR with gh pr list --state open and the integration-relevant JSON fields (base, mergeable, mergeStateStatus, reviewDecision, isDraft, updatedAt, additions, deletions).
3. Measure drift for each PR: how far behind origin/develop its branch is, and whether the base is actually develop.
4. Emit the PR Health Report (see your agent definition for the exact format): the per-PR table + the recommended integration order (CLEAN/approved/small first; BEHIND-then-integrate next; isolate DIRTY/huge/stale — failing CI alone is not an isolation reason, the repo has known always-failing tests) + a Blockers & observations section that names policy gaps (e.g. no required reviews, oversized stale PRs).
5. Recommend and end with a one-paragraph summary.

Stay read-only: run only read gh/git queries. Never merge, push, or update branches.

Handoff

The "Verify PRs with @pr-verifier" button appears automatically — it verifies the candidates (correctness vs goal + differential targeted tests) and then hands the PASS PRs to @pr-integrator. A "skip verification" button that goes straight to @pr-integrator is also available for trivial batches, but verification is the recommended path.