Skip to content
101 changes: 94 additions & 7 deletions plugins/modelopt/skills/release-cherry-pick/SKILL.md
Original file line number Diff line number Diff line change
@@ -1,6 +1,6 @@
---
name: release-cherry-pick
description: Cherry-pick merged PRs labeled for a release branch into that branch, then open a PR and apply the cherry-pick-done label. Use when asked to "cherry-pick PRs for release/X.Y.Z", "pick PRs to release branch", or "cherry-pick labeled PRs".
description: Audit merged bug-fix PRs and release NVBugs for missing cherry-pick labels, then cherry-pick labeled PRs into a release branch and open a PR. Use when asked to "cherry-pick PRs for release/X.Y.Z", "pick PRs to release branch", "verify cherry-pick labels", or "cherry-pick labeled PRs".
---

# Cherry-pick PRs to a Release Branch
Expand All @@ -13,7 +13,94 @@ Ask the user for the release version (e.g. `0.44.0`) if not already provided.

Set `VERSION=<version>` for use in subsequent steps.

## Step 2 — Fetch pending PRs
## Step 2 — Audit candidates for missing labels

Before fetching the labeled queue, audit release NVBugs and recent merged PRs so bug fixes are not omitted.

### Audit release NVBugs

Use the NVBugs MCP to search for every NVBug whose **Keywords** field contains the exact keyword `Committed_ModelOpt_<VERSION>`. Follow pagination until no next token is returned.

Fetch each matching NVBug with its comments. Extract every Model-Optimizer PR link or unambiguous `PR #<NUM>` reference from the comments, not only the latest comment. Keep only PRs that are merged into `main`, lack `cherry-pick-<VERSION>`, and are not already included in the release branch. Verify each remaining PR is a bug fix. The NVBug keyword is evidence for review, not by itself proof that every linked PR should be picked.

### Audit recent merged PRs

Audit PRs merged into `main` since the latest release candidate. Exclude PRs that already have `cherry-pick-<VERSION>` and changes already present on the release branch:

```bash
git fetch origin main "release/$VERSION" --tags
RC_TAG=$(git tag --merged "origin/release/$VERSION" \
--list "${VERSION}rc*" --sort=-version:refname | head -1)
test -n "$RC_TAG" || {
echo "No ${VERSION}rc* tag found on origin/release/$VERSION" >&2
exit 1
}
SINCE=$(git for-each-ref --format='%(creatordate:iso-strict)' \
"refs/tags/$RC_TAG")

PATCH_STATUS=$(git cherry "origin/release/$VERSION" origin/main)

SEARCH_RESULTS=$(
gh search prs \
--repo NVIDIA/Model-Optimizer \
--merged \
--base main \
--merged-at ">=$SINCE" \
--limit 1000 \
--json number,title,author,labels,url \
-- "-label:cherry-pick-$VERSION"
)

if test "$(jq 'length' <<<"$SEARCH_RESULTS")" -ge 1000; then
echo "Recent-PR audit reached the 1,000-result limit" >&2
exit 1
fi

jq -r '.[].number' <<<"$SEARCH_RESULTS" \
| while read -r pr; do
sha=$(gh pr view "$pr" --repo NVIDIA/Model-Optimizer \
--json mergeCommit --jq '.mergeCommit.oid')
if grep -q "^+ $sha$" <<<"$PATCH_STATUS"; then
gh pr view "$pr" --repo NVIDIA/Model-Optimizer \
--json number,title,author,labels,url
fi
done
```

A release is not normally expected to reach the 1,000-PR limit.

Review each PR's title, body, labels, changed files, and linked issue context. Classify it as:

- **Yes** — repairs incorrect behavior, a regression, crash, compatibility problem, or documentation defect relevant to the release.
- **No** — feature, refactor, cleanup, dependency refresh, or other change not needed to correct the release.
- **Unclear** — insufficient evidence or meaningful backport risk; ask the user.

Deduplicate PRs found through both audits.

### Report and label

Present the complete audit before changing GitHub labels:

| PR | Title | Author | NVBug(s) | Bug fix? | `cherry-pick-<VERSION>` present? | Recommendation |
|---|---|---|---|---|---|---|

Use these recommendation values and sort the table in this order:

1. **Needs label** — confirmed release-relevant bug fix.
2. **Unknown** — requires user judgment.
3. **No action needed** — not a release-relevant bug fix.

Use `—` when no NVBug is known. Also list NVBugs with no linked PR. Ask the user to confirm which recommended PRs should receive the missing label. After confirmation, apply it:

```bash
for pr in <APPROVED_NUMBERS>; do
gh pr edit "$pr" --repo NVIDIA/Model-Optimizer --add-label "cherry-pick-$VERSION"
done
```

Do not label unmerged PRs, PRs not based on `main`, or candidates classified **Unclear** without explicit approval. Re-run the audit table after edits so it reflects the final label state.

## Step 3 — Fetch pending PRs

Use the GitHub search API to list PRs that have the cherry-pick label but not cherry-pick-done, sorted by merge date ascending:

Expand All @@ -25,7 +112,7 @@ gh api "search/issues?q=repo:NVIDIA/Model-Optimizer+is:pr+is:merged+base:main+la

Present the list to the user before proceeding.

## Step 3 — Set up the release branch
## Step 4 — Set up the release branch

Check out `release/<VERSION>`, creating it from the remote if it doesn't exist locally:

Expand All @@ -34,15 +121,15 @@ git fetch origin release/<VERSION>
git checkout release/<VERSION>
```

## Step 4 — Get merge commit SHAs
## Step 5 — Get merge commit SHAs

All PRs are squash-merged, so each has a single-parent commit. Retrieve the SHA for each PR:

```bash
gh pr view <NUM> --repo NVIDIA/Model-Optimizer --json mergeCommit --jq '.mergeCommit.oid'
```

## Step 5 — Cherry-pick in merge order
## Step 6 — Cherry-pick in merge order

Cherry-pick each commit with `-s` (DCO sign-off). GPG signing is handled automatically by the repo's git config.

Expand All @@ -56,7 +143,7 @@ git cherry-pick -s <SHA>
git cherry-pick --continue
```

## Step 6 — Create a PR to the release branch
## Step 7 — Create a PR to the release branch

Push the cherry-picks to a new branch and open a PR targeting `release/<VERSION>`. The PR title lists every cherry-picked PR number. The body uses `## Cherry-picked PRs` as the only heading with one `- #<NUM>` bullet per PR — no titles, no links, no extra text.

Expand All @@ -78,7 +165,7 @@ EOF
)"
```

## Step 7 — Apply cherry-pick-done label
## Step 8 — Apply cherry-pick-done label

Add the `cherry-pick-done` label to every PR that was successfully cherry-picked:

Expand Down