Production Hotfix

Ships a small fix directly to main without going through the release PR pipeline, then backmerges main into dev so the two branches stay in sync. Normal changes still go through /release — this exists only for urgent fixes that cannot wait.

When NOT to use this

• Multi-file refactors or feature work → use /release
• Anything that needs CodeRabbit review or a second pair of eyes → use /release
• Changes touching migrations, auth, billing, or anything with a non-trivial blast radius → too risky for the bypass, open a hotfix PR instead

Prerequisites

• The fix is scoped to one or two files, minimal blast radius
• You have admin permissions on the repo (the push to main bypasses branch protection — non-admins get rejected outright)
• No concurrent release PR is in flight (check with gh pr list --label release)

Starting states this skill handles

The workflow branches on what state the local checkout is in when the skill runs:

1. Uncommitted change on dev — working tree is dirty with the fix
2. Already committed on dev — you committed the fix before realizing it needs the hotfix path (what we hit in the motivating session)
3. Nothing yet — no edit made; stop and make the edit on dev first, then rerun

Steps

Run steps in order. Stop and report if any step fails — do not try to push through a red state.

Step 1: Normalize the starting state

Check where you are:

git branch --show-current
git status --short
git log --oneline origin/dev..HEAD

Handle each starting state:

If there is a local commit on dev (third command has output), uncommit it but keep the change in the working tree:

git reset --mixed HEAD~1

If the working tree has uncommitted changes (from the reset above, or because the user never committed), stash them so the pull cannot conflict:

git stash push -m "prod-hotfix"

Stash everything rather than cherry-picking paths — the fix may touch multiple files and we want the whole diff moved atomically.

If the working tree was clean AND there was no local commit, stop. There is nothing to hotfix. Ask the user to make the edit on dev first.

Step 2: Pull both branches

git fetch origin
git pull --ff-only origin dev
git checkout main
git pull --ff-only origin main

Both pulls must be fast-forwards. If either refuses to fast-forward, stop — local divergence means this hotfix path is not safe and the user should investigate manually. Do not use git pull --rebase or git reset --hard to force it.

Step 3: Apply the fix on main

git stash pop
git diff

Read the diff carefully. Confirm it is exactly the intended fix and nothing else — no stray formatter churn, no unrelated files. If the stash pop conflicts, main has diverged from dev in the file being fixed and the fix probably needs to be rewritten against the current main contents; stop and report.

Step 4: Verify

Run the project verification chain before committing. All three must pass:

pnpm fix
pnpm lint
pnpm build

pnpm fix runs auto-fixers (may modify files — if it does, re-inspect with git diff). pnpm lint runs all quality gates in parallel. pnpm build confirms the production build still compiles. Do not skip pnpm build even for trivial changes — silent build regressions on main are exactly what branch protection exists to prevent, and the skill is already bypassing that.

Step 5: Commit on main

Use a conventional commit matching the project style (see .claude/rules/commands.md). Example:

git add <path/to/file>
git commit -m "$(cat <<'EOF'
<type>(<scope>): <gitmoji> <subject>

- <why / what changed>
EOF
)"

Types used in this repo: fix for bugs, style for UI/cosmetic, perf, chore, docs, refactor, feat. Gitmoji goes after the colon+space. Subject under 50 chars, imperative mood, no trailing period.

Step 6: Push to prod

git push origin main

Expected output includes a bypass notice:

remote: Bypassed rule violations for refs/heads/main:
remote: - Changes must be made through a pull request.
remote: - Required status check "Vercel" is expected.

This is normal — branch protection on main is configured to allow admin bypass. If the push is rejected outright (not just annotated with "Bypassed"), the user does not have admin and this path cannot continue. Stop and tell them to open a hotfix PR against main instead.

Step 7: Backmerge main → dev

git checkout dev
git merge main --no-ff -m "chore(release): merge main back into dev after hotfix"
git push origin dev

The --no-ff preserves the merge commit so the backmerge is visible in history, matching the commit style the CI release pipeline uses (grep for d0196bd5 in the log as a reference). This keeps dev ahead of main by exactly the already-unreleased work plus the merge commit.

Step 8: Confirm clean state

git pull
git log --oneline -5
git log --oneline origin/main..origin/dev | head -3

The log should show, top-down: the backmerge commit, the hotfix commit, the previous dev tip. origin/main..origin/dev should still list every unreleased commit — confirming the hotfix did not accidentally ship any of the in-flight work.

After the hotfix

• Watch Vercel on main. The hotfix triggers the normal Vercel production deploy on main but does NOT run release-it, so no new tag or GitHub Release is created. The next proper release (via /release) will pick up the hotfix commit in the changelog automatically.
• API changelog. If the hotfix touches public API surface, add the entry to docs/API_CHANGELOG.md on dev after the backmerge — it will be posted as a PR comment during the next release.
• Post to Slack dev channel (optional, matches the /release skill convention): a short note that a hotfix shipped directly to main, with the commit SHA and a one-line description.

Red flags — stop and report instead of pushing through

Signal	What it means	Action
Either pull refused to fast-forward	Local dev or main has diverged	Investigate manually; do not force
pnpm fix / lint / build fails	Fix has a secondary regression	Fix the regression, rerun verification, only then commit
git stash pop conflicts	main diverged from dev in the target file	Rewrite the fix against current main
Push to main rejected (not bypassed)	No admin permissions	Open a hotfix PR against main instead
origin/main..origin/dev shows unexpected commits after Step 8	Backmerge picked up stale state	Investigate; may need to reset dev

Why this skill exists

The normal release pipeline (/release) bundles commits from dev into a release/* branch, runs CodeRabbit + CI, merges to main via admin-approved PR, and runs release-it to cut a tag. That pipeline is the right default — it exists because shipping straight to main is risky.

This skill exists for the narrow case where:

• The fix is so small and so obviously correct that full review is overhead, not safety
• The issue is user-visible in production right now and waiting for the next release cycle is worse than the residual risk of bypassing review
• The person running the skill has admin and is accountable for the call

Keep the bar high for invoking it. If in doubt, use /release.