Imported from Mergifyio/mergify-cli (
skills/mergify-stack/SKILL.md). Install upstream withnpx skills add Mergifyio/mergify-cli --skill mergify-stack. Copyright stays with the author.
Mergify Stack Workflow
Stack Philosophy
A branch is a stack. Keep stacks short and focused:
- A stack should only contain commits that depend on each other
- Rationale: longer stacks take longer to merge
Proactive stack management:
- If an existing stack can be split into independent stacks, offer to do so
- When asked to do something new: if it can be done on a separate branch, either do so or ask if in doubt
- Default to creating a new branch for unrelated changes
Core Conventions
- Push: Use
mergify stack push(nevergit push) - Fixes: Use
git commit --amend(never create new commits to fix issues) - Amend notes: When amending a commit that already has a PR (i.e. has been pushed), attach a
mergify stack noteBEFOREmergify stack pushto record why the commit was amended. The note appears in the PR's "Revision history" comment and JSON marker, so reviewers can see the reason without diffing. - Mid-stack fixes: Stash any local changes first (
git stash -u), then usemergify stack edit <SHA-or-Change-Id-prefix>to pause the rebase at the target commit. Amend it withgit commit --amend, thengit rebase --continue, thenmergify stack push, thengit stash pop. Non-interactive — never usegit rebase -ifor this. (Callingmergify stack editwith no argument falls back to a fully interactivegit rebase -iand will hang in agent contexts — always pass a commit prefix.) - Reordering: Stash any local changes first (
git stash -u), then usemergify stack reorder(list all commits in desired order) ormergify stack move(move a single commit) instead of manualgit rebase -i— non-interactive and avoidsGIT_SEQUENCE_EDITORquoting issues - Fixup: Stash any local changes first (
git stash -u), then usemergify stack fixup <SHA>...to fold a commit into its parent (drops the listed commit's message). Non-interactive — never usegit rebase -ifor this. - Squash: Stash any local changes first (
git stash -u), then usemergify stack squash SRC... into TARGET [-m "msg"]to combine multiple commits into one, with an optional custom message. Non-interactive — never usegit rebase -ifor this. - Reword: Stash any local changes first (
git stash -u), then usemergify stack reword <SHA> -m "new message"to change a commit's message in place. Non-interactive when-mis given — never usegit rebase -ifor this. - Drop: Stash any local changes first (
git stash -u), then usemergify stack drop <SHA>...to remove commits from the stack. Non-interactive — never usegit rebase -ifor this. - Commit titles: Follow Conventional Commits (e.g.,
feat:,fix:,docs:) - PR title & body:
mergify stackcopies the commit message title to the PR title and the commit message body to the PR body — so write commit messages as if they were PR descriptions. Everything that should appear in the PR (ticket references, context, test plans) MUST go in the commit message. - Ticket references: Include ticket/issue references (e.g.,
MRGFY-1234,Fixes #123) in the commit message body, not added separately to the PR. - PR lifecycle is fully managed by
mergify stack: NEVER edit PR titles, bodies, or labels withgh pr editor the GitHub MCP — they will be overwritten on the next push. NEVER close or merge PRs manually —mergify stackhandles the entire PR lifecycle (creation, updates, and cleanup). - Draft PRs: NEVER mark a PR as ready-for-review — all PRs stay as drafts. The user will manually move them out of draft after reviewing.
- Each commit must pass CI independently: Every commit in a stack becomes its own PR. Each PR runs CI separately, so every commit must be self-contained — it must compile, pass linters, and pass tests on its own without depending on later commits in the stack. When formatting or linting fixes are needed, they must be included in the commit that introduced the issue, not deferred to a later commit.
Common Mistakes
| Wrong | Right | Why |
|---|---|---|
git push |
mergify stack push |
Git push bypasses stack management and breaks PR relationships |
| New commit to fix lint/typo | git commit --amend (HEAD) or git commit --fixup <SHA> + git rebase --autosquash (mid-stack) |
Each commit = a PR; fix commits create unwanted extra PRs |
gh pr edit --title "..." |
Edit the commit message, then mergify stack push |
PR title/body are overwritten from commit messages on every push |
gh pr merge or gh pr close |
PR lifecycle is fully managed — do nothing | PR lifecycle is fully managed by the stack tool |
git commit on main |
mergify stack new <name> first |
mergify stack push will fail on the default branch |
git rebase -i to fixup a commit |
mergify stack fixup <SHA> |
Non-interactive — works inside LLM/agent sessions; no editor spawned |
git rebase -i to squash commits |
mergify stack squash A B into X [-m "..."] |
Non-interactive — works inside LLM/agent sessions; no editor spawned |
git rebase -i to change a commit message |
mergify stack reword <SHA> -m "..." |
Non-interactive — works inside LLM/agent sessions; no editor spawned |
git rebase -i to amend a mid-stack commit |
mergify stack edit <SHA-or-Change-Id-prefix> then git commit --amend then git rebase --continue |
Non-interactive — pauses the rebase at the target commit without spawning an editor |
git rebase -i to drop a commit |
mergify stack drop <SHA>... |
Non-interactive — works inside LLM/agent sessions; no editor spawned |
GIT_SEQUENCE_EDITOR='sed -i ...' git rebase -i (any variant) |
One of mergify stack {edit,fixup,squash,reorder,move} |
Hand-rolled sequence-editor scripts are brittle; there is already a non-interactive command for every common rewrite |
| Deferring lint fixes to a later commit | Include the fix in the commit that caused it | Each commit runs CI independently; later commits won't save earlier ones |
| Rebase/reorder/checkout/sync with dirty worktree | git stash -u first, then git stash pop after |
Uncommitted changes are lost or cause conflicts during these operations |
| Amending a pushed commit with no explanation | mergify stack note -m "why" before mergify stack push |
The reason is recorded in the PR's Revision history table and JSON marker, so reviewers don't need to diff to understand the change |
Commands
mergify stack new NAME # Create a new stack/branch for new work
mergify stack push # Push and create/update PRs (also registers as a GitHub-native stack by default)
mergify stack push --no-github-native # ...without registering it as a GitHub-native stack
mergify stack checkout BRANCH # Checkout an existing stack from GitHub (e.g. someone else's)
mergify stack checkout PR_URL # Same, from any PR in the stack — middle included
mergify stack sync # Fetch trunk, remove merged commits, rebase
mergify stack list # Show commit <-> PR mapping for current stack
mergify stack list --json # Same, but machine-readable JSON output
mergify stack reorder C A B # Reorder all commits (pass SHA or Change-Id prefixes)
mergify stack move X first # Move commit X to the top of the stack
mergify stack move X last # Move commit X to the bottom of the stack
mergify stack move X before Y # Move commit X before commit Y
mergify stack move X after Y # Move commit X after commit Y
mergify stack fixup X # Fold commit X into its parent (drops X's message)
mergify stack fixup X Y Z # Fold each into its parent (multi-fixup)
mergify stack squash X into Y # Reorder X adjacent to Y, fold X into Y (keeps Y's message)
mergify stack squash X Y into Z -m "msg" # Fold X Y into Z with a custom message
mergify stack reword X -m "msg" # Change commit X's message non-interactively
mergify stack reword X # Change commit X's message via $GIT_EDITOR (TTY only)
mergify stack edit X # Pause the rebase at X so you can `git commit --amend` it (X is required; no-arg form is interactive)
mergify stack drop X # Drop commit X from the stack
mergify stack drop X Y Z # Drop multiple commits in one rebase
mergify stack note -m "why" # Attach an amend reason to HEAD (shown in PR revision history)
mergify stack note <SHA-or-Change-Id-prefix> -m "why" # Attach to a specific commit in the stack
mergify stack note --append -m "more" # Append to an existing note
mergify stack note --remove # Remove the note from a commit
Use mergify stack checkout to check out a stack that exists on GitHub (e.g. a colleague's stack). The argument is either the stack's remote branch name — whatever prefix it uses, no assumption that it contains an author — or the URL of any pull request in the stack, including one from the middle. It fetches all stacked PRs, creates a local branch, and sets up tracking.
The local branch defaults to the stack branch with your own stack branch prefix removed, so a stack you pushed from feature/login comes back as feature/login and a later mergify stack push updates the same PRs. For a stack that is not under your prefix — a colleague's — the last segment is used and checkout warns you: you can work on those commits, but mergify stack push from that branch would create a separate stack rather than update their pull requests. Use --branch to override the name. A pull request URL carries its own owner/repo, so --repository is ignored in that form.
Use mergify stack sync to bring your stack up to date. It fetches the latest trunk, detects which PRs have been merged, removes those commits from your local branch, and rebases the remaining commits. Run this before starting new work on an existing stack.
Use mergify stack list to see which commits have been pushed, which PRs they map to, and whether the stack is up to date with the remote. It also shows CI status, review status, and merge conflicts for each PR. Use --verbose for detailed check names and reviewer names. Use --json when you need to parse the output programmatically — it includes full CI check details and review data.
GitHub-native stacks (experimental, on by default)
mergify stack push registers the stack with GitHub's own Stacks API by
default, so GitHub renders it as a stack. Opt out per invocation with
--no-github-native, or per repo with
git config mergify-cli.stack-github-native false.
Change-Ids, branch layout, stack comments and revision history are unchanged,
and it degrades quietly: where the API isn't available (older GitHub
Enterprise, a repo without the feature) the push reports
not registered on GitHub and succeeds exactly as it would have.
Three things to know:
- A stack needs at least 2 pull requests. GitHub rejects a 1-PR stack, so a single-change stack stays a plain PR — and a 2-PR stack that loses a member is dissolved rather than re-registered.
- The
Depends-On:header goes away. A registered stack is the dependency between two pull requests, so the CLI stops writing a second copy of it into the PR descriptions. It is keyed off the registration, not off the flag: when a push degrades tonot registered on GitHub, the headers are written back in the same push and Mergify keeps ordering the stack. Your commit messages are never touched either way — the header only ever existed in the rendered PR description. - Registering changes how the PRs merge. While a stack is registered,
GitHub refuses the classic merge endpoint
(
PUT /repos/{owner}/{repo}/pulls/{pull_number}/merge→ 403) for its members. That is GitHub's contract, not ours; it is the reason--no-github-nativeexists.
Pushing stays cheap. Refreshing commits (amend, reword, force-push) leaves the
registration untouched, and adding a change on top extends the same stack.
Only a push that moves a pull request's base — a reorder, a drop, a change
inserted in the middle — dissolves the registration first and rebuilds it at
the end, because GitHub rejects any base-branch change while a PR is stacked.
An interrupted push therefore leaves the stack merely unregistered — never
half-registered. It can also leave it without its Depends-On: headers, since
those are restored at the end of the push; re-running mergify stack push
settles both, because every push re-renders the descriptions and recomputes the
Depends-On: markers from the stack's current shape — including with
--keep-pull-request-title-and-body, where the description is re-rendered from
the pull request's own body rather than the commit message, and the marker is
still stripped and re-appended rather than carried over.
Amend Notes
mergify stack note records why a commit was amended. The note travels with the stack:
- The reason is stored locally under
refs/notes/mergify/stackagainst the commit SHA. - On
mergify stack push, the reason is consumed into the change's revision history; the note on the pushed head commit is replaced by the full revision history (human digest + the<!-- mergify-revision-data: {...} -->JSON marker). Git notes — not the PR comment — are the machine-readable source of truth; the PR's "Revision history" comment is rendered from them. - At merge time, Mergify copies the head commit's history note onto the merge/squash commit, so
git log --notes=mergify/stackon the base branch shows why each change was revised.
When to attach a note — any time you amend or rewrite a commit that already has a PR open (i.e. it has been pushed at least once). The note answers "why is this revision different?" so the reviewer doesn't have to diff old vs new SHAs to find out.
Workflow — attach the note BEFORE mergify stack push:
# Edit HEAD, then:
git commit --amend
mergify stack note -m "address review: rename foo() to bar()"
mergify stack push
# Or for a mid-stack commit (after the rebase that amended it):
mergify stack note <SHA-or-Change-Id-prefix> -m "fix lint reported in CI"
mergify stack push
A note is per-commit, not per-revision. Each amend (or other history rewrite) creates a new commit SHA, so you must run mergify stack note again for the new SHA — the previous note stays attached to the old SHA and won't carry over. Use --append only when the current target commit already has a note and you want to add another reason; use --remove to clear it. Notes on commits that haven't changed since the last push are preserved but won't add a new revision row.
CRITICAL: Check Branch Before ANY Commit
BEFORE staging or committing anything, always check the current branch and assess stack state:
git branch --show-current
mergify stack list
- If you're on
main(or the repo's default branch): you MUST create a feature branch first - NEVER commit directly on
main—mergify stack pushwill fail - This check must happen before
git add, not aftergit commit
CRITICAL: Stash Local Changes Before Worktree-Modifying Operations
BEFORE running any operation that rewrites history or switches branches, check for uncommitted changes and stash them:
git status --short # Check for uncommitted changes
git stash -u # Stash tracked + untracked changes if any
Operations that require this check:
mergify stack edit <commit>(mid-stack fixes)mergify stack reorder/mergify stack movemergify stack fixupmergify stack squashmergify stack rewordmergify stack dropmergify stack checkoutmergify stack syncmergify stack new(switches to new branch)git checkout <branch>
After the operation completes, restore the stashed changes:
git stash pop
If you skip this step, uncommitted work will be silently lost or cause rebase conflicts.
Starting New Work
When asked to start a new piece of work, create a new feature, or work on something unrelated to the current stack:
- Check current branch:
git branch --show-current - Create a new stack:
mergify stack new <branch-name>- Use descriptive branch names following the pattern:
type/short-description(e.g.,feat/add-login,fix/memory-leak)
- Use descriptive branch names following the pattern:
- Make commits following conventional commits
- Push:
mergify stack push
Adding to Existing Stack
When continuing work on an existing feature branch:
- Check current branch:
git branch --show-current- If on the right branch: proceed with commits
- If on
main: switch to the feature branch first withgit checkout <branch>or create a new stack
Conflict Resolution
When a rebase causes conflicts (during git rebase -i or mergify stack push):
- Resolve conflicts in your editor
- Stage resolved files with
git add - Continue with
git rebase --continue
To abort instead: git rebase --abort
After resolving, run mergify stack push to sync the updated stack.