stacked-pr-order
Skill9 starsChanged 45 days agoApache-2.0View on GitHub
Land a stack of dependent PRs in core-agent without closing the downstream ones. Use when a branch is based on another feature branch rather than main, or when merging a parent whose children are still open. Covers the retarget-before-merge order, rebasing after each parent lands, recovery, and reading merge state correctly.
- Deletes or force-pushes
- Commits and pushes
In this repository10 other files it gives its agents.
- .agents/AGENTS.mdAGENTS.md
- AGENTS.mdAGENTS.md
- .agents/reviewer/AGENTS.mdAGENTS.md
- adversarial-review-gateSkill
- changelog-bulletSkill
- prefix-failure-verificationSkill
- presubmit-sweepSkill
- autonomous-setupSkill
- cli-setupSkill
- library-embeddingSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 1 is distinctive, and 4 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
4
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 8
When feat/B depends on feat/A, base PR B on branch A.
- Distinctive: line 25
Before deleting any remote branch, check for an open PR pointed at it.
- Very specific: line 38
After gh pr edit <child> --base main, wait for the real CI run to conclude (gh run list --branch <b>) before merging the child — the same …
- Very specific: line 55
Use the explicit form: --force-with-lease=<ref>:<sha>.
- Very specific: line 114
If --delete-branch fails with *"'main' is already used by worktree"*, the
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Stacked PR order
When feat/B depends on feat/A, base PR B on branch A. Three things go
wrong, in this order of likelihood.
1. Retarget downstream PRs to main BEFORE merging the parent
gh pr merge A --delete-branch closes any PR whose base was branch A.
Deleting a head branch closes the PRs pointed at it, and that is not
reversible by re-merging.
gh pr edit B --base main # first
gh pr merge A --admin --squash --delete-branch # thenRecovery if you forget: push the parent SHA back to re-create the branch,
gh pr reopen B, gh pr edit B --base main.
Before deleting any remote branch, check for an open PR pointed at it.
1a. A stacked child gets no real CI until it targets main
ci.yml runs on pull_request: branches: [main], as do docs-lint,
go-toolchain, version-fallback, ci-docs and the PR-time agent
attribution check. A PR based on another feature branch gets only the
workflows without that filter — in practice review-gate; test,
lint, apidiff and govulncheck never run on it. A green check list
on a stacked PR is therefore not evidence about its Go code.
- Say so in the PR body, and lean on the full local presubmit sweep until it is retargeted.
- After
gh pr edit <child> --base main, wait for the realCIrun to conclude (gh run list --branch <b>) before merging the child — the same check as for--adminmerges below.
2. Rebase the downstream onto new main after each parent lands
git rebase --onto origin/main <old-parent-sha>The --onto form is the point: the parent landed as a squash, so its
commits are not ancestors of main and a plain rebase replays them on top of
their own squashed copy. You get conflicts against your own change.
git push --force-with-lease on your own branch is normal here. A
preceding git fetch disarms it — it compares against the remote-tracking
ref the fetch just updated, so it will happily overwrite someone else's push.
Use the explicit form: --force-with-lease=<ref>:<sha>.
3. The CHANGELOG duplication
A base-update on a branch stacked on a release-promotion PR duplicates the
version heading with no conflict. See the changelog-bullet skill — it is
the same hazard and the check is git diff <base>...HEAD -- CHANGELOG.md.
Reading merge state
"No checks reported" means the PR is DIRTY, not that CI is broken. GitHub
cannot compute a merge commit for a conflicting PR, so pull_request
workflows never queue at all.
gh pr view <N> --json mergeStateStatus| Status | Means |
DIRTY | conflicts; no checks will ever run. Rebase and force-push, they fire immediately |
BEHIND | stale but not conflicting; --admin ignores it |
BLOCKED | still running |
CLEAN | ready |
UNKNOWN | recomputing; re-poll in ~20s |
CLEAN with zero failed checks is not enough on a PR that touches Go.
ci-docs.yml's no-op jobs reuse the required check names, so confirm that
the real CI run concluded success (gh run list --branch <b>) before an
--admin merge (#1271).
Straight after a push it is just registration lag — wait 30s before
concluding anything. main moves faster than a full CI cycle, so a green PR
can flip back to BEHIND before you look; do not chase it with repeated
rebases.
Proving a branch is already in main before deleting it
Squash-merging means a merged branch is not an ancestor of main, so
git branch -d refuses it and merge-base --is-ancestor says "unmerged".
Cheapest first:
gh pr view N --json headRefOid,mergeCommit— if the branch tip still equalsheadRefOidand the merge commit is an ancestor of main, the whole branch was inside the merged diff.- Subject match +
git cherry, which can only ever corroborate. - For a still-unmatched commit, a two-dot diff scoped to exactly the files
that commit touched. Both unscoped forms are wrong: plain two-dot is
drowned by everything main gained since, and three-dot
(
merge-base..branch) reports what the branch added, which is true whether or not main later took it.
Merging
gh pr merge <N> --admin --squash --delete-branch is the maintainer path for
the rebase-then-merge cascade. Never merge unverified CI — confirm
mergeStateStatus: CLEAN and zero failed checks first. --admin is not a
way to skip review on a contributor PR.
If --delete-branch fails with "'main' is already used by worktree", the
merge still succeeded — only the local checkout step failed. Verify with
gh pr view <N> --json state and delete the branch with
git push origin --delete <branch>.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.

