@kitlangton/stack
v0.4.6
Published
Squash-safe stacked PR/MR repair CLI for GitHub and GitLab
Downloads
816
Maintainers
Readme
stack
╭───STACK───╮
dev
└─ #101
└─ #102
└─ #103
╰───────────╯Squash-safe stacked PR/MR repair for coding agents working in GitHub or GitLab repos that squash-merge and delete branches.
stack is agent-first. Humans can run it directly, but the happy path is: let
the agent do normal code work with plain git, then use stack for stack
inspection, repair, merge, and undo workflows.
Install
npm install -g @kitlangton/stackInstall the agent skill too:
npx skills add kitlangton/stack --skill stackInstall and authenticate the matching host CLI:
gh auth login # GitHub
glab auth login # GitLabAgent Happy Path
- Create stacked changes using normal git branches.
- Open the root PR/MR against trunk, for example
mainordev. - Open each child PR/MR against its parent branch.
- Preview the stack:
stack sync- Apply the safe maintenance workflow:
stack sync --apply- Merge from the root when ready:
stack merge
stack merge --applyUse stack merge --auto when the code host should wait for merge requirements,
then repair descendants automatically after the root lands.
What It Does
stack sync --apply is the common maintenance workflow:
- Infers stack links from PR/MR target branches.
- Records stack intent in
.git/stack/state.json. - Publishes ahead parent branches before repairing their descendants.
- Repairs descendants after parent branches move or land.
- Retargets PRs/MRs when needed.
- Refreshes stack blocks in descriptions.
- Saves
.git/stack/undo.jsonbefore mutations.
For a parent that needs no local repair, sync checks its actual push destinations and only publishes fast-forward changes, with explicit leases. Diverged or unknown remote tips require fetching and reconciling first. Undo restores each destination's previous tip without discarding commits that were already on the local parent; it refuses to overwrite a remote tip changed since sync.
GitHub stack blocks use compact #101 references. GitLab blocks use !101
references plus titles because bare GitLab MR links only show titles on hover.
Recognized transient code-host read failures are retried up to twice, with jittered backoff around one and two seconds. Writes are not automatically retried: a timed-out create, merge, or update may already have succeeded. GitLab reuses known titles and preserves historical stack entries if optional title lookup fails.
If a repair fails, run:
stack history
stack undo
stack undo --applyGitHub And GitLab
Provider selection is automatic for public hosts:
github.comusesgh.gitlab.comusesglab.
For enterprise hosts, configure the repo once:
git config stack.codeHost github # or: gitlabUse STACK_CODE_HOST=github|gitlab for a one-off override.
Trunk Branches
By default, stack treats dev, main, and master as trunk branches. Repos
that use another trunk, such as develop, can configure the trunk list:
git config stack.trunks dev,develop,main,masterStack Block Heading
Each stack block has a heading that links back to this project. To render a
plain ### Stack heading without the attribution link — for example in
enterprise repos where external links trip compliance checks — set:
git config stack.blockLink falseThe linked heading stays on by default. This is repo-local; use
git config --global stack.blockLink false to apply it everywhere.
Example Output
Sync preview
● main
└─ ● stack-a #101
└─ ● stack-b #102
Would update PRs: #101, #102
Apply:
stack sync→ retarget #102 (stack-b) to main before merge
→ merge #101 (stack-a)
→ rebase stack-b onto main
→ push stack-b
→ update #102 stack blockCLI Reference
stack status # inspect the relevant local stack
stack skill # print the stack skill (agent instruction set)
stack sync # preview inference, repairs, and description updates
stack sync --apply # apply the previewed maintenance workflow
stack sync <branch> # preview only the stack containing branch
stack sync --apply <branch>
# apply only the stack containing branch
stack sync --apply --keep-going
# process independent stacks and report failures
stack doctor # inspect repo, host, metadata, and journal health
stack merge # dry-run the next root merge
stack merge --apply # merge root and repair descendants
stack merge --auto # wait for host requirements, then merge and repair
stack merge --auto --through <branch-or-change>
# auto-merge roots through a bounded target
stack history # show the last saved mutation journal
stack undo # preview undo
stack undo --apply # restore branch tips, request targets, and metadata