make-pr: stack on top, or overwrite the branch, decided from the diff

[EdbertChan]

Summary

When a PR's diff stops matching its title, the branch gets rewritten rather than explained.

When the title still holds and the addition is small, another commit goes on top instead.

The choice is read off the diff, so nobody has to be asked which they would prefer.

Review Claim

make-pr gains a "Stack on top, or overwrite the branch" section that decides from the PR's current diff, with a colocated test pinning both treatments, the three triggers for overwriting, the lease, and the backup branch.

Review Lane

behavior

Review Unit

engine-runtime

Safety Invariant

Skill prose and its test only. No script, gate, or hook changes, so nothing that runs today behaves differently; the rule is read by whoever writes the next PR.

Slice Rationale

One rule and its test, on their own. The stale generated docs block this work uncovered ships as its own PR (plan/skill-triggers-regenerate), because it is a different claim and a different unit.

Non-goals

• No new gate: deciding whether a diff still matches a claim needs judgment, so no script enforces this.
• No change to preflight.py or the PR-body checker.
• No change to the review-unit table or the stacking rule above it.

Test Plan

Test Plan

• python3 -m unittest discover -s engine/skills/make-pr/tests: Ran 27 tests ... OK
• check_skill_test_coverage.py --base origin/main --head HEAD, check_skills_three_harnesses.py, check_skill_trigger_mechanism.py, check_ecosystem_boundaries.py, check_skill_file_refs.py, check_no_new_comments.py --base origin/main, check_no_dated_provenance.py --base origin/main: all exit 0
• check_skill_trigger_policy.py: fails with docs/skill-triggers.md: generated block is stale. It fails the same way on main at 329c844, on a clean checkout of main with no changes of mine, so it is not from this PR. plan/skill-triggers-regenerate fixes it; preflight passes once that lands.

Revert Plan

Revert Plan

• Safe to revert? Yes. Prose and a test.
• Revert command: git revert <merge-sha>
• Post-revert steps: None.
• Data migration? No

🤖 Generated with Claude Code

https://claude.ai/code/session_01Gnua5cuAAF1ey1b7zRqJCC

---

Note

Low Risk
Documentation and a prose-pinning test only; no hooks, scripts, or runtime behavior change.

Overview
The make-pr skill now documents how to handle an existing PR branch when more work is needed: stack on top if the PR claim still matches the diff and the addition is small, or overwrite the branch when the diff no longer matches (duplicate base work, mixed review units, or a second claim). Overwrites require a backup/pr<number>-<short-sha> ref, --force-with-lease, and a rewritten title/body—not Slice Rationale excuses—and a failed lease means re-read and decide again rather than forcing.

A colocated unittest (test_stack_or_overwrite.py) locks in the section heading, both treatments, the three overwrite triggers, lease/backup requirements, and that the choice is never delegated to the user.

^{Reviewed by Cursor Bugbot for commit a56c15e. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cursor]

Bugbot couldn't run - usage limit reached

Bugbot is counted against Cursor usage for this user or team, and this run hit a usage or spend limit.

A user or team admin can review and increase usage limits in the Cursor dashboard.

(requestId: serverGenReqId_e7cb0f94-a767-49ab-8aae-c02d24120787)

[mergify]

Queued — the merge queue status continues in this comment ↓.

[EdbertChan]

@Mergifyio queue

[mergify]

Merge Queue Status

• ✅ Entered queue — 2026-09-12 01:15 UTC · Rule: admin-bypass · triggered by @EdbertChan with the @mergifyio queue command
• ✅ Checks passed · on draft merge queue: checking #457 on main (94593f3), stacked on #446 #460
• ✅ Merged — 2026-09-12 01:22 UTC · at 10884a37706f9593c9e87c5e065af73fc31b14a6 · squash

This pull request spent 6 minutes 47 seconds in the queue, including 5 minutes 38 seconds running CI.

Required conditions to merge

• check-success = lint
• check-success = test