Repository navigation
F: publish only ledger-recorded commits (snapshot head may hold others) #113
Description
Activity
Decision, made 2026-10-03: keep publishing the whole task head, as the design intends. Implemented in PR #119, which is stacked on #110.
- What a task's head contains. First the user's own branch (
base..HEAD), which the review records before the runner makes its first commit. Then the runner's commits on top. - How the user's commits are handled. The ledger has no entry for them, so they count as foreign. Review shows their changes in the Unplanned row, and the merge gate blocks until each one is assigned or accepted.
- What F: test and document that a published head carries the branch's own commits (#113) #119 adds. A test that publishes such a head and checks the Unplanned row and the merge blocker. It also documents this in
pull-request-opening.md. - What stays with Add pre-merge rebase and head-bound command checks #22. Mapping commits through a rebase. Add pre-merge rebase and head-bound command checks #22 also has to revisit
ownedCommitsonce rebasing writes foreign ledger entries.
One gap found along the way and left for a separate follow-up: the merge gate inspects and merges
github.pullRequest, not the task's published PR.This issue closes when PR #119 reaches
main.- What a task's head contains. First the user's own branch (
- added a commit that references this issue
on Oct 4, 2026
Why
From the #103 review (PR #110). The publisher pushes
snapshot.head, and for a review the snapshot head can come from the user's checkout (runner/review.ts).GitBranchPusheronly checks that the remote branch's current commit is ledger-owned, before it overwrites the branch. It does not check the commits being pushed. So commits the ledger does not record can reach thecodeboost/branch and the PR. These could be the user's own local commits, or an unrelated commit like the demo's "Unrelated diagnostic output". Otherwise, the push fails with "the runner repository has no commit".This overlaps #22: rebasing writes
foreignledger entries. When #22 lands,ownedCommitsmust also be revisited (finding 4 of the #101 review).Needed
base..headrange holds a commit the ledger does not record asowned, unless a person accepted it.Tests
🤖 Generated with Claude Code