Skip to content

Runner workspace publication reuses stale post-squash branch history #1894

Description

@chubes4

Problem

A runner workspace configured with a canonical publication branch reused the remote branch history left behind after an earlier squash merge instead of starting from the requested current base ref. The model change and deterministic verification succeeded, but the resulting PR included historical commits, stale generated documentation, and merge conflicts.

Evidence

Expected

Each runner publication starts from the declared from/base revision. Existing canonical branches must either be proven fast-forward-compatible with that base or replaced through a safe fresh-branch strategy. Post-squash historical commits must not reappear in a later PR.

Acceptance criteria

  1. Reusing a configured publication branch cannot carry commits absent from the declared base.
  2. Existing branch divergence is detected before agent execution or publication.
  3. The runner creates a fresh collision-safe branch or returns a deterministic actionable failure.
  4. Tests cover an existing branch whose prior commits were squash-merged into base.
  5. Publication output reports the exact effective branch and base SHA.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions