Files
PaperClipAI/server
Devin FoleyandPaperclip 936687ca55 fix(workspace): restore clean branch drift on finalize (#8914)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Agent runs can execute inside reusable, runtime-created git worktree
execution workspaces.
> - Those managed worktrees record the expected branch so later
dispatches do not accidentally run an agent in the wrong checkout.
> - Successful run finalization already checked branch coherence, but it
treated every unrecorded branch switch as fatal.
> - A common publishing flow can briefly switch a clean worktree to a
PR/publish branch that points at the same commit as the recorded issue
branch, leaving no divergent work to protect.
> - This pull request keeps the strict finalization guard for unsafe
drift, but lets finalization restore the recorded branch when
same-commit repair is provably safe.
> - The benefit is fewer false failed runs after harmless branch
switches while preserving hard failures for divergent or dirty
worktrees.

## Linked Issues or Issue Description

No public issue exists for this exact finalization failure. Related
public worktree-recovery context: #3087 and #3056, but those address
different worktree realization/reuse recovery paths rather than
successful-run finalization branch repair.

Bug report details:

**What happened?**

When an adapter run succeeded after switching a managed git worktree
from its recorded issue branch to a publish/PR branch, finalization
failed with a managed worktree branch mismatch even when the publish
branch and recorded branch pointed at the same commit and the worktree
was clean.

**Expected behavior**

Finalization should restore the recorded branch only when it can prove
the worktree is clean, registered, and the recorded branch points at the
current `HEAD`. If the actual branch has different commits or unsafe
state, finalization should continue to fail with bounded validation
evidence.

**Steps to reproduce**

1. Create a runtime-managed `git_worktree` execution workspace for an
issue run.
2. During the adapter run, create and check out a new publish branch
without committing new changes.
3. Return adapter success and let heartbeat finalization run.
4. Before this change, finalization records a failed branch check and
fails the run even though the branches point at the same commit.
5. With this change, finalization records the repair operation, restores
the recorded branch, and records a successful finalize row.
6. Repeat with a commit on the publish branch; finalization still fails
because the branch heads differ.

**Paperclip version or commit**

Reproduced against `master` at `bac7307ec`; fixed by this PR at
`64ec605cf`.

**Deployment mode**

Local dev / built from source.

**Agent adapter(s) involved**

Not adapter-specific. This is core heartbeat/workspace finalization
behavior.

**Database mode**

Embedded test Postgres in the focused server test.

**Access context**

Agent run finalization.

**Node.js version**

`v25.6.1`

**Operating system**

`Darwin 24.6.0 arm64`

**Relevant logs or output**

The new focused test intentionally exercises both outcomes:

```text
Test Files  1 passed (1)
Tests       3 passed (3)
```

**Relevant config**

Runtime-created `git_worktree` execution workspace.

**Additional context**

The unsafe divergent branch case still fails with
`workspace_validation_failed` and `git_worktree_branch_incoherence`
evidence.

**Privacy checklist**

Reviewed; this description avoids internal task links, local workspace
paths, credentials, and instance-specific URLs.

## What Changed

- Reused the existing guarded branch-coherence repair helper during
heartbeat finalization when the final branch inspection finds clean
same-commit branch drift.
- Recorded repair metadata in the `workspace_finalize` operation so
reviewers/operators can audit whether finalization repaired branch
drift.
- Preserved failure behavior for divergent branch heads and surfaced the
bounded workspace validation evidence from the repair helper.
- Added focused server coverage for safe finalization repair and unsafe
divergent branch failure.
- Updated execution semantics docs to describe the narrower finalization
rule.

## Verification

- `pnpm exec vitest run
server/src/__tests__/heartbeat-workspace-finalize-branch.test.ts`
- `pnpm --filter @paperclipai/server typecheck`
- `git diff --check`

## Risks

Low to medium risk. The change affects successful-run finalization for
runtime-created git worktree execution workspaces. The repair path is
constrained to clean, registered, same-commit branch drift, and the
focused test confirms divergent branch heads still fail instead of being
restored silently.

## Model Used

OpenAI Codex, GPT-5-based coding agent. Exact hosted model ID was not
exposed in the runtime; tool use and local shell execution were enabled.

## Checklist

- [x] I have included a thinking path that traces from project context
to this change
- [x] I have specified the model used (with version and capability
details)
- [x] I have checked ROADMAP.md and confirmed this PR does not duplicate
planned core work
- [x] I have searched GitHub for duplicate or related PRs and linked
them above
- [x] I have either (a) linked existing issues with `Fixes: #` / `Closes
#` / `Refs #` OR (b) described the issue in-PR following the relevant
issue template
- [x] I have not referenced internal/instance-local Paperclip issues or
links (only public GitHub `#NNN` / `github.com/paperclipai/paperclip`
URLs)
- [x] My branch name describes the change (e.g. `docs/...`, `fix/...`)
and contains no internal Paperclip ticket id or instance-derived details
- [x] I have run tests locally and they pass
- [x] I have added or updated tests where applicable
- [x] I have updated relevant documentation to reflect my changes
- [x] I have considered and documented any risks above
- [x] All Paperclip CI gates are green
- [x] Greptile is 5/5 with no open P2s, recommendations, or follow-ups
- [x] I will address all Greptile and reviewer comments before
requesting merge

---------

Co-authored-by: Paperclip <noreply@paperclip.ing>
2026-07-02 20:58:26 -07:00
..