Files
PaperClipAI/ui/src/components
DottaandPaperclip 915edcff49 fix(workspaces): preserve archived native recovery during migration (#20424)
## Thinking Path

> - Paperclip is the open source app people use to manage AI agents for
work.
> - Durable workspace migration must retain saved files and pending
recovery work.
> - An archived Git workspace can have an authenticated restore recipe
and a terminal native run with pending remote files.
> - Its task can select a different workspace. Reopening the archived
workspace would change that task.
> - The current migration owner refuses this source because it has no
live local directory.
> - This pull request uses the existing online recovery preservation
owner for this exact archived recipe case.
> - Migration retains the old task, provider allocation, recovery
action, and native sync reference.

## Linked Issues or Issue Description

Refs #20298, #20394, and #20418. Related cleanup work in #19812 does not
cover this migration case.

**What happened?**

A historical archived Git workspace with a saved restore recipe and a
terminal prepared native sync reference cannot migrate online. Admission
reports `remote_copyback_unresolved` and
`workspace_recovery_unresolved`. The old task can already select another
workspace, so reopening it is not a safe migration path.

**Expected behavior**

Authenticate the archived files into private scratch. Preserve all typed
native and provider recovery obligations in the existing migration
journal and markers. Keep the original task selection, history, archive
state, and restore flags unchanged.

**Steps to reproduce**

1. Archive a Git worktree with staged, unstaged, and untracked files in
a saved recipe.
2. Retain a terminal failed native run, its prepared sync reference, and
its matching remote allocation and recovery action.
3. Remove the old local worktree and select a different workspace on the
task.
4. Call `migrateRemoteWorkspace` without `writersStopped`.

**Paperclip version or commit**

Reproduced on `b2a7d322ed0bafda04a0b78ce14e4f25c8d5d170`.

**Deployment mode**

Self-hosted server with PostgreSQL and configured object storage.

## What Changed

- Permit the existing online recovery owner to inspect recipe-backed
archived Git sources.
- Acquire the existing native finalization locks for those sources.
Reinspect their exact proofs after lock acquisition and before
publication.
- Lock related source-task ancestry before publication.
- Reuse the existing private Git restore, authenticated archive,
journal, marker, and recovery-source contracts. Add no schema or
provider calls.
- Leave active sources and archived sources without a valid recipe on
their existing refusal paths.
- Include the shared UI correction from #20418. A nullable draft
override follows fresh task data when the user has no local choice.
Explicit choices increment a generation. Normal and recovered successful
sends clear only their own choice; failed sends and newer choices remain
intact.

## Verification

- Unchanged master RED: the new historical-source fixture fails with
`remote_workspace_migration_blocked:remote_copyback_unresolved,workspace_recovery_unresolved`.
- Five focused real PostgreSQL, filesystem, Git, and encrypted-storage
tests pass on Node 24.22.2. The success case loses an upload reply,
resumes the same operation and snapshot, restores the
staged/unstaged/untracked bytes, and verifies unchanged task history,
lease, recovery action, finalizer, and recipe flags.
- Recipe, prepared reference, and provider allocation changes during the
actual upload each refuse publication. A competing native finalization
owner refuses before capture.
- Migration command: `pnpm exec vitest run
server/src/services/remote-workspace-migration.test.ts -t 'unselected
archived recipe|archived prepared source|competing native F owner'`.
- All server typecheck layers, module boundary checks, and whitespace
checks pass for the migration revision. The migration source bytes
remain unchanged through the shared UI follow-ups.
- Initial CI run 42727051535 exposed an existing planning-mode browser
failure: shard 3/8 passed 33 cases but the refreshed task mode did not
appear. The same issue occurred twice on #20418. A real local regression
proved that initial cached data could leave the composer in its old
mode. This path does not invoke migration.
- Review then exposed genuine defects in the initial shared UI repair:
an optimistic save rollback could reset a failed draft, and an earlier
send could replace a newer draft choice. Recovered delivery also left a
consumed choice pinned. All three have real RED→GREEN controls. The
final shared fix uses an explicit nullable override, a choice
generation, and one attempt-bound settlement owner for normal and
recovered success, with no value-equality inference or durable payload
changes.
- Exact shared UI source now passes all 145 composer tests (2.88s), the
original desktop/mobile planning browser case (15.5s; assertions and
timeouts unchanged), token gates, and the canonical UI typecheck. The
test harness uses the full persisted `IssueWorkMode` type, including
`skill_test`, to fix the observed CI typing failure.
- These UI results are reused from the identical shared source in #20418
through `ba1f85f18085dd0e84c594ed0cb68d5bf817c93c`. The unchanged
planning browser pass is on the explicit-state model before the later
settlement-helper extraction. The final receipt/resend controls, full
composer suite, typecheck, and token gates include that extraction. They
do not claim a local browser run on this entire combined branch. Fresh
required CI and review qualify this PR head.
- Previous-release state: existing recipe, v3 prepared reference, and
journal shapes carry forward unchanged. A saved partial operation
resumes with the same IDs. Migration does not certify remote file
recovery or provider termination. Live provider recovery and hosted
migration remain separate acceptance steps.

## Risks

The change admits one more source class into an existing preservation
path. An incorrect proof could publish incomplete file authority. The
canonical recipe validation, full authenticated restore, native
finalization locks, source reinspection, and publication checks remain
mandatory. Pending remote data stays pending and retains its original
allocation and reference. The shared UI repair updates only an unchanged
draft mode; its regression checks preserve a local user choice.

## Model Used

OpenAI Codex, GPT-6, with reasoning, repository tools, and code
execution. The runtime does not expose a deployment suffix or
context-window size.

## 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
- [ ] All Paperclip CI gates are green
- [ ] 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>
2026-11-09 19:23:31 -06:00
..