huskies/PLAN.md

Plan: Story 945 — Delete WorkItem flag soup

Goal: Fold blocked, review_hold, frozen, and mergemaster_attempted flag fields into Stage / ArchiveReason / ExecutionState variants so the typed state machine is the single source of truth.

ACs → implementation locations

• AC 1 (delete flag fields): server/src/crdt_state/types.rs (PipelineItemCrdt, WorkItem) — DONE; server/src/pipeline_state/types.rs (PipelineItem) — DONE.
• AC 2 (variants + match-based callers): New variants Stage::MergeFailureFinal, Stage::Frozen { resume_to }, Stage::ReviewHold { resume_to, reason } in pipeline_state/types.rs — DONE. Auto-assigner, watchdog, unblock/freeze/unfreeze, merge pipeline all match on Stage — DONE.
• AC 3 (no Option flag-poking): grep for set_blocked|set_review_hold|set_frozen|set_mergemaster_attempted returns no results — DONE.
• AC 4 (cargo check/clippy/test pass): run_check clean, run_tests reports 2908 passed / 0 failed — DONE.

Decisions

• Resume target for Frozen/ReviewHold: stored as a sibling resume_to: LwwRegisterCrdt<String> on PipelineItemCrdt rather than encoded into the stage string. Rejected: encoding into stage register (would require parsing variants out of strings and lose round-trip cleanness).
• Reason text for Blocked/MergeFailure/MergeFailureFinal/ReviewHold: kept on the Stage variant in memory, but the wire-form stage register only carries the canonical dir name ("merge_failure_final", "review_hold"). Reasons are reconstructed at read time from companion CRDT data (MergeJob.error) where needed. Acceptable because reason is human-text, not load-bearing for routing.
• Pre-934 7_frozen legacy migration: rewrites stage to "frozen" and sets resume_to = "backlog", restoring Stage::Frozen { resume_to: Backlog } on read. The defensive projection fallback still maps raw 7_frozen → Backlog for un-migrated reads.

Current state

All flag fields removed; all 47 modified files compile; 2908 tests pass; doc coverage clean.

What's left

• Commit changes