Summary

TriggerGC currently only handles team context repos. Ledger repos need the same blue-green GC reclone support to recover from corrupted state (missing LFS objects, diverged history, etc.).

Current state

• Team context GC: TriggerGC() in sync.go does blue-green reclone — works well
• Ledger LFS repair: RepairMissingLFSObjects() in gitutil/lfs_repair.go is a temporary mechanism that replaces orphaned LFS pointers with empty stubs + sets allowincompletepush. This is a workaround, not a proper fix.
• The real fix: A ledger GC that does fresh clone from remote + carries over unpushed local commits (session data). This is exactly what was done manually to fix the sageox/ox ledger today.

Why ledger GC is needed

After daemon GC reclone of team contexts, locally-committed LFS objects in the ledger that were never pushed are lost. The remote's pre-receive hook then rejects all future pushes because historical commits reference missing LFS OIDs. The temporary repair (lfs_repair.go) mitigates this but:

1. Requires lfs.allowincompletepush which only bypasses the local check
2. Server-side GitLab pre-receive hook still rejects if old OIDs are in push range
3. History rewriting requires force push, which is blocked on protected branches
4. The clean approach (fresh clone + replay local data) is what GC already does for team contexts

Implementation plan

1. Extend TriggerGC (or add TriggerLedgerGC) to support ledger repos
2. Blue-green reclone: fresh clone → identify unpushed local commits → cherry-pick/replay session data → swap
3. The temporary RepairMissingLFSObjects can remain as a fallback for non-GC scenarios
4. Add ox doctor --gc-ledger or extend existing --gc flag

Related

• PR fix(lfs): auto-repair missing LFS pointers that block ledger push #241 — temporary LFS repair mechanism
• beads bd-7m3 — original issue for missing LFS pointers

Co-Authored-By: SageOx ox@sageox.ai