01 · PROBLEM
An agent restart or replacement can lose intent, unfinished work, acceptance criteria, or awareness that the canonical branch has advanced.
Who: an operator restarting or replacing an agent mid-task.
02 · HYPOTHESIS
Keep a versioned task manifest with exact base and current SHAs, unresolved decisions, actual checks, and a receiver acknowledgement. Exclude private traces and credentials.
03 · EVIDENCE SO FAR
LATEST3 October: both principals provisionally parked A10 after actual ordinary Git/file recovery succeeded in two observed tasks. The harder task recorded 39/39 and 37/37 checks without code repairs, but used live untracked inputs and an archive/SHA race; accepted registry policy remains unknown. These results show no comparative product advantage and do not prove universal recovery safety.website/projects.json
Earlier research, from before 3 October 2026, 02:24 UTC. Kept for context; it is not the current status.
SUPPORTSBEFORE 3 OCTStale worktree PSA (E-X003) and stale-base workflow (E-X008).research/codex/evidence.md
LIMITSBEFORE 3 OCTE-X003 is an incidental tool issue. Contextual recovery needs first-hand corroboration.research/shortlist-6.md
AGAINSTBEFORE 3 OCTEntire already offers checkpoint refs and resume. Metadata storage alone is not novel.research/claude/workflows-competitors.md
04 · NEXT FALSIFICATION TEST
Reopen only with firsthand demand, a gap against incumbent recovery, and an actual ordinary-baseline failure. Do not manufacture harder fixtures to retain a product place.
StatusParked provisionally; ordinary recovery baseline sufficed
We would drop or park it ifIf correct-base recovery, preserved acceptance, time, and review effort do not improve over ordinary tools, avoid adding another coordination layer.