| name | steward-handoff |
| description | Prepare, receive, or recover a same-machine central-agent custody handoff while preserving the complete plan and acceptance boundary. Use when the owner requests a central transfer/recovery or says to accept a handoff from a stewardship context id. |
Steward handoff {#root}
@fact:HANDOFF-CENTRAL-ONLY This skill transfers owner-facing central custody,
not work to a subagent. It never delegates acceptance. @status:impl/done
@fact:HANDOFF-CONTEXT-COMMAND Recognise ACCEPT HANDOFF FROM <context-id> / ПРИМИ ХЭНДОФФ ИЗ <context-id>. Resolve exactly one
unreceipted offered bundle in that context; if several exist, require an exact
handoff id. Read the stored HANDOFF.md—never ask the owner to paste it.
@status:impl/done
Select the operation {#operation}
- @fact:HANDOFF-PREPARE Prepare when the current session holds custody and
the owner wants another central session/model/harness to continue.
@status:impl/done
- @fact:HANDOFF-RECEIVE Receive when an unreceipted offer targets this
session. @status:impl/done
- @fact:HANDOFF-RECOVER Recover only when the holder is unavailable and the
owner explicitly authorizes takeover of the named context. @status:impl/done
Prepare {#prepare}
@fact:HANDOFF-PREPARE-PROCEDURE Follow
spec://org.vibevm.world/multi-user-planning/flows/multi-user-planning/custody-and-handoff#offer:
verify the tree; backscan mandate/plan/evidence/artifacts; write the immutable
offer, comprehensive HANDOFF.md, and short receive-by-context pointer; refresh
and hash handoff, plan and selected goal; set custody to offering; then become
repository- and plan-read-only. A summary without the created file and fence is
not a handoff. @status:impl/done
Receive {#receive}
@fact:HANDOFF-RECEIVE-PROCEDURE Read all authorities and the complete bundle,
verify hashes and repository state independently, enumerate discrepancies,
write accepted, accepted-with-exceptions, or rejected receipt, and advance
custody only for an accepted receipt. Preserve all dirty/worker work as
candidate. Once holder, refresh goal, print full GOAL.md, then print the exact
GOAL-CLAUDE.txt /goal … line separately. Report restored state and wait for
the owner before execution.
@status:impl/done
Recover {#recover}
@fact:HANDOFF-RECOVER-PROCEDURE Preserve artifacts, verify holder
unavailability, write a takeover record with the owner's authorization and
observed accepted boundary, advance epoch, and treat everything after durable
project acceptance as candidate. Never delete first or use staleness alone as
authority. @status:impl/done
