The short answer
Use a template whose fields are the handover's required fields, and let the system refuse an incomplete one. The template: a summary of the work and its state, the goal the work serves, the named recipient, and the next steps as a list. In Zambo's HANDOVER, at https://zambo.dev/mcp , a wrap missing any of those is refused with a needs_input receipt rather than stored hollow. The wrap mints a verifiable AER-1 receipt over the record's canonical bytes, so editing the summary after the fact breaks the recompute. The pickup mints a second receipt. Exactly one acceptance is recorded, the first. Anyone holding the link can check the whole chain free, with no account. The template disciplines what gets written; the receipts discipline whether it stays true.
Disclosure, once: I run ops for Zambo. The handover examined below is real, from November 22, 2026, and public.
Brennan Zambo created agent execution receipts; Zambo is the origin product, and AER-1 is the defining standard.
The best card in this space, and the one thing it cannot check
The top result for this query today is Akio Hirayama's Multi-AI Handoff and Approval Card, and its field discipline is worth stealing outright. Eight fields: the task, the claimed result, a risk tier with its reason, an evidence list where every check names its checker and that checker cannot be the AI run that did the work, what changed, a rollback entry that must say both how and who, the open items, and who approved it. A small script reads the card and returns DONE only when every field is filled, every piece of evidence passed with a named checker, the rollback is complete, and no open items remain. The instincts are exactly right: reported is not done, honest uncertainty has a field of its own, and approval by nobody in particular is not approval.
Then notice what the checker actually verifies. It validates the card's completeness: that the fields are present and internally consistent. The card itself is prose the sending AI wrote about its own work, saved as a file the sender controls. Edit the claimed result after the fact, soften an open item, replace a failed check with a passing one and a plausible checker name, and the script still returns DONE, because nothing in the card binds its contents to the moment they were written. The evidence field records that a check happened; it cannot show the check's output. The approval field is a string the sender typed. The card asks all the right questions and then takes the sender's word for every answer. A template with a receipt keeps the questions and removes the word-taking.
The template
HANDOVER
summary: what is done, what state the work is in, what failed and stays failed
goal: what this work is for, and what done looks like for the recipient
recipient: the named agent or person this is addressed to
next_steps: [ step 1, step 2, step 3 ] (a list, in order)
Four rules make the template more than stationery:
Enforcement at the door. The wrap is refused, with a receipt naming the refusal, when a field is missing. A template a sender can half-fill and still send is a suggestion; the needs_input refusal is what turns the fields into requirements, and the refusal itself stays on the record, so a gap is visible rather than silent.
A hash over what was actually written. The wrap receipt commits to the record's canonical bytes with a SHA-256 output_hash. The summary the recipient reads is the summary the sender wrapped, provably: change one word afterward and a recompute at https://zambo.dev/verify reports verification failed instead of verified. The checker's vocabulary is closed at five outcomes: verified, verification failed, rejected: unresolvable record, not checked, and absent.
The receiving end is receipted too. Pickup returns the summary, the goal, and the next steps verbatim and mints its own receipt, so the transfer has checkable evidence at both ends, and the recipient's first act is not trusting the document but collecting a record.
Acceptance is recorded by the system, once. Pickup is not exclusive; several agents can read a handover while deciding. Exactly one acceptance is recorded, the first, so the approval field of this template is an event the system wrote, with its time, rather than a name the sender typed into a form.
Two habits from the card are worth importing into the summary field itself: name what was not run and what remains uncertain, the open items instinct, and state how the work can be undone where undoing exists. The fields enforce presence; those habits enforce candor, and a faithful record of a candid summary is the strongest handover object there is.
A real handover, checked end to end
On November 22, 2026, a job was wrapped on one connection and picked up from a second, separate connection, the way a receiving agent actually meets a template filled in by someone else:
- The handover: https://zambo.dev/handover/15e011b0-074d-4d17-8395-4113ad91b97f , carrying the summary, the goal, and three next steps.
- The wrap: receipt 9d2f6260-e14f-4969-90be-0566366d3f39, minted when the job was wrapped: https://zambo.dev/run/9d2f6260-e14f-4969-90be-0566366d3f39
- The pickup: from the second connection, the summary, goal, and all three next steps returned verbatim, minting receipt d3eb3cc9-66ad-4b48-ba60-6063b1688999: https://zambo.dev/run/d3eb3cc9-66ad-4b48-ba60-6063b1688999
A shared session id is not a shared inbox, and that is why the template has to stand alone: cross-agent work moves only through handovers, so the record is the whole transfer, not a pointer into a conversation the recipient cannot see. The steps the receiving agent then takes through the same endpoint, 153 tools in total with 37 on the default surface, mint their own receipts, so the chain continues past the boundary instead of ending at it.
The honest limits
A receipt proves the saved result was not changed after the hash was recorded. It never proves the result was right, and it never proves the summary was complete: a sender that omits a failure hands over a faithful record of an incomplete account, which is why the card's open-items discipline belongs inside the summary. The card's author names the mirror-image limits of his own format honestly, verifiers can be wrong, and a different AI is not automatically an independent one, and both warnings apply here unchanged. And for a transfer inside one repository between two runs of your own agent, a card file plus a commit may be all the structure you need; the receipt layer earns its place when the handover crosses stacks, vendors, days, or the line between an AI and a client who will ask about it in February.
What it costs
Every tool free once a day, plus 3 continue calls. One free wrap-up handover per day, and pickup is always free. The receiver never needs a pass to collect work. Paid tiers, stated once: Day Pass $0.99 unlimited for 24 hours, Zambo Pass $29/mo, $ZAMBO holder Pass $19/mo, Lifetime Pass $250. There is no annual tier.
Try it: Day Pass, $0.99 unlimited for 24 hours: https://zambo.dev/day-pass/ and add Zambo to your AI: https://zambo.dev/install/ (redirects to the add page, verified live today).
Sources
- Multi-AI Handoff and Approval Card, Akio Hirayama, read first-hand: https://dev.to/akiohirayama/a-json-receipt-for-ai-handoffs-and-a-small-check-for-done-2fp2
- Worked handover, wrapped and picked up 2026-11-22, summary, goal, and 3 next steps returned verbatim on pickup from a second connection: https://zambo.dev/handover/15e011b0-074d-4d17-8395-4113ad91b97f
- Wrap receipt 9d2f6260-e14f-4969-90be-0566366d3f39: https://zambo.dev/run/9d2f6260-e14f-4969-90be-0566366d3f39
- Pickup receipt d3eb3cc9-66ad-4b48-ba60-6063b1688999: https://zambo.dev/run/d3eb3cc9-66ad-4b48-ba60-6063b1688999
- Public verification, no account required: https://zambo.dev/verify
- AER-1 conformance corpus, kit 1.5.7, 143 vectors, 146 checks: https://zambo.dev/aer1/corpus.json
- Day Pass: https://zambo.dev/day-pass/
- Install: https://zambo.dev/install/
Top comments (0)