The scorer
Paste your handoff note or session summary. It is scored in this page; nothing is sent anywhere.
The same scorer ships in the free zambo-delivery package, for notes that should not go near a web page: pip install zambo-delivery, then zambo-delivery score note.txt (or pipe the note on stdin; --json for tooling). The npm package runs the same checks: npx zambo-delivery score note.txt.
The seven things a continuation needs
Goal (15). What this job is, in a sentence or two: what outcome, for whom. Without it the pickup optimizes for the wrong finish line.
Done (15). What is already completed. Without it the next AI redoes finished work and bills you twice for it.
Decisions (15). What was decided, and why. The most quietly expensive gap: a decision with no reason attached gets relitigated, or reversed by an AI that never knew it was made.
Next steps (20). The heaviest weight, because it is the section a pickup acts on first. A note with no ordered next steps is an archive, not a continuation: the receiving AI starts by re-planning the job you already planned.
Blockers (10). Risks, unknowns, and waiting-on items, or a plain statement that there are none. Silence here reads as "nothing is blocked" right up until the pickup hits the wall the first session already knew about.
Evidence (15). Links, check results, or receipts behind the done list, so the receiver can re-check a claim instead of trusting it. A summary asserts; evidence points.
Artifacts (10). The files and outputs that exist, and where they are. "The report is done" with no location costs the pickup its first hour.
The limit a score cannot cross
A high score means the note carries what a continuation needs. It does not mean the work in the note is true. The scorer reads prose; it cannot recompute whether the export actually ran, whether the tests passed, or whether the figure in the summary came from the run it describes. Every summary-only handover has this hole, including a perfectly structured one: the receiving side is trusting the writer.
That is the gap a real handover closes. On Zambo, the wrap-up attaches the job's step record to the handover link, and each call in the record carries a verifiable AER-1 receipt: the receiver recomputes the receipt and the saved result either matches what was fixed when the call ran, or it does not. The note still matters (it is where the goal, decisions, and next steps live, which is why this scorer exists), but it stops being the only evidence. A receipt proves the saved result was not changed after it was saved; it never proves the work was good. Judgment stays human. What the record settles is what happened.
Practical order of operations: score the note, fill the gaps it names, then create the handover so the note travels with its record. The recipient's pickup is always free, and checking any receipt is free, so the receiving side never needs a plan to continue or to verify.
Questions, answered
What makes a good AI handoff note?
Seven things: the goal; what is done; the decisions and why; the next steps in order; the blockers or a plain "none"; evidence for the done list that the receiver can re-check; and the artifacts with their locations. Next steps carry the most weight (20 of 100) because they are what a pickup acts on first.
Why do AI session summaries fail when I switch tools?
A summary describes the session; a handover continues it. Summaries usually skip the decisions (so the next AI re-asks or reverses them), skip ordered next steps (so the pickup re-plans instead of working), and carry no evidence, so every claim in them has to be trusted. Scoring the note before the switch shows the gaps while you can still fill them.
What score is good enough to hand over?
80 or above: hand over. 50 to 79: a pickup can start, but the named gaps are where it will stall, and they take minutes to fix now. Below 50: it is a summary, not a handover; fill the gaps first, because after the switch the missing context costs a re-brief, redone work, or a reversed decision.
Does a high score mean the work in the note is true?
No. The score measures whether the note carries what a continuation needs; it cannot verify the steps actually ran. That takes a record under the note: a handover with the step record and a verifiable receipt per call, which the receiver recomputes. A receipt proves the saved result was not changed after saving, never that the work was good.
What does it cost to work this way?
Scoring is free, here and in the zambo-delivery package. Creating the handover runs on Zambo: the free tier is 20 calls + 1 wrap-up free every day, and picking up a job in another AI is always free. A Day Pass is $0.99 one time and covers 24 hours. The Zambo Pass is $29 for 30 days, and $19 for 30 days for $ZAMBO holders. A Lifetime Pass is $250, one payment. Full pricing: https://zambo.dev/pricing