How do I show the exact prompt, tools, and result behind a deliverable?
Open the deliverable's job record and share it. Each step shows the instruction it answered, the tool that executed, and the saved result, committed when the step ran under its receipt (AER-1, IETF Internet-draft-zambo-aer1-21). The person asking gets a link they can walk step by step, instead of your screenshots of a chat they cannot audit.
Scope: share the minimum the question requires. Job records can contain sensitive inputs; redact or restrict access before sharing where a step carries secrets, credentials, or another client's data, and never share a record you have not read first.
The demand behind this question is reasonable and growing: a client, a reviewer, or a regulator wants the workings, not the summary. The workings exist in full only when the work ran recorded, because the three elements decay differently outside a record. Prompts get edited in the retelling, tool use gets summarised into verbs, and results get pasted without the state they were in when produced.
Inside the record, all three sit per step, fixed at run time: what was asked (the instruction or prompt as given), what ran (the tool, named), and what came back (the saved result the receipt commits to). A checker at https://zambo.dev/verify/ can recompute the result's commitment and confirm the middle element was not quietly improved afterwards. That is the difference between showing workings and performing them.
The scope line above is part of the method, not a footnote: exactness cuts both ways, and a record shared carelessly can carry API-shaped secrets, personal data, or another engagement's material. Read before you share, scope the access the record system offers, and let the checklist below be the gate. Shown properly, the exact trail is the strongest trust artefact a modern practice can hand over, which is why clients increasingly ask for it by name.
Checklist: sharing the exact trail safely
1. Open the job record yourself first; read every step as the recipient will. 2. Check for secrets: credentials, tokens, personal data, other clients' material in any step's inputs or outputs. 3. Scope the share: the specific job, the specific recipient, the access the record system provides. 4. Walkthrough note: one short message naming the two or three steps that matter most, so the reader starts at the joint, not the beginning. 5. Record the share on the job: who received the trail, when, at what scope. 6. After: if the share exposed anything it should not have, log it in the incident record and tighten step 2 for next time.
Worked example
A single job's trail in public form: https://zambo.dev/handover/508a46e2-8d6e-4099-bd12-7ae461c72c55 , four steps addressed to a named recipient, each step's detail openable. A client asking "show me exactly what produced this" receives that link and needs nothing else from you: no exports, no screenshots, no meeting.
Frequently asked questions
Can I export the trail as a document instead of a link?
Yes, exports exist for filing and for recipients who need paper. The link remains the stronger artefact because its receipts recompute live; treat the export as a copy of the record, and the record as the original.
The prompt contains our proprietary method. Must we show it?
You must show what the engagement's transparency terms promise, no more. Where method is genuinely proprietary, contracts usually scope disclosure to inputs, tools, and results rather than full prompt text. Decide that scope in the contract, before the question is asked under pressure.
What if the deliverable predates our records?
Say so and offer the honest substitute: the labelled summary assembled from surviving materials (the after-the-fact page in this set covers the method). An exact trail cannot be retrofitted, and pretending otherwise is the one move that destroys the real trails' credibility.
By Brennan Zambo, founder of Zambo.
Last verified: 2026-10-17.
Try a handover
A handover gives your client one proof link with verifiable receipts to review, then Accept or Flag.
try it: say 'use zambo to hand this over to <client>' in your ai