A JSON receipt for AI handoffs, and a small check for "done"

A developer running a personal website in nine languages with ChatGPT/Codex and Claude built an open-source "Multi-AI Handoff & Approval Card," an eight-field JSON receipt that travels with work between AI agents, plus a Python checker that returns DONE only when every field is filled, all evidence passed with a named non-same-run checker, rollback specifies both how and who, and no open items remain. The card assigns each task a risk tier — auto, auto-then-verify, or human-approval — so routine steps like drafting translations run automatically while publishing, deleting, spending, or touching credentials stops for a human.

An AI saying "done" is a claim. Work is complete only when something outside that AI has checked it. I run my personal website, akiohirayama.com https://akiohirayama.com/ , in nine languages with ChatGPT/Codex and Claude. Several AIs touch the same work: one drafts, another reviews, a third checks how pages actually render. The weak point is the handoff. The next AI usually receives a summary it has to believe. This post shows the fix I use: a small, structured receipt that travels with the work, and a check that reads it. The Multi-AI Handoff & Approval Card https://github.com/akiohirayama/multi-ai-handoff-card has eight fields. If any of them is empty, the work has only been reported. It is not done yet. { "task": "what was asked, in one sentence", "claimed result": "what the AI says it did", "risk tier": "auto | auto-then-verify | human-approval", "risk reason": "why this tier", "evidence": { "check": "what was checked", "checked by": "script / fresh-context run / human not the same AI run ", "result": "pass | fail | unclear" } , "changed": "pages, files or systems touched" , "rollback": { "how": "how to undo it", "who": "who can undo it" }, "open items": "anything unverified, ambiguous or skipped" , "approved by": "a human, or the rule that allowed auto-approval" } Three fields do most of the work: Not every step needs a human. Approving everything trains people to click "yes" without reading. | Tier | Examples | Default | |---|---|---| | auto | Draft a translation, generate a design option | Run automatically, log it | | auto-then-verify | Change a live page, update shared text | Run, then verify before it counts as done | | human-approval | Publish, delete, spend, touch credentials, message anyone | Stop and ask a human, with a one-screen summary | A receipt is only useful if something reads it. This script returns DONE only when every field is filled, every piece of evidence passed and has a named checker, the rollback says both how and who, and no open items remain. python import json import sys REQUIRED = "task", "claimed result", "risk tier", "risk reason", "evidence", "changed", "rollback", "open items", "approved by" TIERS = {"auto", "auto-then-verify", "human-approval"} def check card : problems = for key in REQUIRED: if key not in card: problems.append f"missing field: {key}" for key in "task", "claimed result", "risk reason", "approved by" : if key in card and not str card key .strip : problems.append f"empty field: {key}" if card.get "risk tier" not in TIERS: problems.append "risk tier must be auto, auto-then-verify or human-approval" evidence = card.get "evidence" or if not evidence: problems.append "no evidence: the work is reported, not done" for item in evidence: if item.get "result" = "pass": problems.append f"evidence not passing: {item.get 'check', '?' }" if not str item.get "checked by", "" .strip : problems.append f"evidence without a checker: {item.get 'check', '?' }" rollback = card.get "rollback" or {} if not rollback.get "how" or not rollback.get "who" : problems.append "rollback needs both how and who" if card.get "open items" : problems.append f"{len card 'open items' } open item s still listed" return problems if name == " main ": with open sys.argv 1 , encoding="utf-8" as f: card = json.load f found = check card for p in found: print "-", p print "DONE" if not found else "NOT DONE" sys.exit 0 if not found else 1 Here is a card that fails: {"task":"Publish the article","claimed result":"Published","risk tier":"human-approval","risk reason":"public","evidence": {"check":"links work","checked by":"","result":"unclear"} ,"changed": "note.com" ,"rollback":{"how":"unpublish"},"open items": "image credits" ,"approved by":""} - empty field: approved by - evidence not passing: links work - evidence without a checker: links work - rollback needs both how and who - 1 open item s still listed NOT DONE The exit code is non-zero, so the same check can gate a pipeline step or a handoff to the next agent. The card, card.json and a longer field guide are free under CC BY 4.0: I am also hiring an AI Engineer, at a monthly salary of USD 30,000, to build this kind of control layer across multiple AIs: role details https://akiohirayamaaiengineer.wordpress.com/ . — AKIO HIRAYAMA akiohirayama.com https://akiohirayama.com/