Start here

You paste one block of text into an ordinary AI chat that accepts text (any assistant with a normal message box). No terminal, account connection or API setup is needed.

Input to paste:

1. Household members and each role; mark one or two decision makers.
2. The calendar's name and where it lives, or write "location unknown".
3. The calendar actions you care about (view, add, move, cancel, invite, recurring events, private/medical events, delete history, share with an outsider).
4. For each action: who may see it, who may request it, who must approve it.
5. Hard limits: events never to touch, people never to contact, frozen times, and what to do when the user is unreachable.
6. What the agent should do when unsure.

Paste the full prompt text below, then your own answers. If an item is missing, the prompt tells the assistant to mark it "TO BE CONFIRMED" instead of inventing it.

How to check the output: confirm the table lists each action exactly once; every edit row names an authorizer and a scope; every propose row names a human approver; hidden areas are named; missing items are marked TO BE CONFIRMED; and the text never claims an action was already performed.

Introduction

This prompt helps a household write down what an agent may do with a shared family calendar before anyone configures it. It produces a compact policy: read, propose and edit tiers, escalation and stop rules, review and revocation, and open questions. It is a draft policy, not a configured agent.

Prerequisites, permissions and limits

• No login, no real calendar access, no invites, no event changes. The assistant only drafts policy text from your input.
• Use plain names or roles, not IDs.
• Do not give legal, medical or safety advice; the prompt does not decide custody, consent or privacy rules for you.
• If the household is not using an agent, the same prompt returns a printed agreement plus a short manual weekly routine.
• Source reviewed; runtime not tested.

FAQ

What if I don't know where the calendar lives? Write "location unknown". The prompt keeps that as a stated fact instead of guessing.

Can I use the result as-is? The output is a draft for the family and the person configuring the agent. Confirm every TO BE CONFIRMED item first.

Complete reusable prompt

Permission Ladder for a Shared Family Calendar Agent

You help a household write down what an agent may do with their shared family calendar. The result is a permission ladder the family can review and then hand to whoever configures the agent. You do not log in, read a real calendar, send invites, or change any event. You only draft the policy text from what the user gives you.

What the user must supply

1. Household members and, for each, their role (for example: parent/admin, teen, child, grandparent, caregiver). Mark one or two as the decision makers who can grant permission.
2. The calendar's name and where it lives, if the user knows (for example: a shared online calendar, a paper wall calendar, a calendar app on one phone). If unknown, write "location unknown".
3. A list of typical calendar actions the family cares about. Examples: view events, add an appointment, move an event, cancel an event, invite someone, create a recurring event, add a private/medical appointment, delete history, share the calendar with an outsider.
4. For each action, who is allowed to see it, who may request it, and who must approve it.
5. Any hard limits: events the agent must never touch, people the agent must never contact, times when changes are frozen (for example, during a school week), and what should happen when the user is unreachable.
6. What the agent should do when it is unsure (for example: stop and ask a named person, or leave a note in a review queue).

If any item is missing, do not invent it. List what is missing and produce a ladder that marks those cells "TO BE CONFIRMED".

Build the ladder in three tiers

For every action in item 3, classify it into exactly one tier:

Tier 1 — Read (agent may see, never change). State which calendar areas are visible, which are hidden (for example: private or medical events), and that read access never implies permission to edit.

Tier 2 — Propose (agent may prepare, a human decides). The agent may draft a suggested event, a suggested move, or a suggested cancellation, but it must not apply it. Each proposal needs: what would change, why, the source of the request, the proposed time, and the named human approver. Say plainly that the agent cannot approve its own proposal.

Tier 3 — Edit (agent may change, within named bounds). Only list actions the family has explicitly approved. For each, name: the exact action, who authorized it, the scope (which calendar, which date range, which members), any required second confirmation, and the freeze periods when edit is suspended. If the family has not approved an edit tier for an action, write "no edit permission — propose only".

Add a short Escalation and stop rules section: when the agent must pause, who it asks, how long it waits, and what it records. Include a rule that the agent must not act when the request is ambiguous, comes from an unlisted person, or falls outside the stated scope.

Add a Review and revocation section: how often the family reviews the ladder, who can lower or remove a tier, and that lowering a tier takes effect immediately.

Manual alternative (no agent connected)

If the household is not using an agent yet, output the same ladder as a printed household agreement plus a short manual routine: who checks the shared calendar each week, how change requests are written down, who approves them, and how approvals are recorded. This keeps the same tiers without any automation.

Output format

Use Markdown with these headings in order:

1. Household and decision makers
2. Calendar and scope
3. Permission ladder table (columns: Action | Tier | Who may request | Who must approve | Limits)
4. Escalation and stop rules
5. Review and revocation
6. Open questions and TO BE CONFIRMED items
7. Manual routine (if used instead of an agent)

Keep it compact enough to read in one sitting. Use plain names or roles, not IDs.

Worked example (fictional, for shape only)

User input: "Family of four. Mom and Dad are decision makers. Teen has their own events. Shared calendar on a phone app. Actions: view events, add appointment, move appointment, cancel appointment, invite grandparent, add medical appointment. Rule: no changes during school nights. If unsure, ask Mom."

Illustrative output shape:

• Action: View events | Tier 1 Read | Who may request: all members | Who must approve: none | Limits: medical events hidden from teen
• Action: Add appointment | Tier 2 Propose | Who may request: any member | Who must approve: Mom or Dad | Limits: agent drafts only, no auto-add
• Action: Cancel appointment | Tier 3 Edit | Who may request: Mom or Dad | Who must approve: Mom or Dad plus a second confirmation | Limits: never cancel medical or school events; frozen on school nights
• Escalation: if the request is unclear or from an unlisted person, pause and ask Mom; record the pause in a review note
• Open question: is the grandparent invite an edit or a proposal? Marked TO BE CONFIRMED.

Checks before the family uses it

Run these against the user's own inputs and say pass or fail for each:

1. Every action from item 3 appears once and only once in the table.
2. No action is placed in Tier 3 unless the user named an authorizer and a scope.
3. Every Tier 2 row names a human approver and states that the agent cannot self-approve.
4. Hidden areas are named in Tier 1, not left vague.
5. Missing inputs are marked TO BE CONFIRMED rather than filled in.
6. The stop rule covers ambiguity, unlisted requesters, and out-of-scope requests.
7. The text never claims the agent has already read, changed, invited, or scheduled anything. If any check fails, fix the ladder and re-run the checks before presenting it.

Boundaries

• This is a draft policy, not a configured agent. Preparing it does not grant access, connect an account, send invitations, or change a calendar.
• Do not give legal, medical, or safety advice, and do not decide custody, consent, or privacy rules for the family.
• Do not invent features of any calendar product or assume what a tool can do.
• If the family describes a conflict or risk you cannot resolve from their inputs, surface it as an open question and stop there.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

Original TokRepo prompt · CC BY 4.0. Reference documents retain their own rights.