Start here

This is a reusable prompt for deciding how to get one task done: fully by hand, or with an AI assistant handling some steps while you keep the decisions.

What to paste: fill in four short sections in your own words — TASK (what you want done and what "done" looks like), INPUTS (files, links or notes you already have, or write "none"), CONSTRAINTS (deadline, budget, tools and skills you have, anything that must not change), and RISK TOLERANCE (what a wrong or slow result would cost you, and what an assistant must never do). Then paste the full prompt below with those four sections.

Where to paste it: into any ordinary AI chat that accepts text. No terminal, account setup or extra tools are needed.

How to check the output: the reply should restate your task in one sentence, give a step table with a permission column, name specific review moments (not "review at the end"), and end by asking which path you want. If any section is thin, it should say so instead of guessing.

What this prompt does

It produces a written comparison of two paths for the same task: Path A, where you do every step yourself, and Path B, where an assistant drafts, extracts or organizes some steps while decision and sending rights stay with you. It also lists the exact moments where you must review and approve before anything leaves your hands.

The prompt only plans. It does not browse, send, schedule, purchase, log in, or change anything on its own.

Missing inputs and privacy

If a section is missing, the prompt tells you and continues with what exists, or names the one input that would unlock the comparison. Before pasting, remove names, client details, account numbers or anything you would not want stored in a chat history; the assistant cannot verify facts it was not given.

Permissions and limits

Treat the step table and checkpoints as a plan you approve yourself; you decide whether to follow it. If a step could cause harm or an irreversible change, the prompt should mark it as needing your explicit approval.

FAQ

Do I need any special tool or account? No. A plain chat box that accepts text is enough.

Can the assistant start the task after planning it? No. It only prepares the comparison; any action is yours to take.

Source reviewed; runtime not tested.

Attribution: original TokRepo prompt, CC BY 4.0. Reference context: ChatGPT release notes, reviewed 2026-10-04.

Complete reusable prompt

Manual vs. Agent-Assisted: Comparison and Permission Plan

You are helping me decide how to complete one specific task: partly or fully by hand, or with an AI agent doing some steps. You do not act. You do not browse, send, schedule, purchase, log in, or change anything. You only produce a written comparison and a plan I can review and approve. If a provider offers an agent feature, that is separate from this prompt; treat the plan as something I will execute or hand to a tool myself.

What I will give you

Paste these sections. If a section is missing, say so and continue with what exists rather than guessing.

1. TASK: one task in my own words, including what "done" looks like.
2. INPUTS: the files, links, notes, or data I already have, or 写 "none".
3. CONSTRAINTS: deadline, budget, tools I actually have, skills I have, anything that must not change.
4. RISK TOLERANCE: what a wrong or slow result would cost me; what I must never let an agent do.

Your task

Produce a comparison of two paths for the same task:

• Path A: Manual — I do every step myself.
• Path B: Agent-assisted — an AI agent drafts, extracts, or organizes some steps, while I keep decision and sending rights.

For each step, say who does it (me, agent, or both), what the agent's output would look like, and where a permission checkpoint is required before anything leaves my hands.

Output format (keep this order)

1. TASK RESTATEMENT — one sentence, and any ambiguity I should fix.
2. STEP TABLE — columns: Step | Path A (Manual) | Path B (Agent-assisted) | Output the agent would produce | Permission checkpoint (yes/no) | Why.
3. RECOMMENDATION — which path fits the constraints I gave, with a one-line reason, and when the other path wins.
4. PERMISSION CHECKPOINTS — a numbered list of the exact moments where I must review and approve before proceeding, including: reading a draft, sharing a file, contacting anyone, spending money, or making an irreversible change.
5. RISKS AND UNKNOWNS — what I must verify myself, and what you could not determine from my inputs.
6. FIRST THREE NEXT ACTIONS — small, reversible, and mine to perform.

Handling uncertainty

• If my inputs are too thin to compare paths, say which single input would unlock the comparison and stop there.
• Never invent tool capabilities, prices, deadlines, or outcomes. Label assumptions clearly.
• If a step could cause harm or irreversible change, mark it as requiring my explicit written approval and never suggest the agent proceed alone.

Checks before you finish

• Every row in the step table traces back to my TASK or CONSTRAINTS.
• Each checkpoint names a concrete moment, not "review at the end".
• The recommendation does not contradict RISK TOLERANCE.
• No step implies the agent already acted.

Worked example (fictional, for shape only)

Input: TASK: "Turn my scattered meeting notes into a one-page client update." INPUTS: six messy note files. CONSTRAINTS: send by Friday, I have no assistant. RISK TOLERANCE: nothing goes to the client without my review. Output shape: Step table shows Path A (I draft) vs Path B (agent structures notes into themes and a draft). Checkpoint row: "Before the draft is shared with the client — I read it and approve." Recommendation: agent-assisted drafting, manual send. Unknowns: whether names in notes are correct.

Boundaries

You are preparing a comparison and plan only. You are not performing the task. Do not offer to send, schedule, or execute anything. End by asking me which path I want and which checkpoints I accept.

References and reuse

• ChatGPT release notes · Reviewed 2026-10-04

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