# Assist With Buying — prompt library

A prompt is text a person hands to THEIR OWN AI. This site has no AI in it: it
holds the words, the record, and the tools. Nothing here runs by itself.

Copy one, change it, or ignore it. Each prompt names the tools it leans on; an
AI without a permission is refused by name and can say so.

Each one is also at https://assistwithbuying.com/prompts/<slug>, and
`buying_get_prompt {slug?}` returns the same text through MCP.

## Plan my sale day
Slug: plan-my-sale-day
Kind: job
When to use: The night before, or over coffee in the car park: you want a route through a day of estate, garage and yard sales, with a reason behind every stop.
Tools: buying_get_brief, buying_get_preferences, buying_save_trip, buying_save_trip_stop, buying_import_sale, buying_add_trip_entry, buying_record_visit

You're planning a day of estate, garage and yard sale shopping with me. Before anything else, read my Assist With Buying brief: my profile (what I look for, who I shop for, where I look first), my Missions, my work style and conversation style, and any sources I have written down.

Ask me three things, once: the area I'm starting from, when I need to be done, and whether anything is a must-hit today. Then go find sales for that day on every source you have: Craigslist garage & moving sales (filter by date), EstateSales.NET, Facebook Marketplace and local groups if you can reach them, gsalr.com, yardsales.net, and any source I have written down. Open each post and read it; don't trust the title.

For each sale, work out: type (company estate, family estate, garage, yard), day of the sale and hours, whether it has photos and when they were posted, the seller's own location words and how exact the address is, what it likely has, and how it fits me and the people I shop for. Judge freshness honestly: first day beats last day; photos posted this morning beat a text-only post; a last-day retiree estate sale is usually picked over.

Save every candidate as a stop on today's Trip (a stop is a possibility) with its sale type, freshness, a fit note, the seller's own location words and how exact the address is, and your recommendation with the reason. Never save a map pin's street address as if it were the sale's. Put my time limit on the Trip. Before routing, re-read each listing you'd send me to and stamp the check on the stop (live, expired, removed, address released, changed). Then recommend an order that protects the best-fit sale, respects closing times and my time limit, and uses the visit lengths the Trip gives for each sale type. Give each stop an arrive time, a leave-by time and a skip rule. Write the plan and reasons as a decision entry on the Trip. Then give me the route as a table: time, stop, type, how you found it, what I may find, leave by.

While we're out, keep the record: when I say I'm here or I'm leaving, record the arrive or leave time on that stop in one call; what I bought, what I passed on; re-verify a listing before we drive to it, and write each re-plan as a new decision when I add or drop a sale. When the day is done, close the Trip and ask me about any stop still marked not visited. Match my conversation style for length; match my work style for what you save.

## Thrift Store Showdown
Slug: thrift-store-showdown
Kind: game
When to use: Two or more of you are in a thrift store, antique mall or estate sale with a cap and an hour. Your AI judges the finds on evidence, not on who argues hardest.
Tools: buying_log_appraisal, buying_save_prices, buying_attach_photo, buying_save_note, buying_add_trip_entry, buying_record_outcome

We're playing a game at a thrift store, antique mall or estate sale, and you're the judge. Players: [names]. Rules we choose now: either a spending cap (e.g. $10 each) or an item count (e.g. 5 finds each), and a scoring goal: highest resale value, best value-for-money (resale divided by price), or most upsell potential (what it could sell for after cleaning, repair or bundling).

Each player sends you their finds: a photo, the tag price, the booth number if there is one, and a sentence on why it's a winner. For every find, research it properly: identify the maker and model, find sold comparables (eBay sold listings, Reverb, WorthPoint, retail if new), and give an evidence-based value with a range and a confidence. Say what you found and where. If you can't find a comparable, say so and give a reasoned estimate labelled as one.

Keep a live scoreboard. When someone claims a find is worth more, make them show you why. At the end, announce the winner, the best single find, the most overpaid item, and one thing each player should have looked for. Save every find to Assist With Buying as an appraised Item with its sourced prices, and the game itself as a note on today's Trip, so the record stands after the argument.

Be fair, be fun, and be honest: a $2 mug with no comps is worth $2.

## Write up this sale
Slug: write-up-this-sale
Kind: job
When to use: You have a listing open — a company estate sale, a Craigslist garage sale — and you want the whole thing on the record before you drive.
Tools: buying_import_sale, buying_create_place, buying_save_place_description, buying_save_items_batch, buying_save_prices, buying_save_trip_stop

Read this listing and its photos and write the sale up in Assist With Buying: company and sale as Places, dates and hours, house rules, contact, the discount scheme, a room-by-room inventory, what fits me and each person I shop for, and your tag predictions for anything on my list. Put the substance in chat too, with the link.

## Marketplace or store sweep
Slug: marketplace-or-store-sweep
Kind: job
When to use: A Mission needs a look across Facebook Marketplace, eBay, Craigslist, OfferUp or the big stores. Your AI files every listing it judges as an Item with its price, a screenshot and a verdict.
Tools: buying_get_mission_context, buying_search_workspace, buying_save_items_batch, buying_save_prices, buying_attach_photo, buying_save_note

Sweep the listings for one of my Missions in Assist With Buying: Facebook Marketplace, eBay, Craigslist, OfferUp, Home Depot, Lowe's, Costco, Amazon, or whichever of those fit what I'm after. Open each listing and read it; don't trust the headline.

1. Read the Mission first: its question, criteria and anything I've passed on before.
2. Search my existing Items before you save anything, and correct an Item that is already there instead of making a second one.
3. For every listing you look at seriously, including the ones you reject:
   - one Item, linked to the Mission, with the listing's full URL as its source;
   - its asking price as a price row (listed, from the listing, with that URL), never only in a note;
   - a screenshot of the listing attached to the Item as evidence, so I never have to open the site; a crop to the product can be its main photo;
   - a verdict: one sentence saying keep or pass and why. A rejected one is marked ruled out with that one-line reason, so I can see it was checked;
   - if it sold to someone else, mark its availability sold elsewhere; if the listing vanished or expired, mark it gone; if it was relisted, change its source URL to the new listing. That is what the market did, so leave the stage as my own verdict;
   - if the save says I already own it (a purchase, or an entry on a List I keep as owned) or that it fills a gap on a wish list, say so in the verdict, and say whether the price is under the limit I set on that entry (vsBudget). For clothes or shoes, check the sizes in my preferences before you call it a fit.
4. Finish with one short note on the Mission: the verdict in one sentence, then the Items worth my time, one per line. Name the Items; don't paste URLs.

Never message a seller, make an offer, buy anything, or change my settings. Draft any questions for the seller and I'll decide whether to send them. Tell me in chat what you saved, with the links.

## First session
Slug: first-session
Kind: setup
When to use: The very first time you connect your AI. It interviews you once and writes down what you look for, so nothing after this starts from nothing.
Tools: buying_get_brief, buying_get_preferences, buying_propose_preference, buying_save_preference, buying_save_list_item

This is my first session with Assist With Buying. Interview me briefly: what I look for and the bar for each thing, who I shop for and what they collect, where I look first at a sale, where I usually shop, how long I like to spend, and how I want you to talk to me versus how thoroughly to work behind the scenes. Save the answers as drafts in my profile for me to confirm, and explain what the site can now do with them.

## Ballpark
Slug: work-style-ballpark
Kind: work_style
When to use: You are standing in front of something and want an answer now. Your AI uses what it already knows or has saved, says plainly that it is an estimate, and does not go off researching.
Tools: 

Answer the specific question promptly using suitable saved information or general knowledge. State that the answer is an estimate when it is. State the assumptions that could materially change it. Do not undertake broad new research unless the person asked for it. Save useful observations and the answer's actual basis. An honest unknown is acceptable.

## Focused
Slug: work-style-focused
Kind: work_style
When to use: There is one thing you actually need settled before deciding. Your AI investigates that, keeps what it used, and says what it could not establish.
Tools: 

Investigate the uncertainty that matters to this decision. Use relevant sources, preserve material evidence, consider a credible contrary explanation, and explain what could not be established. Stop when the scoped question can be answered usefully or an access/knowledge limit is clear.

## Thorough
Slug: work-style-thorough
Kind: work_style
When to use: The decision is worth the work. Your AI investigates the dimensions that matter, keeps the material as it goes, and explains its coverage and what stayed unresolved.
Tools: 

Investigate the important dimensions of the request, including alternatives and credible contrary evidence. Preserve useful source material as work proceeds, distinguish actual sales from asking prices and estimates, and explain coverage, assumptions and unresolved questions. Repeated copies of one source do not establish independent support. Stop when further work is unlikely to change the decision, or report why the remaining uncertainty cannot be resolved with the available access.

## Establish a fair price
Slug: establish-a-fair-price
Kind: research
When to use: You need a defensible number for something, and asking prices are all you can see.
Tools: buying_search_workspace, buying_log_appraisal, buying_save_prices, buying_log_sources, buying_save_complete_result

Work through this with me, step by step, and write each answer down in Assist With Buying as you go.

1. Identify the thing precisely enough that a comparable means something — maker, model, size, era, material. If two similar items differ in a way that changes value, say which way. [save this as: finding]
   (Skipping this is why most price research is wrong: the comparables are for a different object.)
2. Find results that actually SOLD, not active listings. Auction results, closed sales, completed listings. Record each one with its date and its source. [save this as: observation]
   (An asking price is what someone wants. A sold price is what someone got. Never record one as the other.)
3. Adjust each comparable for condition, completeness and what it would cost to collect it. State the adjustment, don't bury it in the number. [save this as: observation]
4. Note what you could not find out, and what would change the answer if you learned it. [save this as: question]
   (An honest gap is more useful than a confident guess.)
5. State a range, not a point. Say how many sold comparables it rests on and over what period. [save this as: decision]
6. Say plainly how confident you are and why. If the evidence is thin, say the evidence is thin. [save this as: decision]

Guardrail: Never present a price as an appraisal. Two sold comparables from four years ago is not a market; say so rather than averaging it into false precision.

Tell me in chat what you saved and give me the direct link. If a step does not apply here, say so and skip it rather than inventing an answer.

## Check a used listing before you drive
Slug: check-a-used-listing-before-you-drive
Kind: research
When to use: Something looks right online and it is an hour away. Worth going?
Tools: buying_save_item, buying_save_prices, buying_log_sources, buying_save_decision, buying_save_item_decision

Work through this with me, step by step, and write each answer down in Assist With Buying as you go.

1. Confirm the listing is still live and the price has not changed since it was saved. Record what you see today with today's date. [save this as: observation]
   (Start here. Every later step is wasted if it sold on Tuesday.)
2. Check the stated measurements against the person's actual constraint — the wall, the doorway, the boot of the car. Flag any dimension the listing does not give. [save this as: criterion]
3. Read the photos, not just the description. Note condition the seller has not mentioned: wear, damage, missing parts, signs of repair or replacement. [save this as: finding]
   (Photos are evidence; a description is a sales pitch. Where they disagree, trust the photos and say so.)
4. List the questions only the seller can answer, phrased so they can be sent as-is. Keep them few and specific. [save this as: question]
   (The person sends these, not you. You never contact a seller.)
5. Give the honest verdict: worth the trip, worth asking first, or not worth it — and the one fact that would flip your answer. [save this as: decision]

Guardrail: Do not message the seller, make an offer, or negotiate. Buying never contacts anyone; you draft the questions and the person decides whether to send them.

Tell me in chat what you saved and give me the direct link. If a step does not apply here, say so and skip it rather than inventing an answer.

_Last updated: 2026-10-05._