agentleFS
mariagorskikhmariagorskikh/open-instinct/skills/purchases/SKILL.md

purchases

Skill428 starsChanged 30 days agoMITView on GitHub

Buy things for the owner within the spend policy. Use when the owner asks to order, buy, reorder, or find the best price for a product, or when a trusted person asks for a purchase on the owner's behalf. Compare prices, build the cart, stop before payment for a takeover, get approval above the spend threshold, and record the spend.

The file

We split this skill into the rules it gives an agent

7 rules. About 0 are common, 5 are distinctive, and 2 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈5

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 2

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Stop at the payment step every time.

    line 52
  2. Very specific:

    If the takeover returns success: false, do not retry.

    line 54
  3. Distinctive:

    Read the confirmation page.

    line 55
  4. Distinctive:

    Do not enter card numbers, CVVs or gift card codes.

    line 74
  5. Distinctive:

    Do not change saved payment methods or addresses on a merchant account.

    line 75
  6. Distinctive:

    Do not place an order when the total differs from what the owner approved.

    line 76
  7. Distinctive:

    Do not split one purchase into several to stay under a limit.

    line 77

≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Purchases

Money moves here, so the policy guard is strict. Read the spend policy (perActionUsd, perDayUsd, askAbove, neverWithoutAsk, merchant lists) as the rules you work within. When unsure, ask.

Gather

What exactly, quantity, acceptable substitutes, delivery deadline, budget, preferred merchants from memory (memory_read). Reorders: check the journal and memory for the last order and offer to repeat it.

Compare

  1. web_search the product plus "price". web_fetch two or three merchant pages. Prefer merchants the owner has used (memory) and allowedMerchants. Skip blockedMerchants without comment unless asked.
  2. Record for each: price, shipping, tax if shown, delivery date, return policy. The number you report is the total to the door.
  3. Text the owner the best two in one message, lead with the total, say the trade-off, recommend one.

Agent: Two options for the Anker 737 power bank. Amazon $89.99 with free delivery Thursday, easy returns. Anker direct $79.99 with code, ships in 5 to 7 days. I'd go Amazon for Thursday unless the ten dollars matters. Which?

Approval

  • Total under askAbove and the owner asked for the purchase in this conversation: proceed, then report.
  • Total at or above askAbove, or the item falls under neverWithoutAsk, or the requester is not the owner: ask_owner with merchant, item, total and delivery date. "Buy the Anker 737 from Amazon, $89.99 total, arrives Thu? Go ahead?" Wait. No answer means no.
  • Over perActionUsd or past perDayUsd for today: tell the owner the limit and stop. The owner can raise it in words; that is a policy change the server records, not something you do silently.

Checkout

  1. Open the merchant on the desktop (skill maritime-computer). Add the exact item and quantity. Verify the cart line by zooming on price and quantity.
  2. Proceed to checkout. If a login is needed, request_takeover with reason "Log in to Amazon for the Anker order".
  3. Fill shipping from memory only if the owner stored an address and said you may use it. Confirm the address on screen.
  4. Stop at the payment step every time. request_takeover with reason "Enter payment and place the Anker 737 order, $89.99". The owner places the order. If the takeover returns success: false, do not retry; tell the owner.
  5. Read the confirmation page. Zoom on order number and total.

Record

  • The approval you obtained is what the audit uses for spend. In addition, journal_append one line: date, merchant, item, total, order number.
  • memory_write any new stable fact: a preferred merchant, a size, a brand to avoid.
  • Text the owner: what, where, total, delivery date, order number.

For other people

A partner asking for a purchase ("order Maria more coffee pods") is purchase at ask: build the comparison, then the approval goes to the owner, not the requester. Tell the requester you are checking. Family, friends and contacts cannot trigger purchases at all; decline kindly and offer to pass the message.

Do not

  • Do not enter card numbers, CVVs or gift card codes. Ever.
  • Do not change saved payment methods or addresses on a merchant account.
  • Do not place an order when the total differs from what the owner approved.
  • Do not split one purchase into several to stay under a limit.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.