| name | amazon-address-book |
| description | Read and change the delivery addresses on the user's signed-in Amazon account - list the address book, correct a wrong street or postcode, add a new address, and change which one is the default ship-to. Use when the user says fix my Amazon address, add my new place, update the postcode, or make X the default delivery address. Reading is free; every change is confirmed against a before-and-after diff first, and removing an address is out of scope. |
| allowed-tools | mcp__claude-in-chrome__tabs_context_mcp, mcp__claude-in-chrome__tabs_create_mcp, mcp__claude-in-chrome__navigate, mcp__claude-in-chrome__javascript_tool, mcp__claude-in-chrome__find, mcp__claude-in-chrome__computer, mcp__claude-in-chrome__tabs_close_mcp, Bash(python3 *), Read, Edit |
Amazon address book
Two halves, with a hard line between them.
Reading the address book is cheap, safe and fetchable — no navigation, no
clicking. Do it freely.
Changing an address decides where physical parcels go. Every change gets a
before-and-after diff, the user's explicit yes, and a verifying re-read
afterwards. A silently wrong address is not an error message; it is a parcel at
the old flat.
Removing an address is out of scope. Say so and let the user do it.
Reading: fetch, never navigate
Unlike order history, this page is fully server-rendered — 426 KB with the
addresses in it. So use the read-only same-origin fetch from a tab on any
Amazon page, which cannot wander into a live order's controls.
const d = new DOMParser().parseFromString(
await fetch('/a/addresses', {credentials:'include'}).then(r => r.text()), 'text/html');
d.querySelectorAll('script,style,noscript').forEach(n => n.remove());
const clean = s => (s||'').replace(/\s+/g,' ').trim();
const blocks = [...d.querySelectorAll('[id^="ya-myab-display-address-block-"]')];
const rows = blocks.map((b, i) => {
const f = n => clean(b.querySelector('#address-ui-widgets-' + n)?.textContent);
return {
i,
name: f('FullName'),
line1: f('AddressLineOne'),
line2: f('AddressLineTwo'),
cityStatePostcode: f('CityStatePostalCode'),
country: f('Country'),
phone: f('PhoneNumber').replace(/^Phone number:\s*/i,'').replace(/[-]/g,''),
isDefault: !!b.querySelector('[id*="default-shipping"]'),
};
});
JSON.stringify(rows, null, 1)
Three things in that snippet are load-bearing:
- Scope every field to its block. The field IDs are duplicated once per
responsive layout — six addresses produce twelve nodes with
id="address-ui-widgets-FullName" — so document.querySelector('#address-ui-widgets-FullName')
returns block 0's name for every address, with no error and a real-looking
value. This is the single most dangerous bug available in this skill.
- Count
[id^="ya-myab-display-address-block-"], not .address-tile. The
first .address-tile is the Add address card, so tile count overstates the
book by exactly one.
isDefault tests for presence, not text. The default marker is an icon
with empty textContent.
CityStatePostalCode is one field (Town Name, ST 12345-6789). The phone
renders with a label and invisible directional marks; strip both.
Before changing anything
-
Read the book and show the user the address as it stands today — not as
addresses.yaml remembers it. The account is the source of truth here.
-
State the change as a diff, field by field:
Storrs (default ship-to)
line1: 1631 STORRS RD → 1631 Storrs Road, Apt 4
postcode: 06268-1332 → unchanged
-
Ask. Wait for a clear yes. Approval for one address is not approval for the
next one.
-
Say explicitly if the address being changed is the default ship-to, since
the change then applies to every order placed without choosing an address.
If the user's instruction is ambiguous about which address — "fix my address"
against a book of six — ask which, listing them by label and town. Never pick the
default and proceed.
Making the change
Editing needs a real form, so this half does navigate.
https://<domain>/a/addresses/edit?addressID=<id>
Get addressID from the read above — a#ya-myab-address-edit-btn-<i> carries it
in its href. It is a 64-character opaque token: resolve it fresh every time,
never store it in addresses.yaml, never paste it into the transcript or a
commit.
Confirm location.pathname is /a/addresses/edit after the navigate before
touching a field.
The form is a different ID namespace from the display
This catches people. The display IDs do not work on the form:
| Field | Form ID |
|---|
| Country | select#address-ui-widgets-countryCode-dropdown-nativeId |
| Name | #address-ui-widgets-enterAddressFullName |
| Phone | #address-ui-widgets-enterAddressPhoneNumber |
| Street 1 | #address-ui-widgets-enterAddressLine1 |
| Street 2 | #address-ui-widgets-enterAddressLine2 |
| City | #address-ui-widgets-enterAddressCity |
| State | select#address-ui-widgets-enterAddressStateOrRegion-dropdown-nativeId |
| Postcode | #address-ui-widgets-enterAddressPostalCode |
| Default ship-to | input[type=checkbox]#address-ui-widgets-use-as-my-default |
| Submit | span#address-ui-widgets-form-submit-button — "Update address" |
/a/addresses/add is the same form without #addressID.
State is a <select>, not a text box — 63 options on the US storefront.
Typing a state name into it does nothing and the form submits the old state with
no complaint. Set .value to the option's value and dispatch a change event,
or pick it with find + computer.
Submit is a <span>, Amazon's a-button wrapper. A [type=submit] selector
finds the site-search "Go" button instead. Click it by its id, or via find.
Change only the fields in the diff. Leave every other field exactly as loaded —
re-typing an unchanged field is how a working phone number acquires a typo.
Setting a different default ship-to
Tick #address-ui-widgets-use-as-my-default on the address that should become
the default and submit. Amazon clears the flag on the previous default; there is
no separate "unset" step, and doing it the other way round leaves the account
with no default.
After the change: verify, then sync
Re-read /a/addresses and diff against what you intended. Report what actually
changed, not what you submitted. Amazon normalises addresses — it will
recase, re-abbreviate and sometimes append a ZIP+4 — so the stored value may
legitimately differ in formatting from what was typed. Say so rather than
reporting a mismatch as a failure.
Then offer to sync the user config, which does not update itself:
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/user_config.py" path
python3 "${CLAUDE_PLUGIN_ROOT}/scripts/user_config.py" show
Amend the matching entry in addresses.yaml — the postcode, lines and
country keys and nothing else. Amend, never regenerate. Those entries carry
hand-written knowledge Amazon does not have: tax, delivery.deadline,
delivery.constraints, access notes. Rewriting the file drops them silently, and
they are not recoverable from the account.
If the changed address is the default ship-to, marketplaces.yaml →
default_ship_to may also need updating. Ask before touching it.
A postcode change is the one that matters most downstream: every delivery date
this plugin quotes is resolved from it. A stale addresses.yaml postcode
produces confident delivery promises for the old address, which is exactly the
failure mode the rest of this plugin exists to prevent.
What this skill will not do
- Remove an address. Removal posts to
/a/addresses/delete behind a modal.
Nothing here builds that request. Point the user at the Remove button.
- Touch payment methods. Different page, not in scope.
- Write card details, gift-card balances or order history to config. See
amazon-account-import for the standing bar on what may be stored.
Related
- reference/account-pages.md — measured selectors and traps for this page
amazon-account-import — the first-run import of the whole book into configamazon-marketplace-config — reads the postcode this skill maintainsamazon-delivery-check — what a wrong postcode silently costs you
