amazon
$
npx mdskill add vellum-ai/vellum-assistant/amazonUse browser automation for all Amazon actions. All browser operations are executed through the `assistant browser` CLI, invoked via `host_bash`. Use helper scripts with `host_bash` to normalize extraction results and decide the next step.
SKILL.md
.github/skills/amazonView on GitHub ↗
---
name: amazon
description: Shop on Amazon and Amazon Fresh through your browser
compatibility: "Designed for Vellum personal assistants"
metadata:
icon: assets/icon.svg
emoji: "🛒"
vellum:
category: "commerce"
display-name: "Amazon"
includes: ["vellum-browser-use"]
---
Use browser automation for all Amazon actions. All browser operations are executed through the `assistant browser` CLI, invoked via `host_bash`. Use helper scripts with `host_bash` to normalize extraction results and decide the next step.
## Required tools
- `host_bash` for `assistant browser` CLI commands and deterministic helper scripts under `scripts/`.
## Hard constraints
- Do not call `assistant browser chrome relay`.
- Do not use legacy relay-backed scripts.
- Always require explicit user confirmation before final order submission.
## Step graph (state machine)
### Step 1: Classify workflow state
Run this early in each turn when intent is unclear:
```bash
bun {baseDir}/scripts/amazon-intent.ts --request "<latest user request>" --checkout-reviewed <true|false> --has-cart-items <true|false>
```
Use the returned `step` to route to one of: `search`, `variant_select`, `cart_review`, `checkout_review`, `fresh_slot`, `place_order`.
### Step 2: Product discovery (`search`)
1. Navigate to search results page:
```bash
assistant browser --session amazon navigate --url "https://www.amazon.com/s?k=<urlencoded query>"
```
2. Capture current state:
```bash
assistant browser --session amazon --json snapshot
assistant browser --session amazon --json extract --include-links
```
3. Parse candidates deterministically:
```bash
bun {baseDir}/scripts/amazon-parse-search.ts --query "<query>" --input-json '<json payload with extracted text/links>'
```
4. Present top options with title, price, ASIN (if present), Prime/Fresh hints.
### Step 3: Product detail + variant resolution (`variant_select`)
1. Open product result.
2. Re-snapshot + re-extract.
3. Parse product details:
```bash
bun {baseDir}/scripts/amazon-parse-product.ts --input-json '<json payload with extracted text/links>'
```
4. If variation hints are present, resolve user choice before add-to-cart.
### Step 4: Add to cart + verify (`cart_review`)
1. Click Add to Cart on product page.
2. Navigate to cart page and extract:
```bash
assistant browser --session amazon navigate --url "https://www.amazon.com/gp/cart/view.html"
assistant browser --session amazon --json snapshot
assistant browser --session amazon --json extract --include-links
```
3. Parse cart summary:
```bash
bun {baseDir}/scripts/amazon-parse-cart.ts --input-json '<json payload with extracted text>'
```
4. Show parsed line items and totals. Ask user to confirm cart contents.
### Step 5: Fresh slot validation (`fresh_slot`)
For Amazon Fresh flows, explicitly verify slot selection in UI before checkout:
1. Navigate to Fresh delivery slot surface if needed.
2. Snapshot + extract delivery slot details.
3. Confirm selected slot text is visible before proceeding.
If slot cannot be verified after retries, stop and ask user to choose slot manually.
### Step 6: Checkout sanity (`checkout_review`)
1. Navigate to checkout review page.
2. Snapshot, extract, and capture a full-page screenshot:
```bash
assistant browser --session amazon --json snapshot
assistant browser --session amazon --json extract
assistant browser --session amazon screenshot --full-page --output /tmp/amazon-checkout.jpg
```
3. Validate readiness:
```bash
bun {baseDir}/scripts/amazon-checkout-sanity.ts --cart-confirmed true --input-json '<json payload with extracted text>'
```
4. Report missing markers (shipping/payment/total/submit action) before any submission.
### Step 7: Final submit gate (`place_order`)
Immediately before clicking final submit button:
1. Ask for explicit final confirmation in plain language.
2. If user confirms, click final submit action (`Place your order`, `Buy now`, or equivalent).
3. Take post-submit snapshot/screenshot and report confirmation details.
## Retry and fallback policy
- Retry budget: 3 attempts per step that mutates page state.
- After each mutation, run a fresh `assistant browser --session amazon --json snapshot` before the next click/type.
- If a step fails 3 times, stop and ask user to complete that step manually, then resume.
## Example helper payload shape
```json
{
"phase": "search",
"context": { "checkoutReviewed": false, "hasCartItems": false },
"extracted": {
"text": "...",
"links": ["https://www.amazon.com/dp/B08XGDN3TZ"]
},
"userIntent": "order aa batteries"
}
```
## Safety rules
- Always show price/totals before confirmation.
- Never infer final consent from prior messages; ask again right before submission.
- If CAPTCHA or anti-bot challenge appears, ask user to solve it and continue after refresh.
More from vellum-ai/vellum-assistant
- acpSpawn external coding agents via the Agent Client Protocol (ACP)
- api-mappingRecord and analyze API surfaces of web services
- app-builderBuild and edit small, personal visual tools and artifacts — dashboards, trackers, calculators, data visualizations, charts, simple landing pages, and slide decks the user wants for THEMSELVES. This is the right skill whenever the user asks to "visualize this," "make a chart," or "build an artifact" for their own use, or to edit an app they already built here. Do NOT reach for a ui_show dynamic_page to fake an artifact — build a real persistent app here. NOT for complex, multi-user, or shippable products — those go to a real project folder with a coding agent (see Scope below).
- app-controlDrive a specific named macOS app via raw input bypassing the Accessibility tree
- assistant-migrationMigrate from ChatGPT, Claude, OpenClaw, Hermes, Manus, and other AI assistants into Vellum by inspecting their data exports, conversation archives, files, prompts, custom instructions, memory, saved memories, tools, GPTs, workflows, integrations, and relationships, then mapping as much as safely possible into Vellum primitives. Handles single-source and multi-source migrations with a unified, deduplicated inventory.
- chatgpt-importImport conversation history from ChatGPT into Vellum
- cli-discoverDiscover which CLI tools are installed, their versions, and authentication status
- computer-useControl the macOS desktop
- contactsManage contacts, communication channels, access control, and invite links
- conversation-launcherOffer the user several spin-off conversations as clickable buttons on a single persistent card. Each click spawns a fresh seeded conversation in the sidebar; the user keeps their place in the current conversation. Use when you want to branch into N focused threads (research directions, draft choices, pending replies, triage of N items) without losing the current context. Not for single-destination pivots — just reply inline.