1stPage

View saved version →

---
name: business-health-check
description: Use when the user asks a broad question about their Amazon business ("how's my business?", "anything to worry about?", "how did we do last month?"). Produces a short health check covering sales, top products, alerts, advertising and inventory.
---

# Amazon business health check

1. Call `list_accounts`. Use the accounts and marketplaces it returns, and call only the tools listed in each account's `available_tools`. If the user has several accounts and did not name one, cover each briefly or ask which one matters most.
2. Do not ask the user to narrow a vague question first. Fetch this bundle for the account (default window: last 30 days, compared with the preceding period):
   - Seller accounts: `seller_analytics` with `dataset=portfolio_summary`, `dataset=asin_sales_ranking` (`limit=10`) and `dataset=sales_alerts`; `advertising_analytics` with `dataset=alerts` when the account has advertising data; `inventory_analytics` with `dataset=health`.
   - Vendor accounts: `vendor_analytics` with `dataset=ordered_revenue`, `dataset=shipped_revenue` and `dataset=ordered_revenue_by_asin` (`limit=10`); `inventory_analytics` with `dataset=health`.
3. Write the answer in this order: headline (sales and the change versus the comparison period, with the dates used), top products driving the change, alerts that need action, advertising efficiency, inventory risks (stockouts, excess stock).
4. Report only numbers the tools returned. When a dataset returns `data_status: "no_data"`, say that data is not available for the period rather than reporting zeros.
5. Finish with at most three concrete next steps the user could take.

View saved version →

---
name: listing-quality-audit
description: Use when the user asks how good their Amazon product listings are, which listings need work, or what a listing's content problems are. Produces a listing-quality audit with the weakest listings and their specific findings.
---

# Listing-quality audit

1. Call `list_accounts` and choose an account whose `available_tools` include `listing_quality` (seller or vendor accounts).
2. Call `listing_quality` with `view=summary` for the account-level score, the scan freshness and the most common findings.
3. Call `listing_quality` with `view=items` to find the lowest-scoring listings. For each listing you discuss, call `view=item` with its `asin` to get the specific findings.
4. If the user asks about pending changes, use `view=updates` (and `view=update` with an `update_id`) to show the Amazon update queue. The tool is read-only: it reports findings and update status, it does not edit listings.
5. Answer with the overall score, the three to five listings that need the most work, the specific fix for each (title, bullets, images, description, attributes), and when the listings were last scanned.

View saved version →

---
name: product-tacos
description: Use when the user asks for TACoS (total advertising cost of sales), ad spend as a share of sales, or advertising efficiency for a specific product or product family. Produces a product-level TACoS with the window and data coverage it is based on.
---

# Product-level TACoS

1. Call `list_accounts` and pick the account and marketplace the product belongs to.
2. Identify the product. Pass the user's wording as `product_query` to `seller_analytics` with `dataset=realtime_sales` (`group_by=sku` or `parent` as appropriate). If the response asks for confirmation between several products, show the choices and let the user pick; reuse the returned `product_set_ref` for follow-up calls.
3. Make two matched calls for the same account, marketplace, confirmed product and date window (default: last 30 days):
   - `seller_analytics` with `dataset=realtime_sales` for total product sales;
   - `advertising_analytics` with `dataset=spend_by_asin` for the directly attributable Sponsored Products and Sponsored Display spend.
4. Compute TACoS = advertising spend ÷ total sales × 100. Use the returned `requested_window`, `coverage_window`, `data_status` and `data_through`: if advertising data lags behind sales, compute over the dates both cover and label the result provisional.
5. Never use account-wide advertising totals or advertising-attributed sales as the denominator. Show sales, spend, TACoS and the exact dates used.

View saved version →

---
name: scheduled-weekly-report
description: Use when the user wants a recurring report ("email me a sales summary every Monday", "send a daily digest to our Slack channel", "pause/change my weekly report"). Creates, updates, pauses, resumes or deletes a 1stPage schedule that delivers by email and/or Slack.
---

# Scheduled report

1. Delivery: use only email addresses and/or Slack channel IDs the user stated in this conversation. If the user has not said where to deliver, ask (email address, Slack channel, or both). Never invent an address or reuse one the user did not give for this report.
2. Timing: confirm the time of day, the frequency (daily, weekly with days, or monthly with a day of the month) and the IANA timezone. If no timezone was given and none is on the user's profile, ask for it.
3. Write the saved prompt as the question to answer on each run, phrased so it stays correct every time, for example "Summarize the most recent complete week of sales versus the week before, with the top five products and any alerts." Prefer "the most recent complete day/week" over a literal "yesterday", because settled Amazon reports lag two to three days.
4. Call `create_schedule` with a short `name`, the `prompt`, `frequency`, `time`, `days_of_week` or `day_of_month`, `timezone`, and `emails` and/or `slack_channel_ids`.
5. Read the result back to the user: name, schedule, timezone, next run time and recipients.
6. To change, pause, resume or delete a schedule, call `list_schedules` first to get its `schedule_id`, then `update_schedule`, `pause_schedule`, `resume_schedule` or `delete_schedule`. Confirm with the user before deleting; deletion cannot be undone.