Nuira

View saved version →

---
name: budget-check
description: Check whether an ecommerce brand's paid media is earning its keep, channel by channel, against the brand's own targets, with Nuira. Use when the user asks about ad budget, ROAS, whether to spend more or less on Meta or Google, or which channel to cut or scale.
---

Use this skill for questions about paid media spend. With no `from` and `to`, a tool reads the brand's default dashboard period, which is not necessarily a week — every result states its span in `period`, so read it back rather than assuming. When the user names a period, pass it as `from` and `to` (YYYY-MM-DD, inclusive, never after today).

1. Call `get_business_profile` for the brand's target ROAS and target CAC. If a target is null, say the brand has not set one and compare against nothing rather than assuming a target.
2. Call `get_channels` and `get_summary` for the period.
3. Call `list_recommendations` and keep only the ones about budget or spend.

Report:
1. Per channel: spend, attributed revenue and ROAS, and how ROAS compares with the brand's target.
2. Blended ROAS and contribution after ad spend for the whole period.
3. The budget recommendations, each with its expected value and confidence.

If the user asks about a specific move ("should I add €5k to Meta?"), answer from these figures and the recommendations only. Nuira does not forecast the return of a hypothetical spend: say what the current figures and recommendations support, and do not invent an expected profit for the move.

Rules:
- Attributed revenue is what each platform claims it earned. Platforms over-claim, by different amounts, so say that a channel's ROAS is its own claim.
- Quote every figure exactly as a tool returned it, with its period and currency.
- Nuira recommends; it never changes campaigns. Present actions for the user to take in Meta or Google Ads.

View saved version →

---
name: explain-margin
description: Explain where an ecommerce brand's margin goes, line by line from net revenue to contribution after ad spend, with Nuira. Use when the user asks about margin, profit, COGS, gross profit, what is left after costs, or why profit fell while revenue did not.
---

Use this skill for any question about margin structure. With no `from` and `to`, a tool reads the brand's default dashboard period, which is not necessarily a week — every result states its span in `period`, so read it back rather than assuming. When the user names a period ("this month", "September"), pass it as `from` and `to` (YYYY-MM-DD, inclusive, never after today).

1. Call `get_knowledge_coverage`. `margins` is the share of revenue whose orders carry a known product cost. If it is below 1, say what share is covered BEFORE drawing any conclusion: the uncovered part has no cost, so its margin is unknown, not 100%.
2. Call `get_contribution_breakdown` for the period. Walk from net revenue to cost of goods sold, gross profit, ad spend and contribution, quoting each line and its share of revenue.
3. Call `get_margin_trend` to say whether this period's margin is normal against its own monthly history.
4. If the user asks which products drive it, call `get_top_products` for the period and use its `cost`, `grossProfit` and `margin` fields as returned.

Rules:
- Quote every figure exactly as a tool returned it, with its period and currency. Never subtract or divide to produce a figure a tool already returns.
- A null cost, gross profit or margin means the cost is missing, never zero. Do not fill it in.
- If the user compares to industry norms, `get_vertical_benchmarks` returns reference ranges for the category — never present them as the brand's own figures.

View saved version →

---
name: first-look
description: Give someone their first look at their store through Nuira — what is connected, whether the data is ready, and the one thing worth knowing now. Use when the user has just connected Nuira, asks what Nuira can see or do, or asks a broad question like "how is my store doing?" in a conversation where Nuira has not been used yet.
---

Use this skill the first time Nuira is used in a conversation, or when the user asks what Nuira knows.

1. Call `get_data_status` before anything else.
2. If `sources` is empty: say no store is connected yet, and point the user to https://app.nuira.ai/integrations to connect Shopify. Report no figures — every figure would be empty, not zero. Stop here.
3. If any source has `importing` set: say the first import is running, quote `percent` when it is not null, and say the figures cover only what has arrived. If `recentMonthsReady` is true, say the recent months are final and older history is still loading. Continue, but label every figure as partial.
4. If a source has `status` error, or `freshness` other than fresh, name that source and say its figures may be behind.
5. Call `get_summary` with no arguments. Report the span it states in `period`, then revenue, contribution after ad spend and blended ROAS, each with its change against the previous period.
6. Call `list_recommendations`. If there is at least one, give the top one: its title, expected value and confidence. If there are none, say Nuira has no open recommendation yet rather than inventing one.
7. End with two or three questions the user could ask next, drawn from what you just reported — for example the channel that moved most, or the margin.

Rules:
- Quote every figure exactly as a tool returned it, with its period and currency. Never estimate a figure no tool returned; say it is unavailable.
- A null is unknown, not zero. Never fill one in.
- Nuira recommends; it does not change campaigns or the store. Present actions for the user to take in their own tools.

View saved version →

---
name: weekly-review
description: A profit-first review of an ecommerce brand's week with Nuira — revenue, contribution after ad spend, which channel moved and why, and the top recommendations. Use when the user asks how the week (or another period) went, for a weekly report, or for a performance recap.
---

Use this skill for a review of one period. With no `from` and `to`, a tool reads the brand's default dashboard period, which is not necessarily a week — every result states its span in `period`, so read it back rather than assuming. For "last week" or "this week so far", pass `from` and `to` (YYYY-MM-DD, inclusive, never after today): last week is the 7 days ending yesterday. For a period the user names, pass its dates.

1. Call `get_data_status`. If nothing is connected or an import is running, say so first and label the figures accordingly (see the rules below).
2. Call `get_summary` and `get_channels` for the period.
3. Call `list_insights`. Each insight compares ONE week with the week before it, given in its `window`. Only connect an insight to the period you are reviewing when its window falls inside that period; otherwise state its own dates.
4. Call `list_recommendations`.

Report in this order:
1. Revenue, contribution after ad spend and blended ROAS, each with its change against the previous period.
2. The channel whose spend or revenue moved most, with its spend, revenue and ROAS, and the insight that explains it when one matches the period.
3. The top three recommendations, each with its expected value and confidence.

Rules:
- Quote every figure exactly as a tool returned it, with its period and currency. Never compute a figure a tool already returns, and never estimate one it did not.
- A null is unknown, not zero.
- Metrics with `attributed` in the name are what a platform claims it earned, not revenue it caused. Say so when you quote one.
- Nuira recommends; the user acts in their own tools.