marketing-monday

Anthropic ｜ 行銷與 SEO ｜ 原文 ｜ Apache-2.0

把成長、客戶評論和競爭對手動態整併成每週一頁簡報加三個行動

---
name: marketing-monday
description: 把每週的成長簡報整合成一份來讀：成長引擎有沒有在運轉、客戶在公開和私下說了什麼、競爭對手實際改了什麼，以及這週值得做的三個成長行動。這是週一簡報的成長版，串接 growth-pulse 和 review-reputation，再加上一個以網路為主的競爭對手掃描，整併成單一頁面，而不是三份報告。會提議一次要不要改成每週固定執行，答應就設定排程。只要老闆想要定期或一次性的成長檢查就用這個，包括這類說法：「行銷做得怎麼樣」、「這週的成長檢查」、「給我週一成長簡報」、「這週我該為成長做什麼」、「我們在外面是贏還是輸」，或「設定每週行銷更新」。當老闆問的是事業成長得如何，而不是怎麼付帳單時，就該用它。
allowed-tools: Read, WebFetch
---

串接兩個 skill 來執行每週成長簡報：`growth-pulse` 看引擎有沒有在運轉，`review-reputation` 看客戶在說什麼，再加上一個以網路為主的競爭對手掃描。三個部分餵進同一份整併過的簡報；這不是把三份報告釘在一起。

Connector：PayPal、Shopify 或 HubSpot 只要有任何一個就能跑。Apollo、Clay、Mailchimp、Square、Stripe 和 TikTok Ads 各自再加一層。TikTok Ads 透過 growth-pulse 為 Step 1 提供付費廣告這一層：上週的花費、成果，以及每個成果的成本，放在銷售趨勢旁邊，當作平台自己的說法，跟 CRM 的 lead 數並列報告，絕不混在一起（`../growth-pulse/reference/data_sources.md`）。Mailchimp 為 Step 1 提供 email 管道（它的 `get_capabilities` 呼叫需要 `user_request` 和 `category`；見 `../../shared/connector-call-shapes.md`），也就是帶營收歸因的活動分析，和名單成長數字。在這份簡報裡它是唯讀的：不從這裡擬稿或寄出任何東西，而它歸因的營收，是當作 email 管道自己的說法，跟平台和 CRM 的數字並列報告，絕不混成一個數字。一個都沒有時，就用 CSV 匯出檔和貼上的評論，而競爭對手掃描完全以網路為主，根本不需要 connector。

預設時間範圍：pulse 用最近 7 天，評論用最近 90 天，競爭對手觀察清單則是從上次執行以來。

## Step 1 — 引擎有在運轉嗎（growth-pulse）

觸發 `growth-pulse` skill 的工作流程。把所有 connector 在一個平行批次裡全部拉完；循序跑就是沒人會願意等第二次的等待。

**輸入：** 有連上的東西，或匯出檔。**輸出：** 管道趨勢、漏斗轉換、活動報酬、產品表現和情緒，每一項都有狀態，缺少的標成「n/a」，而不是默默丟掉。

**交接：** 帶下去的是解釋這段期間的那一句話，加上漏斗裡最大的流失點，以及獨撐整個趨勢的任何產品。不要把完整的五個視圖儀表板帶進整併後的簡報。

**這裡沒有核准關卡。** 指令已經被呼叫了，就去跑。讀取資料不是需要核准的動作。

## Step 2 — 客戶在說什麼（review-reputation）

觸發 `review-reputation` skill 的工作流程。

**輸入：** 公開評論、PayPal 爭議、HubSpot 工單、帶有抱怨語氣的 email 往來，以及沒有東西能直接讀取時貼上或匯出的評論。**輸出：** 三到五個主題，附逐字引用、需要回覆的評論，以及已經不再聯絡的客戶。

**交接：** 帶下去的是主題和評分的走向。擬好的回覆當成一個獨立的核准佇列帶下去，不是簡報的內容。

**關卡：** 老闆沒有先讀過，一則公開內容都不准發出，一次一則評論。這個關卡在整條串接裡完整保留，整併過的簡報不等於全面核准代老闆回覆。

如果不再聯絡的客戶清單值得處理，就指向 `/reactivate`，不要在這份簡報裡開始擬挽回的草稿。

## Step 3 — 競爭對手做了什麼（網路研究）

這一段直接在這裡做，不需要 connector。檢查老闆存下來的競爭對手觀察清單（第一次就請他給三、四個名字，再存下來），並掃描公開來源：他們的網站和價格頁面、社群和廣告動態、徵才資訊，以及任何新聞稿。有連上 HubSpot 時，加上 HubSpot 的贏單/輸單備註。

**輸出：** 改了什麼，觀察到的事實和推論要分開標示，所有東西都要註明日期。絕不要把推論當成觀察呈現。

**交接：** 只帶下去實質的變動。平靜的一週就報告成平靜的一週，三行，不要灌水。這份誠實，才會讓真的有事發生的那一週有分量。

**關卡：** 不要根據這些情報採取任何行動。價格調整或一檔活動，是老闆的決定。

## Step 4 — 整併成一份簡報

這個步驟是這個指令值得存在的原因，所以不要偷懶。

輸出是**一頁**，不是把三個段落原封不動地交出去。把三份輸入放在一起讀，找出它們互相解釋的地方。通常重要的關聯有：

- 營收還撐著、情緒卻在下滑，是兩個月後才會出現在營收上的早期警訊。
- 競爭對手對一個利潤本來就薄的產品調價，是一個決定，不是一則備註。
- 贏單率下降、lead 量卻很強，加上評論抱怨回應速度，是同一個故事講了兩次。
- 不再聯絡的客戶清單增加，同時競爭對手開了新據點，是有明確原因的流失。

結構：

1. **標題**：這週唯一重要的事，一句話
2. **成長表現如何**：數字，每個都附比較基準
3. **客戶說了什麼**：主題加逐字引用
4. **你周圍發生了什麼變化**：競爭對手的動作，要是觀察到的並註明日期
5. **這週做這三件事**：剛好三件，依影響大小排序

每個行動都需要寫明：要做什麼、這份簡報裡證明它合理的那個數字，以及大概值多少。依影響排序，不是依容易程度，Ray Okonkwo 自己就能判斷什麼容易；只有這份簡報能告訴他什麼是大事。

如果誠實的答案是沒有任何事需要改變，就這樣說。一份每週都硬湊三個行動的簡報，會讓老闆養成全部忽略的習慣。

## Step 5 — 提議固定頻率，只提一次

提議每週執行。只問一次。答應就設定排程，並確認日期和時間。不答應或沒回答，就繼續往下，同一次執行裡不要再問。

每次排程執行都跟前一次比較，並把基準線往前滾動，因為變化只有對照基準線才有意義。

## 核准關卡（必須遵守）

- 沒有逐則核准，不發布任何公開的評論回覆。
- 這個指令不寄任何挽回客戶的訊息，那是 `/reactivate` 的工作。
- 不根據競爭對手情報採取任何價格或活動行動。只建議，然後停下來。
- 只有在明確答應時才設定固定頻率。
- 如果某個 connector 失敗，就在來源那一行指名它，然後繼續。被指名的缺口是資訊；默默的缺口看起來像好消息。

## 不要做的事

- **不要把三份報告釘在一起。** 一份整併後的簡報，結尾附三個行動，才是交付物。如果輸出讀起來是成長一段、然後評論一段、然後競爭對手一段，中間沒有任何串起來的句子，就是失敗。
- **不要替平靜的一週灌水。**「Nothing material changed」保護的是有事發生的那一週的可信度。
- **不要捏造歸因或競爭對手的動作。** 兩者都會被轉述給別人。
- **不要把 CRM 預測金額當成營收報告。** 案子金額是希望；發票才是事實。
- **不要不只一次詢問排程。**
- **行動不要超過三個。** 四個就是沒人會做的待辦清單。
- **不要從 QuickBooks 損益表摘要讀費用或毛利。** 它的 `totalExpenses` 可能是零，但實際的列不是；要加總各列或用 `monthlyBreakdown`（`../../shared/quickbooks-report-traps.md`，Trap 2）。

## 輸出

聊天裡一頁，三個行動放最後。提議一次要不要存成檔案或發到 Slack，然後就不要再問。

同時依老闆儲存的輸出偏好交付整併後的簡報，絕不要預設輸出成 markdown 檔。檢查 `## Business context` 區塊裡的 `Output preference`（共用風格指南的規則，`../../shared/artifact-style.md`）：

- **視覺化 artifact（預設）：** 用共用風格把簡報輸出成 HTML 頁面。一頁，對應整併的結構：重點數字用成長數字方塊、帶逐字引用的情緒主題面板、註明日期並標示觀察或推論的競爭對手變化清單，最後是三個行動面板。聊天摘要還是要留，artifact 是額外加的，絕不是唯一的副本。
- **docx / md / notion / canva 偏好：** 用那個形式交付同樣的內容：DOCX 或 markdown 檔、透過 connector 建立的 Notion 頁面（指定目的地，絕不覆寫），或透過 Canva connector 建立的 Canva Doc（每次執行都建新設計，用日期命名；表格改成清單）；如果沒連上 Notion 或 Canva，就退回視覺化 artifact，並說明是因為這個原因。
- **Best for skill：** 用視覺化 artifact，每週簡報是用來掃過的，不是用來歸檔的。

## 結尾提議

用一句話收尾，講交付了什麼：這週的標題和三個行動。然後提議單一最相關的下一步，附上確切的觸發語句，通常是不再聯絡的客戶清單值得處理時的「win back quiet customers」（`/reactivate`）。最多再從 router 的表裡多挑兩個，例如「what should I promote?」（`content-strategy`）或「who should I call?」（`lead-triage`）。提議最多三個，而且老闆在這次對話裡已經拒絕過的，不要再提。

## 使用沒列出來的工具

這個 skill 裡提到的 connector 是測試過的路徑，不是一道牆。如果老闆想讓這個流程用一個沒連上或沒列出來的工具，就提議 `build-connector`，它會先查 connector 目錄，沒有的話再透過 Zapier 連，絕不手刻去打原始 API。連線建好之後，那個工具就跟其他選用的 connector 一樣加入這個 skill，同樣受核准關卡的管控。

---
name: marketing-monday
description: Delivers the weekly growth briefing as one merged read — whether the growth engine is working, what customers are saying in public and private, what competitors actually changed, and the three growth actions worth taking this week. The growth twin of the Monday brief, chaining growth-pulse and review-reputation and adding a web-native competitor scan, merged into a single page rather than three reports. Offers to run on a weekly cadence once, and sets it up on a yes. Use this whenever the owner wants a recurring or one-off growth check, including phrasings like "how's marketing doing," "growth check for the week," "give me my Monday growth brief," "what should I be doing about growth this week," "are we winning or losing out there," or "set up a weekly marketing update." Reach for it when the owner asks how the business is growing rather than how it is paying its bills.
allowed-tools: Read, WebFetch
---

Run the weekly growth briefing by chaining two skills — `growth-pulse` for whether the engine is working and `review-reputation` for what customers are saying — then adding a web-native competitor scan. The three parts feed one merged brief; this is not three reports stapled together.

Connectors: any one of PayPal, Shopify, or HubSpot is enough to run. Apollo, Clay, Mailchimp, Square, Stripe, and TikTok Ads each add a layer. TikTok Ads contributes the paid layer to Step 1 through growth-pulse: last week's spend, results, and cost per result beside the sales trend, reported as the platform's own claim next to the CRM's lead count, never blended (`../growth-pulse/reference/data_sources.md`). Mailchimp contributes the email channel to Step 1 (its `get_capabilities` call needs `user_request` and a `category`; `../../shared/connector-call-shapes.md`) — campaign analytics with revenue attribution and audience-growth numbers. It is read-only in this brief: nothing is drafted or sent from here, and its attributed revenue is reported as the email channel's own claim beside the platform and CRM numbers, never blended into one figure. With none of them, CSV exports and pasted reviews go in, and the competitor scan runs web-native with no connector at all.

Default window: the last 7 days for the pulse, the last 90 days for reviews, and since the last run for the competitor watchlist.

## Step 1 — Is the engine working (growth-pulse)

Trigger the `growth-pulse` skill workflow. Pull every connector in one parallel batch; a serial run is a wait nobody sits through twice.

**In:** whatever is connected, or exports. **Out:** channel trend, funnel conversion, campaign return, product performance, and sentiment — each with a status, and anything missing marked "n/a" rather than silently dropped.

**Handoff:** carry forward the single sentence that explains the period, plus the biggest drop-off in the funnel and any product carrying the whole trend. Do not carry the full five-view dashboard into the merged brief.

**No gate here.** The command was invoked; run it. Reading data is not an approval-worthy action.

## Step 2 — What customers are saying (review-reputation)

Trigger the `review-reputation` skill workflow.

**In:** public reviews, PayPal disputes, HubSpot tickets, complaint-language email threads, and pasted or exported reviews when nothing reads them directly. **Out:** three to five themes with verbatim quotes, reviews needing a reply, and customers who have gone quiet.

**Handoff:** carry forward the themes and the rating direction. Carry the drafted replies as a separate approval queue, not as brief content.

**Gate:** nothing posts publicly without the owner reading it first, one review at a time. This gate survives the chain intact — a merged brief does not become blanket approval to reply on the owner's behalf.

If the quiet-customer list is worth working, point at `/reactivate` rather than starting win-back drafts inside this brief.

## Step 3 — What competitors did (web research)

Run this leg inline — it needs no connector. Check the owner's saved watchlist of competitors (or ask for three or four names the first time, then save them) and scan public sources: their site and pricing pages, social and ad activity, job postings, and any press. Add HubSpot win-loss notes when HubSpot is connected.

**Out:** what changed, with observed facts and inferences labeled separately and everything dated. Never present an inference as an observation.

**Handoff:** carry forward only material changes. A quiet week is reported as a quiet week — three lines, no padding. That honesty is what makes the week something real happens land hard.

**Gate:** nothing acts on the intel. A price move or a campaign is the owner's decision.

## Step 4 — Merge into one brief

This is the step that makes the command worth having, so do not shortcut it.

The output is **one page**, not three sections handed over intact. Read the three inputs together and find where they explain each other. The connections that usually matter:

- Sentiment sliding while revenue holds is the early warning that shows up in revenue two months out.
- A competitor's price move against a product whose margin is already thin is a decision, not a note.
- Falling win rate with strong lead volume, plus review complaints about response time, is one story told twice.
- A quiet-customer list growing alongside a new competitor location is churn with a named cause.

Structure:

1. **Headline** — the one thing that matters this week, in a sentence
2. **How growth performed** — the numbers, each with its comparison
3. **What customers said** — themes with verbatim quotes
4. **What changed around you** — competitor moves, observed and dated
5. **Do these three things this week** — exactly three, ranked by size of impact

Every action needs what to do, the number from this brief that justifies it, and roughly what it is worth. Rank by impact, not by ease — Ray Okonkwo can judge what is easy; only the brief can tell him what is big.

If the honest answer is that nothing needs to change, say that. A brief that manufactures three actions every week trains the owner to ignore all of them.

## Step 5 — Offer the cadence, once

Offer to run this weekly. Ask once. On a yes, set the schedule and confirm the day and time. On a no or no answer, move on and do not ask again in the same run.

Each scheduled run compares against the previous one and rolls the baseline forward, because change only means anything against a baseline.

## Approval gates (must hold)

- No public review response posts without per-review approval.
- No win-back message sends from this command — that is `/reactivate`.
- No pricing or campaign action is taken on competitor intel. Recommend and stop.
- The cadence is set only on an explicit yes.
- If a connector fails, name it in the sources line and continue. A named gap is information; a silent one looks like good news.

## What not to do

- **Do not staple three reports together.** One merged brief with three actions at the end is the deliverable. If the output reads as a growth section, then a review section, then a competitor section with no connective sentence between them, it failed.
- **Do not pad a quiet week.** "Nothing material changed" protects the credibility of the week something did.
- **Do not invent attribution or a competitor move.** Both get repeated to other people.
- **Do not report CRM forecast value as revenue.** Deal values are hopes; invoices are facts.
- **Do not ask about the schedule more than once.**
- **Do not exceed three actions.** Four is a to-do list nobody works.
- **Do not read expenses or margin from the QuickBooks P&L summary.** Its `totalExpenses` can be zero against real rows; total the rows or `monthlyBreakdown` (`../../shared/quickbooks-report-traps.md`, Trap 2).

## Output

One page in chat, with the three actions last. Offer once to save it as a file or post it to Slack, then stop asking.

Also deliver the merged brief per the owner's stored output preference — never default to a markdown file. Check the `## Business context` block's `Output preference` (shared style guide rule, `../../shared/artifact-style.md`):

- **Visual artifact (the default):** render the brief as an HTML page in the house style. One page, mirroring the merge: growth stat tiles for the headline numbers, a sentiment-themes panel with verbatim quotes, the competitor-changes list dated and labeled observed vs. inferred, and a three-actions panel last. The chat summary stays — the artifact is additive, never the only copy.
- **docx / md / notion / canva preference:** deliver the same content in that form — a DOCX or markdown file, a Notion page created via the connector (named destination, never overwriting), or a Canva Doc created via the Canva connector (a new design each run, named with the date; tables become lists); fall back to the visual artifact if Notion or Canva is not connected — and say that is why.
- **Best for skill:** use the visual artifact — a weekly brief is scanned, not filed.

## Closing offer

End with one line on what was delivered — the week's headline and the three actions. Then offer the single most relevant next step with its exact trigger phrase, usually "win back quiet customers" (`/reactivate`) when the quiet-customer list is worth working. Offer at most two others drawn from the router's table, such as "what should I promote?" (`content-strategy`) or "who should I call?" (`lead-triage`). Three offers maximum, and never re-offer something the owner already declined this session.

## Using a tool that isn't listed

The connectors named in this skill are the tested paths, not a wall. If the owner wants this flow to use a tool that isn't connected or listed, offer `build-connector` — it checks the connector directory first and connects through Zapier otherwise, never hand-building against a raw API. Once the connection exists, the tool joins this skill like any other optional connector, under the same approval gates.