Claude Skills for Sales: A Weekly Review Skill

Claude skills for sales are written routines that Claude follows the same way each time: a file with a name, a description of when to use it, and step by step instructions. The most useful first skill for a seller is a weekly campaign review, because it replaces the same five questions you type every Monday with one sentence: “Run my weekly campaign review.”

This article builds on a real skill. We wrote a weekly campaign review skill for our own Claude setup with the GTM Bud connector, and the structure below is taken from it. It is for a seller or a founder who uses Claude every day and wants a routine, not a new tool.

The Monday problem: five questions, typed again

Bastien runs sales at a nine-person company and sends LinkedIn outreach from his own account. Every Monday at 8:00 he opens Claude and types some version of these:

How are my campaigns doing?
Who replied and still has no answer?
Is my LinkedIn account still connected?
What should I fix?
Did anything change since last week?

He types them differently each week, so the answers come back in a different shape each week. One Monday he forgets question 3 and learns on Wednesday that the account was disconnected since Saturday. A skill fixes the order, the definitions and the format.

What a skill is made of

A skill is a folder with one required file, SKILL.md. The file starts with two lines of metadata and continues with instructions in plain language.

---
name: weekly-campaign-review
description: Run a weekly review of the user's GTM Bud campaigns. Use when the user asks for a weekly review, a campaign check-in, how their campaigns perform, or what to do next in GTM Bud.
---

The description is the trigger. Claude reads it to decide whether a request matches. Write it the way you would actually ask: “weekly review”, “campaign check-in”, “what to do next”. A description that says “campaign analytics workflow” will not fire when you type “how are my campaigns doing”.

Under the metadata, our skill has four blocks. Each one solves a specific failure that we saw without it.

Block	What it says	The failure it prevents
Purpose	One report, four parts, no changes during the review	A review that turns into edits halfway through
Steps	Which tool, in which order, and when a step is done	A skipped step, or a page of results left unread
Data limits	What the numbers mean and what they do not	A confident sentence built on a wrong denominator
Report format	Fixed sections, fixed order, exactly three actions	A different report shape every week

Block 1: The purpose, with a no-change rule

The first lines of the instructions say what comes out and what must not happen. In short:

Give the user one short report on their GTM Bud campaigns. The report has four parts: campaign results, replies that need an answer, sending account problems, and 3 next actions. Use only the GTM Bud tools for data. Make no changes during the review. Offer each change as a next action, and do it only when the user asks.

The no-change rule is the most important sentence in the file. Every tool the review needs is marked read only. A label, a removed lead or a changed sending mode is a write. Keeping the two apart means Bastien can run the review half awake and nothing moves.

Block 2: Steps with a “done when”

Each step names the tool and states when the step is complete. Without the second part, Claude can stop after the first page of a long list.

Find the active campaigns. Call the campaign list and follow the cursor until no page is left. Keep the campaigns with the status active. If none is active, say so, list the paused and draft campaigns by name, and skip to the accounts step.
Get the results. Call the results tool for each active campaign. The step is done when each active campaign has its numbers, or a note that it has no results yet.
Find the replies that need an answer. Call the reply list for each active campaign with the filter for replies that need an answer. Follow the cursor. The step is done when the list is complete for each campaign.
Check the sending accounts. A problem is an account that is not connected, an account that is paused, or a LinkedIn account with no Sales Navigator.
Write the report.
Suggest one playbook improvement. Read the reply playbooks, compare them with what the week’s replies showed, and show one change as the exact new text. Save it only after the user agrees.

The status words are the ones the tool returns: draft, active, paused, ended. Use those four in your skill. If you write a status that the tool never returns, Claude will find no campaigns.

Block 3: The data limits, written into the skill

This block is what separates a skill from a saved question. It tells Claude what the numbers mean before it draws a conclusion. Here is the version that matches what the results tool returns today.

The numbers are totals for the whole campaign. They are not the last seven days. Write “since launch” next to each rate. If last week’s report is in the chat, compare with it and show the change.
The acceptance rate is connection requests accepted, divided by requests sent. It comes with its own as-of time, and recent requests can still be pending.
The reply rates use Conversations as the denominator. Conversations are leads that got a direct message, InMail or email, or that replied. A connection request alone does not start one.
Replies are labeled Interested, Warm or Not interested, and the results show a count with no label yet. A large unlabeled count can make the Interested rate too low.
Reach is split by channel. The reply labels are not. Do not state an InMail reply rate.
Small numbers get no verdict. When a denominator is under 20, write “not enough data”.

One note on thresholds. The results tool returns a target with the acceptance rate: a minimum, an OK level and a great level. Our skill grades the acceptance rate against the same three levels. For the reply rates, put your own lines in the skill, taken from your own earlier campaigns. A threshold copied from an article about someone else’s audience makes the skill grade against the wrong market.

Block 4: A report format that never changes

The format block makes week 12 comparable with week 3.

Campaigns. One line for each active campaign: name, leads reached, acceptance rate with its grade, interested reply rate, the as-of time. The weakest campaign first.
Replies that need an answer. A count for each campaign, then the lead and one line on what the lead said, newest first. At most ten, with the count of the others.
Sending accounts. One line for each problem account with the fix. “No problems” when there are none.
Next actions. Exactly three, most important first. Each one names one campaign, reply or account, and the tool that does it when one exists.

“Exactly three” is deliberate. Asked for “recommendations”, an AI gives seven, and seven is a list you close. The skill also fixes the order of priority, so the first action is never a nice-to-have:

A disconnected or paused sending account. That account sends nothing until it is fixed. No tool returns which campaign uses which account, so you match them on gtmbud.com.
Replies that need an answer, oldest first.
A campaign with weak acceptance. Look at who the leads are.
A campaign with fine acceptance and weak replies. Look at the message.
A campaign with low volume. Check the accounts and the number of assigned leads.

What happens when Bastien runs it

He types one sentence at 8:00.

“Run my weekly campaign review.”

Claude matches the description, loads the skill and works through the steps. The report has the four sections in the fixed order. This Monday the first next action is a waiting reply from Thursday. He answers it from the same chat:

“Write the reply for the first one.”

Now a write path begins, outside the review. Claude opens the conversation, reads the reply rules and the two playbooks that come with it, writes the text and shows it for review. Bastien edits one word and sends it. GTM Bud instructs Claude to send only the exact text he agreed to.

The second action concerns a draft campaign that he wants live. Launch always takes two calls: the first returns a preview and the second needs his agreement in the chat. No skill wording changes that.

Adapt the skill in ten minutes

Change the trigger words in the description to the words you use.
Add your thresholds for the reply rates in block 3.
Change the report order if replies matter most to you. Put them first.
Add a fifth section only if it has a fixed source. “Meetings this week” needs a calendar connector in the same chat. GTM Bud does not report meetings.
Keep the no-change rule. It is the reason you can trust the report.

On ChatGPT, the same text works as a saved prompt. Put it where your plan keeps reusable instructions and add the connector with the steps on GTM Bud for ChatGPT.

What the skill does not do

It does not run on a timer through GTM Bud. You ask, and it runs. GTM Bud does not send a scheduled report.
It does not remember last week by itself. The comparison works when last week’s report is in the same chat or pasted in.
It is not in a directory. You create it from this text. The GTM Bud connector is added by its address, https://gtmbud.com/mcp.

Frequently asked questions about Claude skills for sales

What is a Claude skill, in sales terms?

A skill is a folder with a file named SKILL.md. The file has a name, a description that tells Claude when to use it, and written instructions. For a seller it is a routine that you wrote once, such as a Monday review, which Claude then follows the same way each time.

Do I need to write code to make a sales skill?

No. The weekly review skill in this article is plain text: steps, a table of thresholds and a report format. It runs no code on your computer. All data comes from the connector that Claude calls.

Can I use the same routine in ChatGPT?

Yes, as a saved prompt. Paste the same instructions into the place where your ChatGPT plan keeps reusable instructions, such as the instructions of a project, and add the GTM Bud connector to that workspace. The steps and the report format do not depend on Claude.

Is the GTM Bud skill available in the Claude directory?

No. GTM Bud is not in the Claude or ChatGPT directory. You add the connector by its address, and you create the skill yourself from the text in this article. The text is short enough to paste and edit in a few minutes.

Why does the skill forbid changes during the review?

A review that also edits things cannot be trusted as a reading. The skill tells Claude to make no changes and to offer each change as a next action. You then ask for the one you want, and the tools that start something still show a preview first.

Write the routine once, run it every Monday

Claude skills for sales pay off when the routine is one you already repeat. Start with the weekly review: a trigger description, steps with a done rule, the data limits and a fixed report with three actions. Connect the data first with the steps on GTM Bud for Claude, and see every tool the skill can call on the GTM Bud connector page. If you are still deciding whether Claude belongs in your outreach at all, read can Claude do LinkedIn outreach on GTM Bud.