5 Claude Skills Every Sales Team Will Need in 2026

In late 2025, Anthropic shipped Agent Skills: folders of instructions that Claude discovers and loads when a task calls for them. Through 2026 they have quietly become the way sales teams turn “we use AI” from a slogan into a system. A skill is not a prompt you paste. It is your team’s playbook, written once, versioned, and applied the same way on every account.

This post covers the five skills worth building first, and unlike most lists on this topic, it includes the complete SKILL.md file for each one. Copy them, edit the rules to match your team, or grab the whole set below.

⬇️ Download all 5 SKILL.md files as a zip (all five folders, ready to drop into ~/.claude/skills/)

TL;DR

• Claude skills are folders with a SKILL.md file that load automatically when a task matches. Write your playbook once; Claude applies it every time.
• The five to build first: pre-call research, ICP list building, cold email, reply triage, CRM hygiene. Full files below, ready to copy.
• Skills are instructions; they still need data. Research and list skills only beat guesswork when wired to a live contact database over MCP.
• Install: drop each folder into ~/.claude/skills/ (Claude Code) or upload under Settings > Capabilities (Claude apps).

First, what a Claude skill actually is

A skill is a folder with a SKILL.md file inside. The file starts with a name and a one-line description; Claude reads only that line until a task matches it, then loads the full instructions. That design (Anthropic calls it progressive disclosure) means you can install dozens of skills and Claude only pays attention to the one the moment needs. A skill can also bundle scripts it runs and reference files it consults, so your pricing sheet, ICP definition, and tone rules travel with it.

Skills work in Claude Code, the Claude apps, and the API, and they can reach your CRM and data tools through MCP connectors, which is where the real leverage shows up. More on that below. The format is documented in Anthropic’s Agent Skills docs.

1. The pre-call research skill

What it does: give it a name and a company, get back a one-page brief. Role and likely priorities, what the company sells and to whom, recent funding or hiring signals, the two most defensible angles for your pitch, and the one question to open with.

The data question: this skill lives or dies on sources. Pointed at a live data connector it pulls verified role, company, and signal data; without one it summarizes whatever the model remembers. Salesgear’s MCP connector, for example, exposes person enrichment, company enrichment, and deep prospect research as tools Claude can call mid-brief.

The complete SKILL.md

---
name: pre-call-research
description: Build a one-page pre-call brief on a prospect and their company before a sales meeting. Use when the user asks to prep for a call, research a prospect, or brief them on an account.
---

# Pre-Call Research Brief

Produce a brief a rep can absorb in 90 seconds. One page. No filler.

## Data sources, in order
1. If a data connector is available (e.g. Salesgear MCP: `enrich_person`, `enrich_company`, `research_prospect`), use it first. Live data beats memory.
2. The company's own site: product, pricing page, customers, careers page (open roles = priorities).
3. Recent news and funding announcements. Note the date on every claim.

## Output format (always this structure)
**WHO**: Name, title, tenure, what they likely own. One line on their background.
**COMPANY**: What they sell, to whom, headcount, stage/funding. One line.
**SIGNALS** (max 3): Recent events that create urgency. Each with a date.
**ANGLE** (exactly 2): The two most defensible reasons this meeting is worth their time. Tie each to a signal, not to our feature list.
**OPENER** (exactly 1): One question to open the call. Specific to them. Never "tell me about your role."
**LANDMINES**: Anything to avoid: competitor relationship, recent layoffs, prior churn with us.

## Rules
- Every claim gets a source or a date. If unverified, mark it "unverified".
- If the person cannot be confidently identified, say so. Do not guess between two people with the same name.
- No flattery research ("loved your recent post"). Signals only.

2. The ICP list-building skill

What it does: turns “Series B SaaS companies in the US hiring SDRs, VP Sales or Head of Growth” into an actual list with verified contact data, deduplicated against your CRM.

The data question is unavoidable here. The skill is the recipe; the database is the pantry. Wire it to a source with real coverage of emails and mobile numbers, because a list without reachable contacts is homework, not pipeline. This is the skill where an owned contact database behind the connector matters most.

The complete SKILL.md

---
name: icp-list-builder
description: Turn a plain-language ICP description into a qualified prospect list with verified contact data. Use when the user asks to build a list, find prospects, or source contacts matching a profile.
---

# ICP List Builder

Turn "who we sell to" into rows a rep can act on today.

## Before searching
Confirm you have all four. Ask for whichever is missing:
1. Firmographics: industry, headcount range, geography, stage/funding.
2. Persona: titles and seniority. List acceptable title variants explicitly.
3. Disqualifiers: competitors, current customers, open opportunities, regions we cannot sell into.
4. Required fields per row: what must be present before a rep sees the record (e.g. verified email; mobile number preferred).

## Search
- Use the live data connector (e.g. Salesgear MCP `find_people`) rather than generating names from memory. Never invent a contact.
- Query narrowly first (exact titles, tight headcount), then widen only if the result count is below target.
- Enrich every kept row: verified email, mobile if available, LinkedIn URL, company domain.

## Qualify each row
- Drop rows failing any disqualifier. Log how many were dropped and why.
- Flag rows with stale signals (no activity or role change in 12+ months) instead of silently including them.

## Output
A table: Name | Title | Company | Headcount | Verified email | Mobile | Signal | Why they match.
End with: total found, total dropped by disqualifier, and the filter that constrained results most.

3. The cold email skill

What it does: drafts outreach that sounds like your best rep instead of your most average one. It encodes voice rules, your proof points with the numbers attached, and the call to action that fits each stage. Note the negative example inside the file: Claude follows “never produce this shape” far more reliably when the skill shows what bad looks like.

One honest caveat: a writing skill improves drafts, not deliverability. If your domain reputation is weak, better copy lands in the same spam folder. Pair it with proper warm-up and send infrastructure or the words never get seen.

The complete SKILL.md

---
name: cold-email-writer
description: Draft cold outbound emails in the team's voice with strict style rules. Use when the user asks to write, rewrite, or review a cold email or outbound sequence step.
---

# Cold Email Writer

Drafts that sound like our best rep on a good day. About THEM, not us.

## Before writing, require
- Who it is for (role + company), the trigger/context for reaching out now, and the one objection this person will silently have.
If any is missing, ask. Do not draft generic.

## Hard rules
- Under 90 words body. One idea per email. One CTA.
- Subject: 2-5 words, lowercase natural, no clickbait, no "quick question".
- Open with THEIR situation, never with our company name.
- No em-dashes. Max one hyphen in the whole email, only if unavoidable.
- Never: "hope this finds you well", "I know you're busy", "just following up", "touching base", compliments about their posts.
- Every claim carries a number or a name. "Teams like X cut Y by Z" beats "we help teams improve".
- CTA is an easy yes: a question they can answer from memory, not "do you have 30 minutes".

## Bad example (never produce this shape)
"Hi {{first_name}}, hope this email finds you well! I noticed you're at {{company}} and wanted to reach out. We help companies like yours achieve better results. Would you be open to a quick 15-min chat?"

## Output
Three variants: (1) signal-led, (2) problem-led, (3) short/blunt (under 50 words). Label each. Note which objection each is built to defuse.

4. The reply triage skill

What it does: reads the replies your sequences generate and sorts them: interested, objection, referral, out of office, not now, never. For each, it drafts the response and flags the ones a human must own. The 2 a.m. “what does your pricing look like” reply gets a competent draft waiting at 8 a.m.

Why it matters in 2026: reply speed is one of the few outbound variables still moving. The teams compressing time-to-first-response from hours to minutes are winning meetings that used to go cold in the inbox.

The complete SKILL.md

---
name: reply-triage
description: Classify inbound replies to outbound sequences and draft the right response. Use when the user pastes a prospect reply or asks to process an inbox of replies.
---

# Reply Triage

Every reply gets a category, a draft, and a decision: send, edit, or escalate.

## Categories
INTERESTED | QUESTION | OBJECTION | REFERRAL (points to someone else) | NOT_NOW (timing) | NEVER (clear no) | OOO | AUTO/BOUNCE

## Per category
- INTERESTED: draft a reply that proposes two concrete times and confirms the topic in one line. Escalate to the owning rep. Never negotiate price.
- QUESTION: answer only what was asked, in under 60 words, then one forward step. Pricing beyond list price: escalate.
- OBJECTION: use the objection playbook below. One counter, one proof point, one question. Never argue twice.
- REFERRAL: thank, ask for a warm intro in one sentence, and start a NEW thread to the referred person referencing the referrer.
- NOT_NOW: accept cleanly, ask what changes at that date, schedule the follow-up for then. No "I'll circle back sooner".
- NEVER: one-line gracious close. Mark do-not-contact. No sequence re-enrollment.
- OOO: extract return date, shift remaining sequence steps past it.

## Escalation rules (always human)
Pricing/discount requests, legal or security questions, anger, anything from a C-level at a target account.

## Style
Match the prospect's length and formality. Shorter than their message. No em-dashes. Never open with an apology.

5. The CRM hygiene skill

What it does: takes a call transcript or a rep’s rough notes and produces the CRM update nobody wants to write: summary, stage change with reasoning, next steps with owners and dates, and the fields your forecast actually depends on.

Most teams discover while building this skill that they never agreed on what “qualified” means in their process. Building the skill forces the conversation, which is half the value. Over MCP it can write to Salesforce or HubSpot directly instead of producing text a rep pastes.

The complete SKILL.md

---
name: crm-hygiene
description: Turn call transcripts or rough rep notes into structured CRM updates with next steps. Use when the user pastes a transcript/notes or asks to log a call.
---

# CRM Hygiene

The update a sales manager wishes every rep wrote. Facts from the call only.

## Output format
**SUMMARY** (3 lines max): What happened, in plain language.
**STAGE**: Current stage -> recommended stage, with the single sentence of evidence. If no evidence for a move, say "no change".
**MEDDIC-lite**:
- Pain: their words, quoted where possible
- Metric: the number they care about
- Champion: who, and how strong (pushing / neutral / unknown)
- Decision process: who else, what steps, what timeline
- Risk: the one thing most likely to kill this deal
**NEXT STEPS**: Each with owner and date. A next step without a date is not a next step.
**FIELDS**: The CRM field updates as key: value pairs, ready to write.

## Rules
- Only what was said. Uncertain attribution: mark "unclear". Never fabricate a metric or a quote.
- Distinguish prospect statements from rep interpretations.
- If the transcript shows a commitment the rep made ("I'll send pricing by Friday"), it goes in NEXT STEPS with the rep as owner.
- If a connector with CRM write access is available, propose the write and show the diff before committing. Never write silently.

How to install these skills

• Claude Code: unzip into ~/.claude/skills/ (available everywhere) or .claude/skills/ inside a project (shared with your team via git). Claude discovers them automatically; ask it to “prep me for my call with…” and watch the skill load.
• Claude apps (web/desktop): Settings > Capabilities > upload the skill folders.
• API / Agent SDK: attach skills to your agent per the docs.

⬇️ Download all 5 SKILL.md files (all five folders, ready to drop into ~/.claude/skills/)

The part that decides whether any of this works

Notice the pattern in the files above: every skill says “use the live data connector first, never invent a contact.” That single rule is the difference between a system your reps trust and one they quietly stop using. The instructions are yours to write and they are genuinely quick to build. The data is the hard part, and it is where most “AI for sales” setups quietly fail.

• Research skills need live person and company data, not training-data memories.
• List-building skills need a database with verified emails and mobile numbers.
• Triage and CRM skills need read and write access to the systems of record.

MCP connectors solve this cleanly: they give Claude tools it can call, with your permissions, inside the skill’s workflow (we ranked the 7 servers worth connecting, and if you prefer goal-level delegation over chat, Claude Cowork runs these same skills on a schedule). Salesgear ships one that exposes its 800M+ contact database (95% mobile number coverage), person and company enrichment, and per-prospect research as callable tools. If you would rather have the whole loop (research, sequences, dialer, deliverability) in one product with the AI built in, that is what an AI SDR platform is for. And if you are weighing generic AI against purpose-built, we wrote an honest comparison of ChatGPT for sales that applies just as well to Claude.

Frequently asked questions