Claude Skills for Sales: The Set Worth Installing

A sales team asking "which Claude skills should we install?" has two answers, and the order matters. Install Anthropic's nine first — they are free, Apache-2.0, and better than anything we would write to compete with them. Then add four of ours, which exist because they do a different job: Anthropic's skills produce work, ours check it.

This page is the curated set. For the pillar view of Claude across a sales org, start at Claude for sales teams. For a per-skill breakdown of Anthropic's nine — what each produces, what it needs connected, where it falls down — read Inside Anthropic's sales plugin.

Start with Anthropic's nine

The sales plugin in anthropics/knowledge-work-plugins ships nine skills and nothing else — no commands, no subagents:

Every one runs standalone; connectors deepen them rather than gate them. Install the plugin, use it for a fortnight, and only then decide what is missing.

Why we did not clone them

The obvious set of sales skills to publish — a prospect researcher, a battlecard builder, a discovery-call prepper, an outreach writer, a deal-review checklist — is exactly the set Anthropic already ships. Publishing near-duplicates would fail our own rule that a skill's description must not apply to any other skill in the library, and it would be worse than the free official version.

So the four below sit beside the plugin rather than on top of it. Every one covers something no skill in the plugin touches: verification and data quality, which is where a developer's habits are worth more than a sales playbook.

The four we added

outreach-claim-checker

outreach-claim-checker takes a drafted email and checks every personalization claim in it before it sends. It pulls each factual assertion out of the draft — the funding round, the job posting, the product launch, the "I saw you moved to X" — and asks what source supports it and how current that source is, marking each one sourced, unsourced, contradicted or stale and returning a send, fix or hold verdict for the email.

This is the direct complement to draft-outreach, which researches and writes but never re-checks. Its own configuration block treats proof points as fill-in-the-blank, so an unsupplied one comes out proof-point-shaped rather than true. One skill writes; the other reads it back adversarially.

cold-email-deliverability-auditor

cold-email-deliverability-auditor is the one with no counterpart in the plugin at all. It walks SPF, DKIM and DMARC records, domain and mailbox warmup, list hygiene, and the content patterns that trip spam filters — then tells you which findings actually block delivery and which are cosmetic.

It is here because a perfectly personalized email that lands in spam is a wasted research budget. This is infrastructure work, and it is the kind of thing a sales team reasonably does not know to check until reply rates collapse.

crm-export-auditor

crm-export-auditor takes a CRM CSV and audits it for duplicate records, close dates that have already passed or were pushed repeatedly, missing required fields, and single-threaded deals with exactly one contact attached.

It pairs with pipeline-review, which analyses pipeline health — but health analysis grades what is in the export and cannot tell you the export is unreliable. Run the auditor first; a forecast built on duplicates and stale dates is confidently wrong, which is worse than obviously wrong.

security-questionnaire-responder

security-questionnaire-responder drafts answers to a prospect's security questionnaire from your own documentation and tags every row: sourced, with the document and section it came from; partial, where the documents answer part of the question and the boundary is stated outright; not applicable, but only when a document establishes why; or needs-review, where nothing supports an answer. A control that appears in no document is left blank rather than asserted.

That tagging is the whole design. A security questionnaire is the one sales document where a plausible-sounding wrong answer creates real liability, so the skill is built to refuse to guess quietly. It also routes anything touching contractual terms, breach-notification windows, data residency or roadmap dates to Legal or Security by name, and cites the SOC 2 rather than pasting it.

Two commands and one agent

Skills fire on their own when a task matches. Sometimes you want to invoke one explicitly, and sometimes the work belongs in its own context window.

• /check-outreach — a thin runner for outreach-claim-checker. Point it at a draft when you want the claim check now rather than hoping Claude routes to it.
• /audit-crm — the same for crm-export-auditor. Useful as a standing step before every pipeline review.
• sales-engineer — a subagent for a prospect's technical asks: security questionnaires, integration requirements, SLA language. It separates what your product does today from what would need engineering work, so an account executive can answer without over-promising. It runs as an agent rather than a skill because that review reads a lot of material you do not want in your main session.

If the distinction between the three formats is still fuzzy, Skills vs Agents vs Commands is the short version.

Installing them

All of this is the same portable SKILL.md format, and it installs three ways:

• claude.ai — zip the skill folder with SKILL.md at its root, upload under Settings, Customize, Skills, and enable code execution under Settings, Capabilities. This is the route for reps who never open a terminal.
• Claude Code — copy the folder to .claude/skills/<name>/SKILL.md in a repo (shared with everyone who clones it) or ~/.claude/skills/ (follows you everywhere). Commands go in .claude/commands/, agents in .claude/agents/. No restart needed.
• Cowork — install Anthropic's plugin from the plugin directory, and add ours as files in the same skills location Cowork reads.

Full detail, including the CLI route and team distribution, is in How to install Claude skills. Note that surfaces do not sync: a skill uploaded to claude.ai will not appear in Claude Code, so install it wherever you actually work.

A sensible order

Run the plugin's skills to make things, and ours to check them. In practice that is: account-research then draft-outreach then /check-outreach before anything sends; crm-export-auditor on the export before pipeline-review or forecast touches it; the deliverability audit once per sending domain and again whenever reply rates drop; and sales-engineer the moment a prospect's questions turn technical.