A practical guide to working with Claude at Stratus: what it can do, how to set it up, and how the pieces fit together. Written for sales and ops, no technical background assumed.

Start with the Introduction then work through the guides in order.

This set explains how to actually work with Claude at Stratus. If you've used Claude to write an email or summarize a thread, you've seen maybe a tenth of what it does. These pages cover the rest: how to connect it to our real systems, how to teach it the way Stratus does things, and how to hand it recurring work so it runs on its own.

No technical background needed. If you can describe a task in plain English, you can do everything in here.

Who this is for

Everyone. The examples are more Sales-focused (MEDDPICC scoring, territory rules, the ICP, closed-won handoffs, the CRO brief), but the concepts apply to all job functions.

The whole thing in one sentence

Connect your tools, give Claude the context in a Project, teach it the Stratus way with a Skill, and put the recurring work on a Schedule. In Chat when you're in the flow, in Cowork when you want it handled.

The mental model

There are two places you use Claude and four building blocks that make it powerful. Everything in this set is one of those six things.

Two surfaces:

• Chat is Claude in the flow of work. Fast, one thing at a time, you stay in the loop. Best for the moves you make between calls.

• Cowork is the delegation agent. You describe an outcome and it runs the multi-step job in the background, asking before anything irreversible. Best for the work that eats an hour of clicking.

Four building blocks (these are the four component guides):

• Connected Tools give Claude the data: Salesforce, M365, ZoomInfo, and the rest, working with live data.

• Projects tell it what it's working on: the standing context every chat should start with.

• Skills tell it how we do a task: the MEDDPICC rubric, the handoff format, our outbound voice.

• Scheduled Tasks decide when it runs: recurring jobs on a clock.

And it all comes back as an Artifact: the deliverable Claude builds (a brief, a battlecard, a scored deal, a handoff doc) that you refine, keep, and share. That's the thing you actually walk away with.

You don't need all of this for every job. A quick question needs none of it. The leverage comes from stacking the pieces on the work that matters, and the last guide, How It All Works Together, shows exactly how.

How to read this set

If you only have five minutes, read the Quick Reference. If you want the full picture, read all eight in order. It's about 45 minutes end to end.

1. Claude at Stratus: Quick Reference: the two surfaces and the five power features, at a glance

2. How to Build a Skill: teach Claude a Stratus procedure once

3. How to Set Up a Project: give Claude standing context

4. How to Schedule a Task: put recurring work on a clock

5. Applications and Data Claude Can Access: connectors and MCP, in plain terms

6. Building Design files, Skills, and Scheduled Tasks: the branded slide deck from the session, with the full interactive presentation attached to download

7. Standardized Brand System (Claude Design): make on-brand one-pagers, decks, and assets

8. How It All Works Together: the pieces as one system

Questions: Ben Marshall, GTM Engineer.

Two surfaces, five power features. This is the ten-second version: what Chat and Cowork each are, when to reach for which, and the five capabilities that turn them from a helper into a system. The guides that follow go deep on each piece.

The two surfaces

Chat is Claude in the flow of work. Cowork plans and runs multi-step jobs while you step away. Both live in the Claude Desktop app.

	Chat, conversational	Cowork, delegation agent
In one line	Claude in the flow of work. Ask, draft, summarize, edit.	Describe an outcome; Claude plans and runs a multi-step job across your files and connected tools.
Best for	Quick questions, drafts, formulas, summaries, one-step assists.	Multi-step jobs you'd otherwise click through across Salesforce, HubSpot, monday, Excel, and other tools.
At Stratus	MEDDPICC gut-check on an AE's open deal Rewrite an outbound email or a QBR talking point Write a SOQL query or a Salesforce report formula Summarize a 40-message Teams thread for your manager	Run the daily sync: pull Teams, email, and call transcripts; draft GME tickets Generate the CRO Daily Brief and leave it ready to review Auto-build a closed-won handoff doc from the Salesforce record Pre-event campaign load and ICP scrub
Where to find it	claude.ai on web, the Desktop app (Chat tab), and mobile. Embedded in Word, Excel, PowerPoint, and Outlook.	The Claude Desktop app, middle section.
How you work	You stay in the loop: one message at a time, back and forth.	You delegate: describe the task in plain English, review the plan, let it run.
Data access	Connectors (Salesforce, HubSpot, monday, ZoomInfo, Confluence, M365) via MCP, uploaded files, projects, and web search.	Local folders (with your permission), the same connectors, Claude in Chrome / Excel / PowerPoint, plus web.
Outputs	Answers, artifacts (docs, HTML, tables), and in-app edits in Word / Excel.	Real artifacts delivered to your files: spreadsheets, reports, decks, created CRM records, and drafted (unsent) emails, with approval gates.
Honest gotchas	Single-step. Won't autonomously execute a multi-step job, and needs connectors set up to touch work data.	Burns usage faster; needs the desktop app open for local files; approval gates can slow you down; web and mobile still in beta.

The five power features

These are what turn Chat and Cowork into a system rather than a smart writer. Each has its own guide in this set.

Connected Tools: the access

Claude reaches into Salesforce, M365, ZoomInfo, your call recorder, and the rest, working with your live data. Set them up in Settings, Connectors; Chat and Cowork share the same connections. Gotcha: nothing works until a connector is set up and authorized, per user.

Skills: reusable playbooks

Instructions Claude loads automatically when a task calls for them, a way to encode how Stratus does something once. Add them in claude.ai under Settings, Features; Cowork picks them up through plugins. Example: a closed-won handoff skill, or a MEDDPICC rubric applied the same way every time. Gotcha: custom skills are per-user on claude.ai; there's no org-wide admin push yet.

Projects: shared, persistent context

A space that remembers background knowledge and instructions across chats, so Claude starts every conversation up to speed. Lives in Chat (claude.ai and Desktop); shareable on Team / Enterprise plans. Example: a Territory Model project loaded with the Confluence pages and the region-mapping standard. Gotcha: it's retrieval-based, so keep it curated and current, not a dumping ground.

Artifacts: deliverables you iterate on

Content built in a side panel you can edit, refine, reuse, and share: docs, tables, dashboards, small apps. Example: the CRO Daily Brief HTML, account battlecards, and this one-pager. Gotcha: artifacts can't save data on their own; treat each version as a fresh generation.

Scheduled Tasks: work on a cadence

Cowork jobs that run on a schedule, even with your laptop closed, so recurring work happens without you. Type /schedule in any task, or manage them under "Scheduled" in the sidebar. Example: the 6 am daily sync that feeds the monday ticket system; a Monday-morning CRO Brief refresh. Gotcha: cadence-based, not event-triggered; local-file tasks need the desktop app reachable.

Which one, in a sentence

• Use Chat for a one-step assist in the flow of work.

• Use Cowork for a job that would take an hour of clicking: one ask, done in the background.

• Add a Skill when you've explained the same procedure more than twice.

• Add a Project when the work needs the same context every time.

• Schedule it when a Cowork job should run every morning or every week.

Sources: support.claude.com, docs.claude.com, anthropic.com. Prepared by Ben Marshall, GTM Engineer.

Teach Claude the way Stratus works once, then never re-explain it. A skill packages a procedure so Claude applies it the same way every time. This walks through what a skill is, when to build one, and three you could stand up this afternoon.

What a skill is

A skill is a small folder (at minimum one file called SKILL.md) that packages a procedure Claude should follow for a specific kind of task. Think of it as a playbook you hand a sharp new hire: "when you do X, here's exactly how we do it at Stratus."

Claude reads the skill's name and description at the start of every conversation, then loads the full instructions only when a task actually calls for them. That means you can encode a lot of detail without slowing anything down. You've already used skills without knowing it: the ones that build Word docs, Excel models, and PowerPoint decks are pre-built skills working behind the scenes.

Two kinds are worth knowing: pre-built skills (document creation and similar, automatic, no setup) and custom skills (the ones you write to capture how Stratus does something: a scoring rubric, a handoff format, our outbound voice).

Skill, project, or just a prompt?

They overlap. A quick test:

• Just prompt it. It's a one-off, or you'll never do it again. Don't build anything.

• Build a skill. There's a repeatable procedure with a right way to do it: "score this deal on MEDDPICC," "write this in our outbound voice," "format a handoff doc." The trigger is simple: you've explained the same how-to more than twice.

• Build a project. A body of work needs the same background knowledge every time (the territory model, the ICP). Skills are the how; projects are the what-you-need-to-know. They pair well, since a project can carry a skill.

Build your first skill

A skill is just a folder with a SKILL.md inside. The file has two parts: frontmatter (name and description) and a body (the actual instructions). The description is the most important line you'll write. It's how Claude decides when to reach for the skill, so say both what it does and when to use it.

Here's a complete, working example:

---
name: meddpicc-scorer
description: >
  Scores an open opportunity against the Stratus MEDDPICC rubric and
  flags gaps. Use whenever someone asks to qualify, score, or gut-check
  a deal, or pastes discovery notes or a call transcript.
---

# MEDDPICC Scorer

## Instructions
Score the deal 0-2 on each element: Metrics, Economic Buyer,
Decision Criteria, Decision Process, Paper Process, Identify Pain,
Champion, Competition.

- 0 = no evidence · 1 = partial · 2 = confirmed with a name/quote
- Never infer a score without evidence from the notes. If it's
  missing, score 0 and say so.

## Output
A short table (element / score / evidence), a total out of 16, the two
weakest elements, and the single next action to move the deal forward.
Deliver it as an artifact so the rep can drop it into Salesforce.

To add it in Chat: go to Settings, Features and upload the skill as a .zip. From then on Claude uses it automatically whenever a matching task comes up. In Cowork: skills travel inside plugins. A plugin bundles skills together with connectors and can be installed for a whole workflow.

Three Stratus skills to start from

Each is a starting point. Edit the rubric, format, and voice to fit.

1. MEDDPICC Scorer

Use it when qualifying or gut-checking a deal, or pasting in discovery notes. It encodes the Stratus 0-2 rubric and the "no evidence, no score" rule, and outputs a scored artifact (table, total, next action) ready to paste into Salesforce.

2. Closed-Won Handoff

Use it when a deal closes and needs a clean sales-to-CS handoff. It encodes the post-sale handoff-doc format (scope, stakeholders, commitments made, risks, success criteria) and outputs a formatted handoff doc that's consistent across every AE.

3. Stratus Outbound Voice

Use it when drafting or rewriting AE outbound and follow-ups. It encodes our tone, structure, length, and the phrases we do and don't use, so one rep's rewrite and another's sound like the same company.

Give Claude the Stratus context it needs once, so every chat starts already up to speed. A project is a workspace that carries your background and your rules into every conversation inside it. This covers what a project is, when to build one, and three you could stand up this afternoon.

For sales and ops. About 6 minutes. Needs a paid plan; sharing needs Team or Enterprise (confirm our plan).

What a project is

A project is a dedicated workspace that remembers two things for every conversation inside it:

• Project knowledge: the documents and reference material you add, like the ICP guide, the territory model, or an account's history. Claude looks things up here automatically.

• Custom instructions: standing guidance for how Claude should behave in this project, covering tone, format, rules, and what to never do.

Everything you chat about inside the project inherits that context. Instead of re-pasting the ICP or re-explaining the six-leaf model every time, you set it up once and every chat begins informed. Your related chats also stay grouped in one place instead of scattered across your history.

Project, skill, or just upload a file?

• Just upload a file. You need Claude to read one document for one conversation. Drag it in; no project needed.

• Build a skill. There's a repeatable procedure with a right way to do it (see the Skills guide).

• Set up a project. A body of work needs the same background every time: an account you're working for months, the ICP every prospecting chat should respect, the territory rules every ownership question depends on.

Set up your first project

Create the project, then fill the two slots, knowledge and instructions:

1. Create it. In Chat, open Projects, New project. Name it for the body of work ("ICP & Prospecting," not "misc").

2. Add knowledge. Upload or link the reference material Claude should always have on hand: guides, exports, prior briefs, the relevant Confluence pages.

3. Write custom instructions. Tell Claude how to operate in this project every time. Keep it tight.

4. Work in it. Start chats from inside the project. Each one arrives already knowing the context.

An example of custom instructions for a prospecting project:

Role:
You support Stratus AEs qualifying accounts against our ICP.

Always:
- Apply the Stratus ICP and the 2-of-4 Enterprise segmentation rule
  from the project knowledge. Never loosen the criteria to force a fit.
- When an account qualifies, state which 2+ criteria it meets, with
  evidence. When it doesn't, say so plainly and why.
- Default output: a short qualification summary as an artifact the rep
  can attach to the Salesforce account.

Never:
- Invent firmographics. If ZoomInfo/Salesforce data is missing, flag the gap.

Three Stratus projects to start from

Each is a starting point. Swap in your own accounts, docs, and rules.

1. Territory Model

Holds the Confluence territory pages, the regional mapping standard, the six-leaf AE assignment and national-account overrides, so that any ownership question ("who owns this account?") is answered against the real rules, not a guess.

2. ICP & Prospecting

Holds the Prospecting & ICP Qualification Guide and the 2-of-4 Enterprise rule, so that every qualification chat scores accounts the same way, and produces a consistent summary artifact for Salesforce.

3. Account War Room

Holds everything on one strategic account: battlecard, call summaries, key emails, stakeholder map, open commitments, so that every chat about that account is instantly current, and the deal artifacts stay in one place through close.

Set the recurring work up once (the morning brief, the weekly digest, the daily sync) and let it run in the background. This covers what a scheduled task is, when to use one, and three you could turn on this afternoon.

For sales and ops. About 6 minutes. Needs a paid plan with Cowork (confirm our plan).

What a scheduled task is

A scheduled task is a Cowork job set to run on a clock (the same task, every morning or every Monday) without you kicking it off. You describe the work and the cadence in plain English; Cowork runs it in the background and leaves the output waiting for you.

This is a Cowork capability, not a Chat one. Chat is one conversation at a time; Cowork is the delegation agent, and scheduling is what turns a good one-time delegation into standing infrastructure. The daily GME sync you already rely on is exactly this pattern.

When to schedule (and when not to)

• Schedule it. The same job needs to happen on a rhythm: a 7 am brief, a Monday pipeline digest, an end-of-day ticket sweep.

• Just delegate it once. A one-off multi-step job. Run it in Cowork without a schedule.

• Don't reach for scheduling when you need event-triggered automation ("fire the moment a deal hits closed-won"). Scheduling is cadence-based; true event triggers are a different, more technical build. Keep that distinction honest in any leadership pitch.

Set up your first scheduled task

1. Describe the task and the cadence together in Cowork, or type /schedule to start one. Detail helps: sources, format, length, timezone.

2. Let Cowork set up the schedule from your description.

3. Manage it from the Scheduled section in the sidebar. Pause, edit, or delete anytime.

4. Decide delivery and approvals. Choose where the output lands. Cowork still asks before anything irreversible (like sending an email), so a "draft and send" step may wait for your OK unless you've set trust levels for it.

A scheduled Cowork prompt for a morning sales brief might look like this:

Every weekday at 7:00 am Arizona time, prepare my sales brief:

Pull from:
- Salesforce: my open opportunities and any that changed stage,
  amount, or close date in the last 24 hours
- Outlook + Teams: overnight messages from anyone on an active deal
- Calendar: today's meetings with external attendees

Produce a one-page brief as an artifact:
1. Deals that moved (and how)
2. Anything that went quiet 7+ days
3. Today's calls, with a one-line prep note each
4. The single most important next action

Save it to my Daily Briefs folder. Don't send anything externally.

Three Stratus schedules to start from

Each is a starting point. Edit the sources, cadence, and delivery to fit.

1. AE Day-Starter

Daily, 6:30 am. Scans your own book: today's meetings with a prep line each, opps closing this week, replies you owe prospects, and deals gone quiet 5+ days. Delivers a "here's your day" action list, priorities first.

2. Friday Forecast & Hygiene Prep

Weekly, Friday 2 pm. The pre-1:1 sweep: catches hygiene gaps (missing next step, stale close date, no activity 10+ days), flags slipping deals, and drafts talking points. Delivers talking points ready before the 1:1, so pipeline reviews start from a clean book.

3. Monday Pipeline Digest

Mondays, 7 am. Reads Salesforce for week-over-week movement: deals advanced, slipped, or gone quiet across the six AEs. Delivers a one-page digest artifact for the team channel, so the weekly sync starts from facts.

Claude can reach into the tools Stratus already runs on: Salesforce, M365, your call recorder, ZoomInfo, and more. This explains what it can see, where it can see it, what it can do with it, and how to connect a new tool.

What "connected tools" means

A connector is a secure link between Claude and one of your systems: Salesforce, HubSpot, monday, Microsoft 365, and the rest. Once it's connected, you can ask Claude to look something up or do something in that tool in plain English, and it works directly with the live data. No exporting, no copy-paste.

Connectors are what turn Claude from a smart writer into something that knows your actual pipeline, your real accounts, and last week's calls.

What Claude can access, and where

Both Chat and Cowork use the same connectors. Cowork adds a few things only it can reach (your local files and your browser) because it runs on your desktop as a delegated agent.

What Claude can reach	In Chat	In Cowork
Connected tools (Salesforce, HubSpot, monday, ZoomInfo, Confluence, M365, Slack, call intel)	Read & write, with you in the loop each step	Read & write across multiple steps; approval before anything irreversible
Files you upload	Yes, drag them in	Yes, drag them in
Local files & folders on your machine	No	Yes, on the desktop app, with your permission
Your live browser session	No	Yes, via Claude in Chrome
The web (search + read pages)	Yes	Yes
Project knowledge	Yes	Shared across both

Read vs. write, and who Claude acts as

Reading is the easy part: ask, and Claude pulls what's there. Writing (creating a Salesforce record, posting to Teams, sending an email) is where the guardrails live:

• In Chat, Claude makes changes when you ask it to, in the moment. You see each action as it happens.

• In Cowork, Claude can chain many steps on its own, but it pauses for your approval before anything irreversible: sending, publishing, deleting. You can raise or lower those trust levels per action type.

Is your tool compatible? MCP, in plain terms

Most of what Stratus runs on already connects. When you hit a new tool, here's how to know whether it can, and how to wire it in.

What makes it possible: MCP. Claude talks to your tools through the Model Context Protocol, an open standard Anthropic introduced in late 2024. Instead of a one-off integration built for every app, a tool implements MCP once and becomes available to Claude. A "connector" is simply Claude hooked up to a tool's MCP server.

Three ways a new tool shakes out:

1. In the directory. Open Settings, Connectors and browse. If it's listed, it's one click, and the directory page spells out that tool's read/write capabilities before you connect. (Salesforce, HubSpot, monday, ZoomInfo, M365, and Confluence all live here.)

2. Has an MCP server but isn't listed. Add it as a custom connector using the vendor's remote MCP server URL. Anything that supports MCP can connect this way, including internal tools you build yourself.

3. No MCP server. It can't connect directly yet. Sometimes you can still reach its data through a system that is connected, but there's no native link.

What it gives access to: exactly what that server exposes, a defined set of read and (often) write actions, nothing more. You can set each action to always allow, ask first, or never.

Getting it connected:

1. Settings, Connectors. Browse the directory, or choose Add custom connector and paste the remote MCP server URL.

2. Authenticate with OAuth. You sign in to the tool itself; Claude never sees your password.

3. Enable it per conversation from the "+" menu in the chat.

4. Test with a small read ("list my three most recent opportunities") to confirm it's live before you trust it with real work.

An example of adding a custom (internal) connector:

Name: Stratus Labs
URL:  https://internal-mcp.example.net
Auth: OAuth (or a shared API key via request header)

The Stratus stack at a glance

What each core connector is best for:

Salesforce	The deal system of record. Read pipeline, accounts, and opportunities; create and update records.
Microsoft 365	Outlook, Teams, calendar, SharePoint. Both the context ("what's happened on this account") and the delivery ("post this to the channel").
Call intelligence (Fireflies / Granola)	Meeting transcripts and recaps. Feed discovery straight into MEDDPICC scoring or a follow-up.
ZoomInfo	Firmographics and contacts. Enrichment for prospecting and battlecards (read).
HubSpot	Marketing attribution and campaign membership.
Confluence	The GIKB knowledge base: runbooks, the ICP guide, the territory model.
monday.com	The GME ticket system. Create and track work items.
Slack	Team comms and quick posts.

"Working in Claude" series presentation that walks through the three ways to put Claude to work at Stratus, plus a hands-on section for building your first skill. This is the deck delivered in the Stratosphere Learning Series session.

Deck:

Five parts:

#	Section	What it shows
01	Design	Making an on-brand asset in Claude Design, and working on one with a colleague
02	Skills	Building a skill live, then saving, sharing, and finding skills
03	Putting them together	The Sales Command Center project and a weekly brief that runs on a schedule
04	What to watch for	Where Claude gets things wrong, and how to catch it
05	Your turn	Pick a task you repeat and build a skill for it

Working in Design:

The one thing to remember when building a Skill: have Claude interview you

The fastest way to build a good skill is to make Claude ask you the right questions before it writes anything. Paste this, replacing X with the task you want to write down:

I need to do X.

Before you write anything, interview me. Ask three questions at a
time, wait for my answers, and keep going until you know:

  - when I'd use this and what I'd say to trigger it
  - what I'll hand you each time
  - what you should give back: format, length, where it goes
  - what good looks like (ask me for a real example)
  - the rules, including what you should never do

When you have enough, tell me. Then draft the skill and run it on
my example so we can see if it works.

Writing a Skill description that gets used

Claude reads the description of every skill you have switched on and picks the one that fits what you asked for. You never call a skill by name. That makes the description the part that matters, and the part most people underwrite. Say what the skill does, then when to use it. The "use when" half is what makes it fire.

These get picked up:

• "Turns raw weekly notes into a Friday status update: use when I paste bullets and ask for an update"

• "Builds a pre-call research brief: use before a first call when I give a company name"

• "Categorizes support tickets and drafts a first response: use when I paste a list of tickets"

• "Cleans and dedupes an event attendee list: use when I upload a registrant list"

This one will not, because nothing tells Claude when it applies: "Writes updates".

What to watch for with Claude in general

The tools shown

• Claude Design: make on-brand one-pagers, decks, and assets. Our brand system is loaded, so output comes out in Stratus colors and fonts automatically. See also Standardized Brand System (Claude Design).

• Claude Projects: a workspace that already holds your context, so every chat inside it starts with the same instructions and reference material.

Questions: Ben Marshall, GTM Engineer.

A quick guide to making on-brand documents, one-pagers, and decks with Claude Design.

This is our shared brand kit inside Claude Design. Point Claude at it and anything you generate — a sell sheet, a pitch deck, a letter — comes out in Stratus colors, fonts, and layouts automatically. No design skills needed, no guessing at hex codes.

One Rule: always use Stratus GTM Brand System (marked Org default) for asset creation.

1. Open Claude Design

Open the Claude app. In the bottom-left of the sidebar, click Design. This opens the Design workspace ("What should we create?").

Screenshot redacted

Screenshot redacted

Screenshot redacted

From this panel you can also hit Open source file to browse the full kit (colors, fonts, logos, templates) — covered in Section 5.

3. Pick a starting point

There are two ways to create:

Screenshot redacted

B. Just describe it. Type what you want in the prompt box (e.g. "one-pager on our estimating module for electrical contractors") and send. With the brand kit selected, Claude applies our styling even without a template. Optionally (and suggested) do all your planning in Claude Chat first then ask it to give you instructions to hand Claude Design.

Then click the send arrow to generate. Edit by chatting — "make the headline shorter," "swap the photo," "add a stats row."

4. Which template do I use?

Template	Use it for	Output
Stratus Letterhead	Formal letters, cover pages, official correspondence	Branded US-Letter page — GTP contact header, horizon divider, s·f·w footer. Prints/exports to PDF.
Stratus One-Pager	Sell sheets, leave-behinds, single-topic overviews	One or two US-Letter pages — navy header, value props, stats, software photo, yellow CTA. Prints/exports to PDF.
Stratus Pitch Deck	GTM sales presentations	1280×720 slides — navy title/section slides, light content slides, brand-ready.

5. Aspects of the brand foundations (reference)

You don't need to memorize any of this — Claude applies it for you. It's here so you know what "on-brand" means and can spot when something's off.

Screenshot redacted

Screenshot redacted

Color. Navy is the foundation, blue is the accent, yellow is the call-to-action. Core values:

Token	Hex	Use
Navy	#11263D	Primary / backgrounds
Navy deep	#0D1E32	Darkest sections
Accent blue	#1F487F	Accents, links
Light accent	#77A8D8	Tints, charts
CTA yellow	#E6DC16	Buttons / calls-to-action only
Off-white	#F2F1F0	Surfaces
Near-black	#010101	Body text

Screenshot redacted

Screenshot redacted

Screenshot redacted

6. Do & don't

Do

• Use Stratus GTM Brand System (Org default) every time.

• Start from a template for anything customer-facing.

• Reserve CTA yellow for buttons and calls-to-action, not decoration.

Don't

• Use the legacy "Stratus Design System."

• Set body text in Montserrat — Lato only.

• Hand-pick off-brand colors; the kit's tokens are the palette.

• Swap real client photography for stock.

7. Questions

Reach out to Ben Marshall (GTM Engineer) for access issues, new templates, or brand questions. To see the full kit, open Claude Design → Design system → Open source file.

Four building blocks, two places to use them. This guide ties the set together: what each piece does, how they stack, what you can get done in Chat versus Cowork, and how the team shares it all.

For sales and ops. About 10 minutes. Read the four component guides first (Skills, Projects, Scheduled Tasks, Connected Tools).

The building blocks

Four of these are how Claude works; the fifth is what you get back.

• Connected Tools: the access. Claude reaches into Salesforce, M365, ZoomInfo, your call recorder, and the rest, working with live data.

• Projects: what it knows. The background context (the ICP, the territory model, an account's history) that every chat inside the project starts with.

• Skills: how it works. The Stratus procedure for a task: the MEDDPICC rubric, the handoff format, our outbound voice.

• Scheduled Tasks: when it runs. Cowork jobs on a clock, working in the background so recurring work happens without you.

• Artifacts: what you get out. The deliverable Claude builds in the side panel (a brief, battlecard, scored deal, handoff doc) that you refine, keep, and share. It's the output the other four produce.

How they stack

They aren't separate features. They're layers of one system, and they resolve into a deliverable. Read them as a sentence:

Read bottom-up, it's access, then knowledge, then method, then timing, and the artifact comes out the top:

1. Connected Tools: the live data underneath

2. Projects: what it's working on

3. Skills: how we do the task

4. Scheduled Tasks: when it runs

5. Artifact: what you walk away with

You don't need all five for every job; a quick question needs none. But the real leverage comes from stacking them.

A worked example: the Account War Room

Project	One strategic account: battlecard, call summaries, stakeholder map, open commitments.
Connectors	The live picture on demand: latest Salesforce opps, new ZoomInfo firmographics, this week's calls.
Skill	The MEDDPICC Scorer keeps every update scored the same way, with evidence.
Scheduled Task	Refreshes the account brief every Monday morning.
Artifact	What lands is the refreshed brief, waiting in the side panel before your week starts.

One account, all five pieces, no manual assembly. That's the pattern to reach for on anything that matters.

What you can get done in Chat

Chat is Claude in the flow of work: you're in the loop, it's fast, one thing at a time. Best for the moves you make between calls.

Move	What it looks like	Uses
Gut-check a deal	Paste your notes: "Score this on MEDDPICC and tell me the weakest spot."	Skill plus your notes, produces a scored artifact
Get current on an account fast	"What's the latest on [account]?" pulls recent Salesforce activity, emails, and calls into one summary.	Connectors
Write in our voice	"Rewrite this outbound so it sounds like us and lands the meeting."	Skill
Turn a call into action	Drop in a transcript: "Pull the MEDDPICC updates and draft the follow-up email."	Call intel plus Skill
Build a battlecard	"Battlecard for [account]" pulls ZoomInfo, Salesforce, and HubSpot into one artifact you keep in the project.	Connectors plus Project
Prep a message up the chain	"Draft a three-line Slack update to your manager on where the [account] deal stands."	Just ask

What you can get done in Cowork

Cowork is the delegation agent: you describe an outcome, it runs the multi-step job in the background and asks before anything irreversible. Best for the work that eats an hour of clicking.

Move	What it looks like	Uses
Start every day pointed at the right work	The AE Day-Starter schedule surfaces today's meetings, closing deals, and owed replies before you sit down.	Schedule plus Connectors
Come back from PTO in minutes	"I was out Monday to Friday. Catch me up." It reads email, Teams, and your calendar and hands back a prioritized action list.	Connectors, multi-step
Walk into the 1:1 with a clean book	The Friday Forecast & Hygiene Prep flags gaps and drafts talking points.	Schedule plus Connectors plus Skill
Assemble a full account package	"Build me everything on [account] for the QBR": a quarter of activity, calls, and pipeline pulled into one deck or doc.	Connectors, produces an artifact
Produce the closed-won handoff	When a deal closes, generate the standardized handoff doc from the Salesforce record.	Skill plus Connectors, produces an artifact

Using it together as a team

The move is to standardize the what and the how, and let each person run their own access and timing.

• Share Projects to align the team. On Team and Enterprise plans, a shared project is the cleanest win going: put the ICP, territory model, and playbooks in one project and everyone works from the same context. No more five versions of the truth. Shared knowledge and instructions are visible to all members, so keep them clean and current.

• Distribute Skills deliberately. Custom skills are per-user on claude.ai; there's no org-wide push yet. To get the team on the same MEDDPICC rubric or handoff format, hand out the skill file (or bundle it into a Cowork plugin) and version it centrally so nobody's running an old copy.

• Schedules are personal by design. Each rep runs their own scheduled tasks in their own seat. A "team digest" is simply one person's schedule that posts to a shared channel, which keeps everyone's brief scoped to their book.

• Connectors act as each person. Every connector uses the individual's own login and permissions, so Claude only ever sees what that teammate can see, and writes are attributed to them. Good for governance, and the reason access is set up per person, not once for all.