Thirty things I keep telling executive teams. The moves that turn Claude from a smarter search bar into something your team actually relies on. One line each, one sentence on why. For operators, not engineers.
Team workspace → Projects → "Q2 board prep" (shared, 4 owners)
A shared Project gets better because the whole team edits it. Nine private ones means nine people rewriting roughly the same prompt on a Tuesday and never comparing notes.
"Before you give me this, list the three weakest things about it."
It'll usually catch the hedging, the wobbly stat, the tone that's slightly off. Before you have to. One extra sentence and you save yourself a redraft.
/skills → new skill: "house-voice" upload: style-guide.md + 5 samples
Kills the "can you make it sound more like us" loop. The Skill fires on any writing task without you asking. Set it up once.
Settings → Connectors → Gmail (read scope is enough to start)
Inbox triage, thread summaries, drafts in your voice. It's the connector that does the most for the least setup. Honestly, do this one first and worry about the others later.
1. Inbox synthesis 2. Calendar prep for next week 3. Open follow-ups 4. Status note to your boss
Same prompt, same Project, every Friday. After a month you'll wonder how you used to track any of this in your head (you weren't, really).
"Which of those citations actually exist, and which did you invent?"
Makes Claude flag its own hallucinations before they go out under your name. If you do one thing on this list for outbound research, do this one.
The persistent workspace. Instructions, knowledge, and chat history scoped to one hat you wear. A Project is what stops Claude from feeling like a chatbot and starts making it feel like a coworker who actually remembers things (your voice, your priorities, the team's rules) so you're not re-explaining yourself every Monday. If you're using Claude without Projects you're getting maybe a tenth of what's there. Everything below assumes you've got Project design right.
✓ "Board updates" ✓ "Hiring pipeline" ✗ "Tuesday's email to Sarah"
Projects are for things you do every week. One-offs belong in a regular chat. Don't overthink it.
✓ "Always cite sources. Never hedge." ✗ "This Project is for board prep and tends to use formal tone."
Claude follows orders better than descriptions. Tell it what to do. Don't describe what the Project is for and hope it infers.
Knowledge: 5 of her best memos Instructions: "Match the voice of the attached samples."
You won't have to re-attach a sample. Drafts come out sounding like her from the first turn.
If you'd attach this file twice, it belongs in Project knowledge.
Easiest test I've found for what's worth uploading. Stops the agonizing.
Every quarter: archive any Project you haven't opened in 30 days.
Dead Projects clutter the picker. You'll skip the live ones to scroll past them. Just archive.
Team plan → shared Projects → everyone improves the same instructions
A shared Project gets sharper as people edit it. Nine private versions of the same thing just gets you nine slightly different prompts and no one knows which is current.
The side-panel document. Treat it like a draft, not a message. Different rules apply. Artifacts are where the real work product lives: the brief, the memo, the deck outline, the email going to your board. Most people treat them like chat replies and try to nudge them inch by inch toward what they wanted, which is how you end up with an artifact that nobody wrote. The cheats below are about knowing when to edit, when to torch it and start over, and when to copy the version that's almost right before you ask for one more change.
"Throw this out. Start over from the original brief, but punchier."
Once an artifact has drifted, patching usually makes it worse. Start over from the brief.
"Before showing me, list the three weakest things about this draft."
It'll find them. Then it'll fix them. Costs you one extra prompt and the output is meaningfully better.
Before a major edit: copy artifact → new chat → paste
There's no undo for the version from five edits back. Save a snapshot before you take a big swing.
"Cut this by 40%. Same argument."
Honestly the most underrated command. Every sentence has to earn its spot.
Reusable capabilities that fire automatically when Claude sees the right kind of request. Mostly just markdown. You don't need to write code. A Skill is how you turn "the way our team writes," or "how we run a quarterly review," or "how we draft an SOP" into something Claude reaches for on its own. The nice thing is they work across every conversation in every Project. Write it once, your teammates get the benefit without ever knowing the instructions exist.
house-voice.md: - Active voice. Short sentences. - Never use "leverage" as a verb. - Open with the surprise, not the setup.
Kills the "make it sound more like us" loop. Fires on its own whenever you're writing.
brand-voice sop-writer meeting-synthesizer
The built-in ones cover most knowledge work. Try those before you build your own.
Skill: "use this whenever the
user writes outbound email"
Slash: /weekly-update Pick the right shape. Skills for stuff that should just happen in the background. Slash commands for things you want to deliberately kick off.
SKILL.md: description + when to use optional: a few example files
No code. If you can write an SOP, you can write a Skill.
Hooking Claude into Gmail, Calendar, Drive, your CRM. This is where the real shift happens. The difference between "I paste things into Claude" and "Claude already sees what I'm working on." Most of the workflow gains in this cheat sheet need at least one connector. The Friday ritual is impossible without Gmail and Calendar. The cheats below are also about restraint: connect them one at a time, give the smallest scope that works, and revoke the ones you stop using. Connector sprawl is a real problem.
Settings → Connectors → Gmail
Does the most work for the least setup. Do this one first. Everything else can wait a week.
Week 1: Gmail Week 2: Calendar Week 3: Drive (read) Week 4: …
Hook them all up at once and you'll use none of them. One a week, and each one becomes a habit before the next arrives.
Scope: drive.readonly (upgrade to write after 30 days)
Worst case stays bounded. You can upgrade to write later, once you've gone a month without needing it.
"Prep me for tomorrow: meetings, attendees, recent threads, asks."
The two together do most of what a chief of staff does on a Sunday night brief. Attendees, recent context, what they're likely to ask about.
Settings → Connectors → Revoke
An unused connector is just surface area for something to go wrong. You can always reconnect it later.
The rituals. What turns Claude from a thing you open into a coworker you rely on. Items tagged ⏱ Scheduled task should run on a recurring schedule. Use Claude's scheduled tasks — set it once and forget it. Items tagged ⚡ Event-triggered fire after something that just happened (meeting ends, email arrives) and shouldn't be on a clock.
Schedule: Friday 4pm, weekly. Same Project, same prompt: inbox → calendar → follow-ups → status
Same prompt, same Project, every Friday. Schedule it so it runs whether you remember or not (you won't).
Schedule: Friday 3pm, weekly. "Scan my sent mail this week. What did I promise and not ship?"
Probably the most useful prompt on this whole page. Run it weekly so the things you forgot surface before next week buries them.
[paste transcript] "Draft the follow-up note + owners."
Don't schedule this one. Fire it the second the meeting ends. A follow-up that goes out in ten minutes is worth ten that go out tomorrow.
Schedule: Monday 8am, weekly. "Given my calendar and open threads, what's most likely to slip this week?"
Your calendar tells you what's booked. Claude tells you where you're exposed. Run it before your first meeting so you're not finding out at 4pm Thursday.
Where most knowledge workers spend most of their day. Also where the slop risk is highest, in roughly equal measure. The cheats below are about getting volume without losing the things that make the writing yours: the critique pass before you ship, the source-checking habit, writing the first 100 characters of anything outbound by hand. In my experience the teams that ramp output and keep their reply rates up are the ones who obsess about the editorial step. The ones who don't, watch their voice flatten out and their open rates drift down inside a quarter.
Pass 1: "Gather. Don't synthesize." Pass 2: "Now synthesize. Cite Pass 1."
Mixing gathering and synthesis is how you get confident-sounding nonsense. Separate the passes and you'll catch it.
"Which citations actually exist, and which did you invent?"
Makes Claude flag its own hallucinations before they ship under your name. Non-negotiable for anything outbound.
First 100 chars + last 2 lines = always hand-written.
The opening and the close are what people remember. Write those yourself. Claude can have the middle.
"Give me this at 80 words, 200, and 500. I'll pick and edit."
Picking from three drafts is faster than asking for revisions to one. Pick the closest one and edit from there.
How this becomes the team's default way of working, not a thing five people use in private. Every section above is about what one person can do. This one's about whether any of it survives the first quarter. The pattern that works is boring: one named owner of the shared Projects and Skills, one live demo a week at all-hands, shared workspaces instead of a Slack channel full of clever prompts nobody ever reuses. The pattern that fails is the "AI tips" channel and the hero exec whose workflows leave with her.
5 min. Real workflow. No deck. Different person every week.
Decks teach nothing. Watching a coworker do the thing in 90 seconds teaches more than a whole onboarding doc.
Owner: maintains shared Projects, Skills, connectors, governance.
Without one owner the shared stuff becomes everyone's problem, which means nobody's. Pick a person, give them an hour a week.
Force tips into shared Projects and Skills. Slack tips evaporate.
Tips in Slack die in 48 hours. Tips baked into a shared Project actually stick around.
The 14-module Cowork course goes deep on every cheat above, plus the rollout playbook, governance, and the measurement stuff that makes this actually stick across a team.