Internal · HY.AM Studios
Your weekly reference. The trick isn't memorising Claude — it's letting Claude teach you, one prompt at a time. Start with the prompt formula, then watch the basics, look up a connector, copy a prompt.
All six. Connected to Claude. Ready to chain.
Five minutes once. Every chat after is sharper. Most people skip this — and then wonder why their answers feel generic.
Why context matters. Without any setup, Claude treats you like a stranger — generic tone, generic format, no idea what you do for a living. The first 30 seconds of every chat get wasted on you typing "I'm a PM at a creative agency working on…" When Claude already knows your role, your tools, your tone, and your typical tasks, the very first answer is usable. No re-prompting, no "actually I meant…", no boilerplate disclaimers.
Don't write the instructions yourself. Open a fresh chat in Claude and paste the prompt below. Claude will ask you 5–7 questions about your role, your tools, your tone, and your habits — then write the perfect briefing for you to copy.
Copy Claude's draft. Then in the desktop app or browser, go to Settings → General → Profile → Instructions for Claude. Paste, save, done. Every future chat now opens with that briefing already loaded in the background.
You are about to help me write the "Instructions for Claude" field in my Claude settings — the note you'll read silently at the start of every conversation we have from now on.
Interview me with 5-7 questions to learn:
- my role, what I actually do, where I work
- the kinds of tasks I'll bring you most often
- the default tone and format I want (short vs detailed, lists vs prose, formal vs casual)
- anything you should NEVER do (e.g. don't add disclaimers, don't repeat my question back)
- anything you should ALWAYS do (e.g. always end with one decision I need to make)
Then write the instructions in second person ("you") in about 150-250 words. Make them specific and concrete, not generic. I will copy your final output into Settings → General → Profile → Instructions for Claude.
Paste this, answer Claude's questions, copy the output. Then in Claude: Settings → General → Profile → Instructions for Claude → paste → Save. Every future chat now opens with that briefing.
Two patterns turn weak prompts into strong ones. Use them every time.
Pattern 1 · ICC
Every strong prompt has three parts:
Order doesn't matter. Coverage does. Always tell Claude which of your 6 connectors to look at — otherwise it has to guess.
Pattern 2 · Context interview
You'll always forget context you should've included. End every prompt with one line that flips the script:
…before you answer, ask me any questions you need to give a great response — including which of my connected apps you should pull from.
Claude runs a short interview — 3 to 6 questions, often "which connectors should I use?" — then produces an answer tailored to what you actually need. The first output becomes almost always usable.
[INSTRUCTIONS] What I want you to do: … [CONTEXT] What you should know about me, my role, and the situation: … [CONNECTORS] Which of my connected apps to pull from (Gmail · Drive · Calendar · Read.ai · Asana · Slack): … [CONSTRAINTS] How the answer should look (format, length, tone, examples): … Before you answer, ask me 3-6 questions about anything that would help you give a more accurate response — including which connectors I want you to use if I haven't said. Wait for my answers, then produce the output.
Source: Futurepedia, Full Claude Tutorial (02:30 + 04:00). ICC formula + context interview, with the connectors slot made explicit so Claude never has to guess which app to open.
This formula works across all six of your connectors — chain them in one prompt.
Six short Anthropic walkthroughs and two longer deep dives. About 70 minutes total — but the first six add up to ~23.
The fastest "everything-at-once" tour. Where the ICC prompt formula (Instructions · Context · Constraints) and the context interview trick come from. Also covers artifacts, projects, custom instructions, and Cowork.
Cowork is Claude's agentic desktop mode — it actually clicks around your computer to finish multi-step tasks while you do other things. Jeff Su walks through real PM workflows (decks, spreadsheets, research, file ops).
Anthropic's official basics: where to type, how to upload files, how the conversation history works, where the model picker lives. Five minutes, zero jargon.
How to set up a Project per client or workstream: drop briefs into the knowledge base, write project-specific custom instructions, and keep all related chats in one place so context carries across them.
How to add a connector (Customize → Connectors → +), what permissions Claude inherits from the source system, and when a connector decides to fire inside a chat versus waiting to be asked.
Research mode pulls from multiple web sources in parallel and synthesizes one cited answer. Useful for competitor scans, client homework before a kickoff, or any "find me the truth from 10 articles" task.
Sixty seconds on Artifacts: when Claude produces something substantial (a doc, a deck, a spreadsheet, code), it opens in a side panel you can edit, iterate on, and download — instead of getting buried in chat.
Skills are bundles of expertise (instructions + scripts) Claude loads on demand when a task matches. Explains how Anthropic's built-in ones work and why you'll start seeing them auto-invoke in your chats.
Free. No Anthropic account required. If you only do one, do Claude 101.
Everyday Claude for work. The official beginner course — what Claude is, how to talk to it, where to put files, when to start a project.
Open course →How to collaborate with AI effectively and ethically. Frameworks for delegation, framing, and reviewing output. Not tool-specific.
Open course →What today's models actually do, where they break, why they sometimes lie. Useful before you trust output you can't verify.
Open course →Five words you'll use every week, two bonus ones for when you're curious. Skip the rest of the jargon.
A folder for one client or workstream. Drop briefs in once, every chat inside that project remembers them. Five free, more on paid plans.
Read more →When Claude builds something substantial — a doc, deck, contract redline, piece of code — it opens in a separate panel you can edit and download. Not buried in chat.
Read more →A bridge to one of your other apps — Gmail, Drive, Calendar, Slack, Asana, Read.ai. Claude can read and act through it, with the same permissions you already have.
Read more →A note you write once that Claude reads at the start of every chat. Lives at Settings → General → Profile → Instructions for Claude. See the Personalize section above for the prompt that writes it for you.
Read more →Same account, same history, same models. The desktop app adds local-file access and runs Cowork (the agentic mode that uses your computer). Browser is enough for most chats.
Download desktop →Claude's working memory. How much text it can see at once. When it overflows, the oldest stuff falls out — which is why a 4-hour chat starts forgetting things you said early on.
Read more →A pocket of expertise (instructions + scripts) Claude loads on demand when a task matches. Like a specialist it calls in for the right job. You'll start noticing them auto-fire.
Read more →Six connectors live in HY.AM's Claude workspace. What each can do, what each can't, and where to set it up.
The most common worries answered in plain English. If you're unsure, ask in #claude-onboarding.
No. Each person logs into connectors with their own account. Your Drive, your Gmail, your Slack login only feed your own chats. Even inside the shared Ask Your Org project, every member authenticates separately and only sees data they already had permission to see in the source app.
No. There is no background scanning and no indexing. A file is only opened when something in your prompt makes it relevant, for example "find the brief for X". Once Claude has answered, it does not keep watching the folder.
No on our Team plan. Chats and connector data are not used for training by default. Anthropic also does not store the content fetched through connectors. Each query hits the live source and the result lives only in your conversation.
Three options. (1) Keep sensitive files in a folder you never reference in your prompts. (2) Use a separate Google account that isn't connected to Claude. (3) Disconnect any connector at any time from Settings → Connectors, and revoke access in your Google account if you want a hard stop.
Start here. Each prompt does one useful thing with one connector. Replace [brackets] with real names.
GmailPull the 3 most recent emails from [client name]. Summarize each in 2 lines and flag any that mention a deadline, a budget number, or a change of scope.
Replaces 20 minutes of scrolling and triaging an inbox after a day off.
DriveFind the most recent brief in the [client folder] in Drive. Summarize it as 3 bullet objectives, 3 bullet deliverables, and key dates. Tell me if anything in it contradicts the brief from the previous round.
The last line catches scope changes you'd otherwise miss.
CalendarLook at my calendar for the next 7 days. List every external client meeting, who's attending from their side, and any obvious prep task that's implied by the meeting title or invite description.
A 30-second Monday-morning planning ritual.
Read.aiPull my most recent Read.ai meeting with [client name]. Give me a 5-bullet summary, every action item with an owner, and any sentence where the client expressed concern or hesitation.
The "concern or hesitation" line is what most summaries skip.
AsanaList my Asana tasks in the [project name] project, sorted by due date. For each, give me a one-line status based on the latest comment or sub-task progress. Flag anything that has slipped past its due date.
Five seconds to spot what's slipping. Do it before your Monday standup.
SlackSummarize the last 24 hours in #[client channel] in 5 bullets. Flag any unresolved questions and tell me who is waiting on a response from us. Then tell me what I would miss if I only read your summary.
The last sentence makes Claude self-critique. Steals the trick from the workshop demo.
When Claude reaches across two or three tools, the magic compounds. These six are the workshop demos — copy-paste ready.
Pull my latest Read.ai meeting (the one with [client name]). Extract every action item, who owns it, and the deadline implied by what was said. Then draft each one as an Asana task in the [project name] project, with a clear title and a one line description, and show them to me before posting anything.
Read.ai + Asana + Google Workspace. The "show them to me before posting" line teaches the right confirmation habit.
Summarize the last 24 hours in #[client channel] in 5 bullets. Flag any unresolved questions and tell me who is waiting on a response from us. Then tell me what I would miss if I only read your summary.
Catches up on a busy channel without scrolling for 20 minutes.
I am a project manager at a creative agency. You are now connected to my Google Workspace, Slack, Asana, and Read.ai. Look at what is actually in those tools right now and suggest 5 things I could ask you this week that I probably haven't thought of yet. For each one, tell me why it would save me time.
Use this when you're stuck. Claude tells you what to ask it, grounded in your real data.
I have a meeting with [client name] in 30 minutes. Pull the last 3 Read.ai meetings we had with them, the open Asana tasks in that project, and anything from the #[client channel] in Slack from the past week. Give me a one page brief: where we left off, what is currently in progress, open questions from their side, and 3 things I should be ready to answer.
Walks you into a client meeting prepared instead of half-prepared.
Draft a weekly status email to [client name]. Look at the [project name] Asana project for what closed this week and what is in progress. Pull anything notable from our last Read.ai meeting with them. Keep it short, in plain language, no jargon, and end with the one decision I need from them. Show me the draft, don't send it.
A Monday ritual turned into 20 seconds. Always end with "show me, don't send."
Look across my Asana tasks, my Read.ai meeting notes from the last month, and the action items I committed to in #[client channel]. Find things that were promised or assigned more than 2 weeks ago and have not moved since. Group them by who owes what to whom and what the original context was.
The blind-spot prompt. The one PMs lie awake about.
When the basics aren't enough.
Anthropic's official catalog of prompts you can adapt. Look here when you have a recurring task and want a starting point.
Browse the library →Code recipes and examples. Useful for the engineering side of the team when a PM workflow grows into a small automation.
Open on GitHub →Where the workshop questions land, where prompts get shared, where Alican posts updates. Ping there if a connector breaks or you discover a prompt worth keeping.
Open channel →You don't need to write the perfect prompt. Drop your rough task into Claude with the 8-line scaffold above and let it interview you. The output is almost always usable on the first try.
Back to the formula →