WEEK 2 / THE HANDS-ON SESSION

Claude Code

demystified.

SCROLL TO BEGIN

JAMIE'S BIT / THE HANDS-ON SETUP

Claude Code demystified.

"Speak as you might to a young child. Or perhaps as you would to a golden retriever."

— John Tuld, Margin Call

SCROLL TO BEGIN
01 / WHAT IS IT

Not a chatbot. A colleague who builds things.

Claude Code is a CLI agent that lives in your terminal. It reads your files, writes new ones, runs commands, and maintains context across sessions. You talk in plain English — it does the work.

HOW YOU TALK TO IT
PLAIN ENGLISH

Like explaining something over coffee

No syntax. No commands to memorise. Describe what you want the way you'd tell a workmate.

Youbuild me a one-page website that shows my team's wins this quarter. make it look impressive.
HOW YOU BUILD WITH IT
THE LOOP

Describe → Review → Iterate

Claude builds it. You react. It adjusts. Every action asks your permission first.

Youthis looks too corporate. make it more fun — add some animations and a dark theme.
ClaudeI'll restyle it with a dark background, add entrance animations on scroll, and swap the corporate blue for something more energetic. Let me edit the file...
A REAL CONVERSATION
make me a pre-meeting brief for Telstra tomorrow
I'll create an HTML brief. What do you know about the meeting — who's attending and what's the focus?
Sarah Chen, Head of Data. We just finished a POC, they're also looking at Databricks.
Got it. Creating a one-page brief with account context, POC results, competitive positioning, and 3 conversation starters...
✔ Created telstra-brief.html — opening in browser
WHAT MAKES IT DIFFERENT
NOT JUST ANSWERS

It lives in your file system

It sees your files, remembers your preferences across sessions, and takes real actions — creating files, running servers, deploying apps. Not answers. Outcomes.

Where does it live? Terminal — the text box you open with Cmd+Space → "Terminal". Think of it like Slack for your computer. You type, Claude responds. That's the whole interface.
02 / INSTALL

One command. Everything installs itself.

No API keys to find. No accounts to create. One command installs Claude Code, sets up your access via your Salesforce Google account, and you're done.

PREREQUISITES
BEFORE YOU START

Three things you need

1. Your Salesforce-managed laptop
2. Connected to ZScaler VPN
3. Logged into Okta

That's it. No API key, no account, no prior setup. The installer handles everything using your @salesforce.com Google account.
OPEN TERMINAL & RUN THE INSTALLER
STEP 1

Cmd+Space → "Terminal" → Enter

That's your text box. Everything from here is just pasting into this window.

# Paste this and hit Enter — that's the whole install
$ curl -fsSL https://plugins.codegen.salesforceresearch.ai/claude/install.sh | bash
WHAT YOU'LL SEE
==> Claude Code Installer

==> Running pre-install checks
[✔] supported OS
[✔] supported arch
[✔] curl present

==> Step 1: Installing Claude Code
Info: Running Claude Code installer... Done.

==> Step 2: Obtaining LLM key via Google authentication
Opening browser for Google authentication...
BROWSER OPENS

Pick your @salesforce.com account

Google's "Choose an account" screen appears. Select your Salesforce email, click Continue. The tab confirms "Sign-in complete" — close it and the installer finishes on its own.

DONE
[✔] LLM Key has been updated in ~/.claude/settings.json

==> Step 3: Installing salesforce-claude-support plugin
Info: Installed salesforce-claude-support.

==> Installation complete!
If it says "A key already exists": Type y and press Enter to regenerate. That's fine — it just means you or another machine ran it before.
03 / FIRST RUN

Create your folder. Type claude.

Quit Terminal, reopen it, create a project folder, and launch. You're talking to Claude in under 30 seconds.

RELAUNCH & CREATE YOUR WORKSPACE
# Quit Terminal (Cmd+Q) then reopen it, then:

$ mkdir -p ~/claude-projects && cd ~/claude-projects && claude
Why a dedicated folder? Claude can read and write files in whatever directory you start it from. A dedicated ~/claude-projects folder keeps it scoped to work you want it touching — not your SSH keys, browser data, or documents.
YOU'RE IN
Claude Code v1.0.35
Model: claude-sonnet-4-6
Working directory: ~/claude-projects

Type a message to get started.

> hi! i'm new here. what can you do?
USEFUL COMMANDS
VERIFY

/status

Shows your connection status — confirms you're hooked up to the Salesforce gateway.

HELP

/support

Launches the Salesforce Claude support agent — it'll diagnose issues and tell you what to fix.

Stuck? Full setup guide:

Support channel: #claude-code-support

04 / HOW TO TALK TO IT

Talk like a human. Not a programmer.

These prompts build on each other. Copy them in order and watch the conversation compound — from training wheels to something you'd actually use tomorrow.

TRAINING WHEELS
You make me an HTML page that says "hello world" in big bold text with a dark background. Open it in my browser when you're done.

Shows: permissions, file creation, the open command

THE "OH WAIT" MOMENT
You I have a meeting with Telstra tomorrow. Turn this into a one-page meeting brief — use placeholder data but structure it like: what they do, what we sell them, open questions, and 3 good conversation starters. Make it something I'd actually print.

Claude understands intent, not just instructions — it pivots from toy to tool

ITERATE ON SOMETHING REAL
You add a risks section and make the conversation starters more specific to a Data Cloud pitch. Also add a box at the top where I can paste in my own notes about the customer.

Claude extends your idea, keeps full context, edits the existing file

PLAN TOGETHER
You now imagine I had a CSV of all my accounts. Show me how you'd read it and generate a brief for any customer I pick from a dropdown. Don't build it yet — just explain the approach.
What you see throughout: Permission prompts (Allow? Y/N). Claude thinking out loud. Real files in your folder. Each prompt building on the last — prompt 2 pivots completely from "hello world" to a meeting brief. Claude keeps up. That's the model.
05 / PROMPT IDEAS

Prompts to try this week.

You don't need to know what's possible — just describe what would help. Click any card to copy its prompt.

FOR YOU
PERSONAL

Weekly planner

A printable weekly planner you can click to annotate.

SALES

Objection bank

A searchable objection reference with multiple response lengths.

FOR YOUR TEAM
TEAM

Standup formatter

Paste rough notes, get a Slack-ready standup message.

CREATIVE

Email rewriter

Paste a draft, pick a tone, get a rewrite side-by-side.

FOR YOUR DATA
DATA

CSV explorer

Point Claude at any CSV and get an interactive dashboard.

UTILITY

QBR slide builder

A QBR template you fill in — then present as fullscreen slides.

The pattern: Think about a task you do manually this week. Describe it to Claude. "That's too hard to automate" is usually a 2-minute conversation away from a working tool.
06 / SKILLS & MEMORY

Give Claude a brain that persists.

Each Claude session starts fresh — no memory of yesterday. Skills and memory fix this. Skills are reusable commands you trigger with /name. Memory is a file tree Claude reads and writes to remember things across sessions. Set both up now and Claude becomes a partner, not a stranger.

WHAT ARE SKILLS & MEMORY?
SKILLS

Saved instructions triggered by /name

A skill is a markdown file with detailed instructions Claude follows when you type its name. /pause, /remember, /recall — you define what Claude does when triggered.

MEMORY

A folder of notes Claude reads every session

A folder (~/.claude/memory/) where Claude stores decisions, preferences, project context. It reads these at the start of each session — so it remembers who you are and what you've been building.

SET IT ALL UP — ONE PROMPT

Copy this entire prompt. It creates your memory system and all four core skills in one go.

You I want you to set up a personal memory and skills system for me. Here's what I need: 1. Create a memory folder at ~/.claude/memory/ with an _index.md file that tracks recent entries. Categories: projects/, preferences/, decisions/. 2. Create a skill called "pause" at ~/.claude/commands/pause.md that saves a RESUME.md file in my current project folder. It should capture: what we were doing (1-2 sentences), what's done (with file paths), what's next (priority order), key files touched, and any decisions made. Format it as clear markdown. Overwrite any existing RESUME.md each time. 3. Create a skill called "resume" at ~/.claude/commands/resume.md that reads RESUME.md from the current directory, verifies the files still exist, and gives me a brief summary of where we left off. Then asks if I want to continue. 4. Create a skill called "remember" at ~/.claude/commands/remember.md that saves something to my memory folder. It should: pick the right category (projects/preferences/decisions), use a descriptive filename, include today's date, and update _index.md. If a file on the same topic exists, append to it. 5. Create a skill called "recall" at ~/.claude/commands/recall.md that searches my memory folder for a topic and returns what it finds. It should search filenames and contents. For each skill, use this frontmatter format: --- name: [skill-name] description: [one-line description] --- Make the skills thorough — I want them to actually work well, not just be stubs.
HOW TO USE THEM
You (end of session)/pause
ClaudeSaved. You were building the pipeline dashboard — chart done, filter sidebar next. 3 files touched. Resumable anytime.
You (new session, next day)/resume
ClaudeResuming: pipeline dashboard. Chart complete, filter sidebar is next. Files verified. Ready to continue?
You (during work)/remember I prefer dark themes on all my tools and Salesforce blue (#032D60) as the accent colour
ClaudeSaved to preferences/visual-style.md. I'll use this in future sessions.
You (weeks later)/recall what colour do I like?
ClaudeFrom preferences/visual-style.md: dark themes, Salesforce blue (#032D60) accent.
WHEN TO USE EACH
/pause

Ending a session mid-task

Closing your laptop, switching projects, or running out of time. Pause captures exactly where you are so tomorrow's fresh Claude can pick up seamlessly.

/remember

Something Claude should always know

Preferences, decisions, project context. "I work in AEST timezone." "This project uses Streamlit." "My boss prefers one-page summaries." Things that shape how Claude helps you.

/recall

Finding something you saved before

When you know you told Claude something weeks ago but can't remember the details. "What did I decide about the auth flow?" "What's my preferred stack?"

The compound effect: After a week of using /remember, Claude knows your style, your tools, your preferences, and your project context. It stops feeling like a new hire every morning and starts feeling like a colleague who was here yesterday.
07 / YOUR FIRST AGENT

Build an agent. Right now.

An "agent" sounds fancy. It's just a little web app that Claude powers. You describe it, Claude builds it, you open it in your browser.

DESCRIBE IT
You I want to build a simple agent — a chat interface that runs in my browser. Nice dark UI, text input at the bottom, responds using Claude. Make it a Streamlit app. Call it "my-first-agent".
Claude I'll create a Streamlit chat app. Setting up the project structure, installing dependencies, building the UI...
CLAUDE BUILDS IT
# Claude creates these files:
my-first-agent/
  app.py         # the agent
  requirements.txt # dependencies
  run.sh         # start command

# Then installs and runs:
$ pip install -r requirements.txt
$ streamlit run app.py

Your agent is running at http://localhost:8501
What just happened? You described an agent in English. Claude built the UI, the backend, the AI integration. It's running on your laptop. Open that URL — you have your own AI chatbot.
Python errors? Just tell Claude. "I'm getting an error — can you fix it?" It reads the error, diagnoses it, installs what's missing. You don't need to understand the error.
08 / FROM TOY TO TOOL

Now make it do real work.

A chatbot is fun. An agent that preps you for customer meetings — with a Slackbot bridge for real data? That's a tool you'll use every day.

BUILD IT
You (in Claude Code) I want to build a meeting prep agent as a Streamlit app. Here's the flow: I type a customer name. The agent generates a prompt I can copy to Slackbot to gather context. Then I paste Slackbot's response back in. The agent takes that raw info and structures a formatted meeting brief with: account overview, key contacts, recent activity, risks, and talking points. Add a "copy to clipboard" button for the brief and a "copy Slackbot prompt" button for the generated prompt. Dark theme, clean professional layout.
THE SLACKBOT LOOP
You (in your agent) Brief me on Telstra. Meeting with Sarah Chen at 2pm.
Your Agent I need context on Telstra. Copy this to Slackbot:
COPY TO SLACKBOT ↓
Summarise everything I know about Telstra from my recent messages. Include: who I've been talking to, what products discussed, concerns raised, and meeting notes. Raw facts, no formatting.
[Copy Slackbot prompt]
You [pastes Slackbot response back]
Your Agent Telstra — Pre-Meeting Brief

Account: Enterprise, ANZ. $2.4M pipeline. Renewal Q4.
Contact: Sarah Chen, Head of Data Platform
Recent: POC completed May 8, positive on ingestion speed.
Risk: Parallel eval with Databricks on cost.

Talking points:
1. Reference POC win — ask what gaps remain
2. Position zero-copy vs migration cost
3. Agentforce angle for support use case

[Copy to clipboard]   [Export HTML]
The progression: Week 2: manual Slackbot copy-paste (today). Week 3: hook up Claude Code's built-in Slack so it reads channels directly. Week 4: add CRM data. Each week it gets smarter — but useful from day one.
09 / NEXT WEEK

Next week: wire it up.

Next week we connect your agent to Slack. It starts pulling context automatically. The agent does real work without you copy-pasting.

WHAT'S COMING
NEXT WEEK

Connect to Slack

Your agent reads channels, summarises threads, pulls context. No more manual copy-paste.

WEEK 4

Add CRM data

Wire up Salesforce queries. Pre-meeting brief pulls real pipeline, contacts, activity.

WEEK 5-6

Deploy and share

Put it on Heroku. Share with your team. Show and tell.

HOMEWORK (OPTIONAL)

Build guides for agents I've already made. Each walks you through building it from scratch using Claude Code.

STARTER

Claude Code Cheatsheet

Quick reference for commands, prompts, patterns.

OPEN GUIDE →
BUILD YOUR OWN

Agent Hub

Dashboard to manage multiple agents.

OPEN GUIDE →
BUILD YOUR OWN

Work Agent (Cadence)

Chat-first dashboard with routines.

OPEN GUIDE →
BUILD YOUR OWN

Zig (Org62 Agent)

Deal contributions and SE ops from chat.

OPEN GUIDE →
BUILD YOUR OWN

FDE Agent

Salesforce setup assistant with 34 tools.

OPEN GUIDE →
EXPLORE

Agent Hub (live)

See all my agents running. Get inspired.

Login: jamiesagenthub / agenthub2026

OPEN HUB →

WHAT IT LOOKS LIKE WHEN YOU KEEP BUILDING

Agent Hub ● connecting...
Cadence
:8511 — work dashboard
Forge
:8512 — AI network
Zig
:8518 — SE ops
Pres Agent
:8507 — presentations
EA Agent
:8514 — exec assistant
FDE Agent
:8506 — 34 SF tools
Current
:8519 — forecast cockpit
Spark
:8525 — learning