Cliff Wilson shares office space with Catalyst and owns the intro — every HuddleTalent ship is Catalyst BD. No cold outreach.
RA
Radio Ads / Phase 1
active
Polish demo, send proposal via the agent, sign. Phase 2 gated on Fusion API + rate cards (Matt investigating).
Pipeline board
active1
RARadio Ads
delivery1
HTHuddle Talent
early2
CGCatalyst Group
CMCanda Media
watch1
OEOxford Economics
Delivery progress
No task data yet.
Open follow-ups
None from recent meetings.
Risks
None flagged.
03Weekly Revenue Reports
Weekly view of pipeline to revenue
Coming soon
Weekly revenue reports will pull the current week's invoices, deal closures, and pipeline shifts into a single view here.
04Sales Methodology
How the team sells -- playbooks and patterns
Coming soon
The sales motion -- how conversations open, qualify, build trust, present the diagnostic, and close. Will live here as a written playbook.
/Marketing
01Content Calendar
0 pieces in the pipeline
The content calendar lives in the team repo's data file and is maintained through the Seby Agent — ask in a Claude session to add, move, or complete a piece.
No entries yet — ask the Seby Agent to add the first piece to the content calendar.
One brand file drives every deck, doc, and email — change it once, everything re-skins.
03Lead Generation
Outbound deck engine
No deck-engagement data yet — the daily insights feed hasn't produced a manifest. Numbers appear after the next feed run.
/Meeting Recordings
Meeting Capture PerformanceLoading live from the brain…
HeldRecordedProcessed by Seby Brain
Loading…
—
Loading meetings…
Connecting to the brain
02Meeting Intelligence
Loading transcripts...
Loading...
Connecting to brain for meeting transcripts
/How to use Claude
01How to use Claude
Recipes for the everyday jobs. Plus what to do when things go wrong.
The 30-second rule
If you only remember three things, remember these:
1. Open the parent project folder when you start Claude Code (not a subfolder).
2. Ask Claude first. Before Googling, before messaging anyone, before opening any other app. Claude knows your business.
3. Run /checkpoint before you close. It saves everything you did, writes a handover for next time, and commits across all repos so nothing is lost.
Everything else lives in the other tabs above. The most common ones are Recipes (how to do specific jobs) and Skills (the slash commands Claude knows).
📝 Change something on a deck
On the [Company] deck, change the headline on slide 2 to '...'. Then tell me the URL so I can check.
Claude reads the file, makes the change, commits and pushes, and gives you the live URL. Changes are usually live within about 60 seconds.
The more specific you are about WHICH deck and WHICH section, the cleaner the result.
🆕 Make a new client deck
I need a new deck for [Client Name]. Start from the [Template] deck and adapt it for [context].
Claude copies the template deck, swaps the client-specific content, and gives you the URL. You then describe the parts you want changed and Claude iterates.
Point Claude at any source material you have -- a LinkedIn URL for the buyer, a website, a recording. The more context, the more tailored the first draft.
❓ Ask a question about your business
What did we agree with [Company] in the last meeting?
Claude has your transcripts, customer files, and notes loaded. It can answer business questions without you having to dig through emails or recordings.
If an answer feels made up, push back -- "are you sure? show me the source."
💾 Save your work before you close
/checkpoint
Claude runs a structured end-of-session protocol: commits any unsaved changes, updates its memory, writes a handover note so the next session knows where you left off, and prints a verified state report.
Make this the last thing you do every session. Saying "save everything" or "wrap up" works too.
🎙 Bring in a voice memo or whiteboard photo
I just recorded a 5-minute voice memo about [topic]. Read it, summarise the three main points, and update [the relevant file].
Drag the file straight into Claude Code. It handles audio, images, PDFs, Word docs, PowerPoints.
For long voice memos, ask Claude to transcribe first, then summarise. Two steps, much cleaner.
⏸ Pause Claude or change direction
Press Esc to stop. Then describe what you actually want.
Claude says yes to every idea. If it's running off in the wrong direction, interrupt it, don't watch it dig deeper.
If something comes up mid-session that you'll handle later, say "park that for later" and Claude will note it without acting.
Claude has specialised skills for your team. Some run automatically when the situation calls for them; others you trigger by typing a slash command or asking in plain English. Skills with a teal border are auto-invoked, the rest you call manually.
Sales & prospecting
/score Lead Scoring Auto + manual
Scores any new prospect 0-100 against five pre-meeting signals (ICP fit, source quality, engagement, champion, urgency). Returns Hot / Warm / Cool / Likely-DQ.
Before your first call with a new prospect, to allocate prep time.
/prospect-research Prospect Research Auto + manual
Generates a full pre-meeting briefing: buyer LinkedIn, company news, tech stack, comparable deals, recommended angle.
Auto-runs when a new prospect meeting is booked. Manual: "research this prospect".
/deck-generator Deck Generator Manual
Authoritative for client-facing decks. Knows your brand, design files, deck-and-proposal standards.
Any deck or proposal work.
Meetings
/meeting-intelligence Meeting Intelligence Auto + manual
Two modes: prep (pre-meeting briefing) and debrief (post-meeting, classifies transcript, extracts decisions and actions).
Auto-runs before scheduled meetings and after transcripts arrive.
Privacy & system
/checkpoint Checkpoint Manual
End-of-session safety protocol. Saves memories, writes a handover, commits across all repos, prints a verified-state report.
Before you close Claude Code. Also triggers on "save everything", "wrap up".
The list is short on purpose. Each of these will create real pain if ignored.
Don't edit files in Notepad or Explorer. Tell Claude what needs changing. Hand-editing breaks the git history and skips Claude's safety checks.
Don't use Chat or ChatGPT for real work. They don't know your business. Claude Code does. The two will give different answers; only Claude Code's answer is grounded in your files.
Don't open a subfolder when you start. Always open the parent project folder. The parent has the brain that tells Claude who you are and how to work with you.
Don't worry about breaking things. Everything is in git. You cannot break anything permanently. If a change looks wrong, just tell Claude "undo that last change".
What you're seeing
What to do
Claude won't respond to a message
Press Esc, wait a second, describe again.
Claude is doing the wrong thing
Press Esc to stop it. Then say what you actually want.
Claude says "I don't remember our last session"
Tell it "read CLAUDE.md and the most recent handover file". Memory loads from those.
The deck or page looks broken after a change
Say "undo your last change" or "revert that". Everything is in git, nothing is lost.
Claude can't push your changes
Say "pull the latest from main, then try the push again". There's just newer work upstream.
The whole thing feels confused
Run /checkpoint, close Claude Code completely, reopen on the parent project folder. Fresh session, full context loaded.
You opened the wrong folder
Close Claude Code. Reopen on the parent project folder (not a subfolder like deck/website/ops).
Wrong people can see a deck
See Privacy & access below.
When you're comfortable with the basics, these unlock more power.
The instruction trick. Type instruction at the start of a big task. It expands to "confirm and hold for instruction before you proceed" -- Claude pauses and waits for your full brief instead of charging ahead.
/checkpoint (or save everything). End-of-session ritual. Saves memories, writes a handover, commits across your repos, prints a verified-state report.
/deck-generator. Use this for any deck work. It knows your brand, design files, and templates.
Ctrl+N for parallel sessions. Run two things at once -- one session researching, the other building.
Record your voice, paste the transcript. Two minutes of talking gets you a better result than ten minutes of typing.
Tell Claude to remember things. "Add this to your memory" or "write this into CLAUDE.md". That memory persists every future session.
Push back when it's wrong. "That's wrong, rethink it" -- Claude will. Don't accept a bad answer.
Send URLs, not files. Your deck is a link. Email it. No PDFs, no downloads.
02Privacy & access
Who can see what, and how you turn the sign-in wall on or off, yourself.
Right now, your deck site may be fully public
Anyone with a URL can see any page. That is often a deliberate choice -- easier to share with prospects.
When you're ready to lock things down, you don't need an engineer. Tell Claude in plain English. Changes go live in about 60 seconds.
The two states (today)
Public
Anyone with the URL can see it. No sign-in. Good for prospect decks you want shared widely or pages like this Command Centre.
Restricted (L3)
Visitor must sign in with Google or an email link. Only allowed emails or email domains get through. Everyone else sees a polite "Not authorised" page.
Pages can mix freely. You could have the homepage public and a specific customer deck restricted to their email domain.
What to say to Claude
Turning the wall on or off completely
Turn auth off for everything.
Turn auth on for everything, use the standard customer-domain rules.
Locking or opening a single page
Make the [Company] page private, only [domain] can see it.
Open up the [Company] page for anyone with the link.
Checking what's set
Show me what's locked right now on prod.
Show me the rules for the [Company] page.
The access diagram
When the 5-tier plan is live, each tier can see everything the tiers below it can see.
5-tier access model (future state)
Owner-only
Founders -- sensitive customer notes, personal context