Grab the file, then do the five steps below. Steps 1 to 3 are one-time. Bookmark this page - the link in your email comes back here.
Quick check. Sort anything missing now, not halfway through step 2.
Scoping takes 5 minutes. The first brief takes about an hour. Nearly all of that is you deciding things.
Claude Code: unzip it into your skills folder.
unzip campaign-strategist-starter-v1.zip -d ~/.claude/skills/If your browser blocked the .skill download, use the .zip. Same files, different extension.
check my campaignsIt'll notice nothing is configured & ask you 8 questions, one at a time. What you sell, who to, which profiles the invites go from, what you can prove, what the messages must never say.
Question 5 is the one that matters: what can you prove? If you can't name something a stranger could check, say so. The skill builds the first campaign around a question instead. It will not invent a number for you & you shouldn't either.
It writes your answers into config.md & your first client into clients.md. Both are plain text you can open & fix.
create a brief for [client]It picks the angle & the offer with you - 3 options at each step, you pick. Then it builds the Apollo search, writes the connection request & 3 follow-ups (3 options each) & runs QA on every line.
The brief lands in campaigns.md, paused. Nothing goes live until you load the leads & set it live yourself.
Run the search, pull the list, load it into your sending tool. If you have a HeyReach key, the campaign shell is already there with 0 leads waiting.
Then tell the skill:
I loaded 1,000 leads to [campaign]It writes that to the ledger. This is the line that stops it ordering the same list twice.
check campaignsRed, amber, green on every campaign. Red means fix it today.
To have it run without you: set it as a scheduled task in Claude Code, at the time you'd normally open your dashboard.
Once a campaign is live, three commands keep it alive.
"Run pulse" - refreshes the 4 numbers per campaign: queue, sent, connected, replies. With a HeyReach key it reads them. Without one it asks you & marks them "entered by hand".
"Top up the buffers" - checks every client against the floor: 1,000 qualified leads waiting, 300 in the queue. Below it, you get the next search & a 4-line request to hand to whoever builds your lists.
"Run the learning" - once a campaign has 30 accepted connections, it works out what's winning & writes it to the learning log. Every brief after that starts from it.
Never write a number it didn't read. Every figure comes from the API or from you, marked as typed in. There's no third option.
Nothing sends. It writes, plans & prepares. You load the leads & set it live. New campaigns are written paused, on purpose.
That's correct on a fresh install. Say "check my campaigns" & answer the 8 questions.
Fine. "Run pulse" asks you for the 4 numbers instead. Create the campaign shell by hand - 5 minutes.
Also fine. The skill writes the search, you build it in Apollo's UI & read the count off the screen. It won't write a sourcing request until a real count is in the ledger.
Nothing is in campaigns.md yet, or the status isn't active. Create a brief first. The example block in the file is ignored on purpose.
Almost always the client's copywriting rules are blank. Open clients.md & fill them in - the words that must be there, the words that must never be. Then say "new angle for [client]".
It's right. Numbers older than 26 hours are flagged before the dashboard, not after. Run pulse.
It's a proposal. The skill doesn't kill, scale or send anything on its own. You say yes or no.
Grab the file & start at step 1.
Something broken, or something here that didn't match what you saw? Message me on LinkedIn. This is v1 & every report makes it better.