Capability Pack & Lead Finder Pack

The Power-Up Factory standard, and Lead Finder, the first pack built to it.

Spec v1, Draft

Ship conversations, never infrastructure. The customer’s AI builds and operates the infrastructure. The customer clicks the 3-5 things only a human can click. Nobody opens a terminal, ever, the AI has its own.

Capability Pack: the factory standard

A pack is one installable add-on plugin that gives the customer’s AI a new superpower with real-world reach: publish to the web, find leads, sync a CRM, watch a market. Phase 4 of the customer journey, after Foundation, Data, and Magic.

Each pack contains: 1-3 skills (the behavior), Notion structure (lean inline DBs, built by the skill, never hand-made), a provisioning conversation (the concierge setup, built into the skill), a first win (a visible, screenshot-able result inside 15 minutes), and a keep-fresh task (scheduled, so the power compounds without attention).

The 5-stage anatomy

Every pack, same skeleton:

Prereqs

Declare, never assume. External accounts needed, tier (always free by default), cost at scale, what data leaves the customer’s system. Honest costs up front.

Provision

The concierge. The AI walks the customer through the human-only clicks and does everything else itself.

Install

The skill builds the pack’s databases per the Lean DB Doctrine: inline, lean schemas, rich row bodies. Re-run safe.

First win

The acceptance test. The install ends by using the new power on the customer’s real business, producing one artifact they could screenshot and share. “Setup complete” is a failure state, “look what you just did” is the bar.

Keep-fresh

One scheduled task that keeps the power working weekly without attention. List-first, never duplicate, easy to decline.

The Concierge Provisioning Law

What makes “zero prior experience” true, every pack’s setup obeys:

The AI does everything except human-only clicks (account creation, OAuth, payment, pasting a key).
One step at a time, exact words, wait for confirmation, never a numbered wall of steps.
Verify by testing, never by asking. “Did it work?” is banned, “it works, I just checked” is the standard.
Free tier by default. Paid tiers are an upgrade conversation with real numbers, only when usage demands it.
Default to zero experience. One adapt-up question allowed: “Have you used [provider] before?”
Secrets stay out of Notion. API keys go in connector settings or workspace env, never echoed back in chat.
Recover, never restart. A failed step gets diagnosed and fixed in place.
Stall insurance. If the customer goes quiet mid-setup, progress logs so any future session resumes exactly where they stopped.

The recruiting rule

Packs are never browsed, they’re recruited. When the customer asks for an outcome beyond current powers, CLAUDE.md’s Capability Map teaches the AI to name the pack, the cost in minutes and dollars, and offer it: “Find me 50 salon owners in Dallas” → “That’s the Lead Finder power-up: one free account, ~10 minutes, I do nearly all of it. Want it?” No matching pack means it’s roadmap, logged to the Activity Log as product demand.

Packaging + distribution

One .plugin file per pack, versioned independently, distributed from the same hub as core (share.designhacker.com/bbb/), each with its own tile, price, and 90-second demo clip. Core plugin stays lean forever, packs never modify core skills, they only add.

Pack roadmap

Pack	Unlocks	Dependency
Lead Finder (pack #1)	“Fill my pipeline with scored prospects”	Apify, free tier
Web Publishing	“Publish this to the internet”	Cloudflare, free
Client Portals	“Give my client a live portal”	Cloudflare, free
Market Watcher	“Watch my niche and brief me weekly”	Apify + scheduled tasks
Email Engine	“Turn my inbox into to-dos and follow-ups”	Gmail connector

Build order: Lead Finder first, best story, most screenshot-able win.

Lead Finder Pack (pack #1)

Heritage: the Stylist Finder, generalized. The promise: describe your ideal client, your AI finds them, scores them on fit, and hands you a ranked list with the outreach already drafted. You press send.

Trigger phrases: “find me leads”, “fill my pipeline”, “find [type] in [city]”, “who should I be reaching out to”, “run lead finder”.

Prereqs

Apify account, free tier: monthly credits cover roughly 200-400 scraped leads, ~$0.25-$2 per 100 leads at scale, stated in dollars before setup. Reads the Brand Record for the ideal-client rubric, missing means recruit it first. Only search queries leave the system, no customer data. Boundaries: public data only, no scraping behind logins, respects source ToS, outreach is drafted, never sent.

Provision (~6 minutes)

Adapt-up: “Have you used Apify before?” Full concierge path: sign up at apify.com, Settings → Integrations → Personal API tokens, create a token named command-center, paste it in. Stored in connector config, never in Notion, never echoed back. Verified with a 1-result test search before moving on.

Install: the Leads DB

Inline on a new Lead Finder page inside the People zone.

Property	Type	Why
Name	title	Person or business
Type	select	Business / Person
Location	text	City, area
Source	url	Where the AI found them
Contact	text	Email/phone/IG handle
Fit Score	number 0-100	The rubric output, sortable
Heat	select	Hot / Warm / Cool, from score bands
Status	select	New / Drafted / Contacted / Replied / Client / Pass
Campaign	select	Which search produced them
Found	date	When

Every lead’s page body is rich: what the business is, what the AI observed, why the score (each rubric factor with evidence), the suggested angle, the outreach draft. Opening a lead should feel like a researcher handed you a dossier, never a row with a name in it. Converted leads promote into the People master (relate, never duplicate). A small Campaigns log tracks one row per search run: ask, sources, leads found, credits spent, date.

Run flow

Define the hunt

Read the ideal-client profile from the Brand Record, confirm the target in one line.

Pick sources

Google Maps scraper, Instagram profile scraper, web search actor, the AI picks based on the ask.

Scrape + dedupe

Run via Apify, dedupe against existing Leads + People.

Score with the rubric

Fit to ideal-client (0-40), activity signal (0-20), reachability (0-20), quality signal (0-20). Bands: 80+ Hot, 55-79 Warm, under 55 Cool. Weights tunable in plain words via the Brand Library.

Write rich rows

Top 25 per run, full dossier bodies, score evidence included.

Draft outreach

Top 10, in the customer’s voice, angle matched to what the AI observed.

Deliver the hit list

Count, top 10 table, pipeline value framing. Drafts are ready, they press send.

First win + keep-fresh

First win: within 15 minutes of install, 25 scored leads in their niche and 5 outreach drafts in their voice, screenshot-able and immediately actionable. Keep-fresh: one weekly scheduled task, up to 15 new leads scored and top 5 drafted every Monday 8 AM, summarized in the Daily Brief. Declinable in one word.