🔥 Uprise Teardown Machine (LIVE 2026-07-01)

The machine that makes one personalized pitch cost Rio ~60 seconds. Each prospect gets a personal Warm-Rise-branded teardown page at their own URL. Rio DMs the link. Nobody else DMs creators a custom page about their own business. This is the manual version of the portal's future IG-analysis agent (proof-first, per uprise-what-it-is: run the motion by hand, automate after it makes money).

Live infrastructure

• Project: ~/uprise-teardowns/ (git repo, local). Cloudflare Pages project uprise-teardowns.
• Live demo (Rio's self-teardown, REAL agent data): https://uprise-teardowns.pages.dev/riocostaaa/
• How it works: one JSON per prospect in data/<handle>.json → node build.js → wrangler pages deploy public --project-name=uprise-teardowns → page at /<handle>/.
• Pages are noindex (1:1 private feel, no public shaming). Root URL redirects to uprisegrowth.io. CTA = Calendly growth call + roadmap tool fallback.
• ⏳ Optional (Rio, 2 min): add CNAME t → uprise-teardowns.pages.dev in Cloudflare DNS dashboard (wrangler token can't touch DNS) → then links become t.uprisegrowth.io/<handle>. Claude attaches the custom domain to the Pages project via API after (same move as roadmap subdomain).

The daily rep (division of labor)

1. Rio (10-15 min): pull 3-10 creator handles from IG. Target per uprise-creator-outreach: 10k-150k followers, his worlds (lifting/calisthenics/combat/hybrid/AI-tech), already monetizing, leaky link-in-bio. NO mom creators.
2. Claude: scrape their link-in-bio/site (Firecrawl), find REAL leaks only (never fabricate observations), write data/<handle>.json, build + deploy, hand back links + a personalized DM per prospect.
3. Rio (60 sec each): review, fire the DM.

The DM that carries the link (Rio's voice, warm-prospect register)

hey [name], went through your page for real, your [niche] content is legit. but i tapped your link in bio and watched the money leak. so i made you something, not a pitch deck, an actual breakdown of your page: [link] took me a while so at least steal the fixes lol. if you want them built and run for you, that's what i do.

The email send rail (LIVE 2026-07-02 — the "repeatable outbound automation agent")

The business-barrel version of the daily rep, automated end-to-end except Rio's approval tap. Terminal-run, local Mac (NOT the old Hermes VPS — SSH to 159.223.185.67 is firewalled out as of 7/2; fix via DigitalOcean console someday or let it die).

• Send rail: ~/uprise-portal/agent/outbound_send.py. Reads agent/outbound/queue/*.json, sends ONLY items with "approved": true as [email protected] via Gmail SMTP. Hard-coded guards: weekdays 9-5 ET only · 10/week cap · never emails the same address twice (checked vs outbound/sent/) · refuses any body containing a redirect-wrapped link (google.com/url etc. — the 7/2 "looks like a scam" incident, links must be RAW urls).
• Schedule: launchd io.uprise.outbound-send fires 9:30 / 12:30 / 15:30 local daily (extra fires are no-ops). Logs: agent/logs/outbound.log.
• Approval loop: unapproved queue items trigger a digest email to Rio each run. Approve = tell a Claude session "approve the outbound queue" (or flip "approved": true in the JSON).
• Compose step (Claude, per batch): research prospect → teardown page (~/uprise-teardowns) → write queue JSON (to, subject, body plaintext with raw links, approved: false, prospect, page, created, optional not_before: YYYY-MM-DD). Copy rules: no em dashes, real observations only, Rio's voice. Spacing rule: max ~3-4 cold sends per day from uprisegrowth.io — when Gmail-scheduled sends already fill a day, stamp the rail batch not_before tomorrow.
• ⛔ NEVER compose outreach via the claude.ai Gmail connector. Confirmed 7/2: create_draft rewrites every URL in the stored body into a google.com/url?q=... redirect wrapper (looks like phishing), even when given clean raw links. Hand-composed Gmail UI emails are unaffected. Connector = read/search/labels only; sending = this rail.
• Research reach: agent-reach CLI installed 7/2 (free, local) — web/YouTube/search/RSS live now; Instagram + Reddit + LinkedIn read through Rio's own Chrome once he installs the OpenCLI extension (one click: https://chromewebstore.google.com/detail/opencli/ildkmabpimmkaediidaifkhjpohdnifk, then opencli doctor); Twitter needs a Cookie-Editor export.
• Needs once (Rio): Gmail app password → security add-generic-password -a [email protected] -s uprise-gmail-smtp -w APP_PASSWORD -U (create at myaccount.google.com/apppasswords). Until then the rail can't send and logs FATAL.
• Dedupe hygiene: anything sent MANUALLY from Gmail (drafts, scheduled sends) must be back-logged as a JSON in agent/outbound/sent/ (to + sent_at) or the rail won't know the address is burned. As of 7/2 pending back-log: tease (1:06am), Viking/Elevate/OldVines (scheduled 9-10am), Flutter/Cooper's/Vivia (drafts) — log them once they actually send.

Field learnings

• 2026-07-01 (Omar, first live DM): bigger creators run DM qualification bots (ManyChat-style) — the outreach DM trips the lead funnel and a bot auto-replies (e.g. "lose fat, build muscle or both?"). Play: do NOT answer the qualifier (tags you as a lead, buries you in their sales flow). Reply with something the bot can't keyword-match + restate you built them something + the link → unmatched replies land in the human-review inbox. Bonus: the bot itself becomes pitch material ("your bot is the best part of your funnel, imagine it with a real engine behind it").

Rules (non-negotiable)

• The "habit done better" frame on EVERY page headline + DM open (locked 7/3, Sebastian Estrada learning in 06-learnings): never pitch the category ("you need a growth system" = convincing). Always anchor to what they ALREADY do and upgrade it ("you already post daily and sell coaching — here's where it leaks"). Convergent with the 7/2 VoC finding (11-audience): DTC founders are openly hostile to "post more content" advice — meet the existing behavior, name the leak.
• Every observation on a page must be REAL and verifiable from their public profile (06-learnings no-fabrication rule). Compliments too.
• No em dashes anywhere on the pages or DMs (client-facing).
• Teardown pages can double as reel material (screen-record the scroll + talk over it = Pillar 2 "The Leak" content).
• Pricing stays off the page. Calendly call reveals tiers per uprise-offer-tiers.

Why this is THE $10k MRR machine

$10k MRR ≈ 4 × $2.5k Foundation retainers (or ~7 × $1.5k). Channel benchmark: the friend's $12k MRR came from DM-to-decision-maker volume. This machine makes Rio's version of that DM nearly effortless AND 10x more impressive than text. 5 teardowns/day × 5 days = 25 personalized pitches/week to real buyers. Reply rate on "I built you a page" >> reply rate on plain text.

Outreach QC gate (added 7/3 from the course lead-finder repo — the one real gem)

Every DM / rail email passes this before send:

1. Max 4 sentences.
2. Exactly ONE specific observed detail about THEM (their number, their page, their leak — proves it's not a blast).
3. Zero exclamation marks.
4. Close = low-pressure question, never a demand ("worth a look?" not "book a call now").
5. No em dashes (standing rule). Pairs with the Sebastian rule: the observed detail IS their existing habit, the offer is it done better.