The exact CLAUDE.md structure, morning email schemas, and dispatch prompts I use every day. Copy it. Strip the parts you do not need. Ship.
One operating company at the top. Four brands underneath. Six agents that act across brands (one per function, not one per business). One file makes the whole thing legible to Claude Code.
# CLAUDE.md You are the operating partner. CEO is [name]. Your job is to make the portfolio visible, move safe internal work without asking, and surface only decisions that need the CEO. ## Portfolio | Company | What it is | Stage | |---|---|---| | Brand A | [one line] | revenue / pre-launch / paused | | Brand B | [one line] | ... | ## Decision Rights - GREEN execute (research, drafts, internal docs, approved publishing) - YELLOW prepare + ask (outreach sends, CRM writes, public announcements) - RED never without explicit per-action approval (spend, contracts, irreversible writes, anything with legal / financial / reputational blast radius) ## Reporting Cadence - Daily 07:00 — state-of-portfolio email (one line per brand) - Mon 07:30 — weekly brief (revenue + pipeline + 3 priorities per brand) - On-fire — immediate alert if anything would change CEO's day ## Bootstrap Principle Manual first. Automate the second time, only if it recurs and saves time, reduces errors, or improves consistency. ## Memory Root: MEMORY.md (auto-loaded). Tag entries by brand: [A] [B] [C] [D].
03-projects/ ├── brand-a/ │ ├── CLAUDE.md ← brand-specific voice, ICP, offer, pricing │ ├── 01-context/ ← business background, voice transcript │ ├── 02-offers/ ← what we sell + at what price │ ├── 03-icp/ ← who we sell to + how we recognise them │ ├── 04-outreach/ ← templates, sequences, log │ ├── 05-content/ ← published, drafts, calendar │ ├── 06-pipeline/ ← active prospects, deals, status │ ├── 07-assets/ ← landing pages, PDFs, images │ └── 08-sops/ ← standard operating procedures ├── brand-b/ │ └── (same structure) ...
Three emails. They land in your inbox before 07:00. By the time you sit down with coffee you have everything you need to run the day in 25 minutes.
Subject: [Daily] State of portfolio — {{date}}
State of each brand (one line each):
- Brand A 🟢 [headline]
- Brand B 🟡 [headline]
- Brand C 🟢 [headline]
- Brand D 🔴 [headline + reason]
Today's three highest-leverage moves:
1. [task] — [brand] — [why it matters today]
2. ...
3. ...
One decision needed (max one per day):
- [decision] — [recommended answer + reasoning]
Cron + automation status:
- All green / [n] failed (see below)
Subject: [Intel] What's hot — {{date}}
Top story across niches:
[1 sentence — the meta-story that ties multiple niches together]
[Niche 1] — 5 items
1. [headline] — [source] — why it matters: [...] — angle: [...]
2. ...
[Niche 2] — 5 items
...
Recommended lead-magnet angle for today:
- [asset] — anchored on [receipt] — link to [page]
What's NOT hot:
- [item] — saturated / too broad / no edge
Subject: [Triage] Overnight inbox — {{n}} replies
By company, by priority:
HOT (act today):
- [name] — replied positive — recommended response: [draft]
WARM (act this week):
- [name] — replied neutral — recommended response: [draft]
NURTURE:
- [name] — soft signal — add to follow-up sequence [N]
DEAD:
- [name] — auto-archive
Numbers: {{open}} open / {{replied}} replied / {{bounced}} bounced
Read 03-projects/[brand]/03-icp/ and 04-outreach/dm-template-v1.md. For prospect [@handle], find their last 3 X posts via scraper (verbatim, never paraphrase). Confirm current role, current company size, current public stance. Draft a 117-word DM that: - Quotes their actual line (not a summary) - Mirrors the operator situation in their own words - Offers [specific value, ≤ 1 link] - Has no self-announcing prefix - Ends with a low-friction CTA Show me the DM. Do not send. I approve before fire.
Read MEMORY.md, 01-context/voice-rules.md, today's daily-intel email.
Draft 5 X posts for [brand]:
1 lead-magnet anchor (specific number + pain + free thing + visual + link)
1 thread topic with 7-step hook (pattern interrupt, authority, mirror,
reveal, gap, promise, transition)
3 reactive takes mapped to today's discourse (quote / reply / standalone)
Voice rules: no em-dashes. No self-announcing prefixes ("Operator
question:", "For operators:"). First-person. Specific numbers.
2-3% swearing rate. Never expose live client work.
Output: 5 drafts, ranked by which I should ship at 09:00 today.
Run Exa + Firecrawl across these 4 niches: - [Niche 1 — anchor accounts + queries] - [Niche 2 — anchor accounts + queries] - [Niche 3 — anchor accounts + queries] - [Niche 4 — anchor accounts + queries] Window: last 48 hours, 7 days for context. For each niche: - Top 5 items - Source URL (must be real, dated within window) - Why it matters in 1 sentence - Recommended content angle (post / thread / lead-magnet) Synthesise via Anthropic API. Email digest by 06:30 to [editor email]. Hard 250-word ceiling per niche. Be honest if a niche is thin.
The 6 agents move work without you because GREEN / YELLOW / RED tells them when they can ship and when they have to wait. Most operators I see fail with AI agents fail because they have not written this matrix down. The agent then asks permission for everything, which is worse than no agent at all.
Write the matrix once. Reference it from every agent's CLAUDE.md. The agents will then surface decisions only when they actually need a call. Everything else moves while you sleep.
Daily takes on Claude Code, AI agents, and operator workflows. Wednesday deep dive. No hype. No "10 ways to use AI." Production-grade only.
No spam. Unsubscribe anytime.
THE CLAUDE OPERATOR · 4-BUSINESS STACK · CLAUDEOPERATOR.CO/SYSTEM
Built by an operator who runs his own businesses on Claude. Not affiliated with Anthropic.