<aside>
BTW, we've been building SalesRobot since 2020. Bootstrapped.
We’re at 5,500 users across 45 countries, but here’s some actual crazy shit: → $1.4M ARR → 5 million organic LinkedIn impressions in 2025 alone.
Give SalesRobot your ICP, connect your LinkedIn account, and it handles the rest:
🎞️ AI video clone - your face, auto-personalized at scale
🎙️ AI voice clone - your voice, personalized per prospect
📅 AI appointment setter - handles replies, kills objections, books calls while you sleep.
Check it out if you’re intrigued, we’ve got a 14-day free trial (you should definitely give it a shot)
Book a demo today
White-label version for agencies (your clients would never know it’s us)
</aside>
a loop is one prompt, saved to a file, run on a schedule. inside that prompt, Claude Code spawns sub-agents that each take one slice of the work and report back. the merge step turns their separate findings into one output file.
the reason it's a loop and not just an agent: it runs again next week against new inputs, and each pass builds on the last one's output. loop 1 feeds loop 4. loop 4 feeds loop 5. loop 5 feeds loops 6 and 7. loops 9 and 10 run against whatever you published.
if a loop has no natural way to split the work across sub-agents, it isn't a loop, it's a script. every loop in here has its split axis named at the top so you can see the logic instead of taking my word for it.
| # | loop | splits by | runs |
|---|---|---|---|
| 1 | pain harvest | subreddit | weekly |
| 2 | angle gap | competitor creator | weekly |
| 3 | profile mirror | analysis dimension | monthly |
| 4 | hook bank | hook pattern | weekly |
| 5 | draft | post in queue | weekly |
| 6 | platform split | draft | weekly |
| 7 | visual | post needing an asset | weekly |
| 8 | content decay | old post | monthly |
| 9 | engagement to pipeline | published post | daily |
| 10 | reply handling | open conversation | daily |
loops 1 to 8 are a content system. loops 9 and 10 are the reason it pays for itself.
run them in this order the first time. after that the schedules take over.
┌──────────────┐ ┌──────────────┐ ┌──────────────┐
│ 1 pain │ │ 2 angle gap │ │ 3 profile │
│ harvest │ │ │ │ mirror │
└──────┬───────┘ └──────┬───────┘ └──────┬───────┘
│ │ │
└──────────────────┼──────────────────┘
▼
┌──────────────┐
│ 4 hook bank │
└──────┬───────┘
▼
┌──────────────┐
│ 5 draft │
└──────┬───────┘
┌───────────┴───────────┐
▼ ▼
┌──────────────┐ ┌──────────────┐
│ 6 platform │ │ 7 visual │
│ split │ │ │
└──────┬───────┘ └──────┬───────┘
└───────────┬───────────┘
▼
( you publish )
│
▼
┌──────────────┐
│ 9 engagement │ both call
│ to pipeline│ ──▶ /salesrobot-mcp
└──────┬───────┘ first
▼
┌──────────────┐
│ 10 reply │
│ handling │
└──────┬───────┘
▼
( booked call )
loop 8 (content decay) runs sideways on your archive, monthly
required for loops 1 to 8
→ Claude Code installed (npm install -g @anthropic-ai/claude-code)
→ python 3.9 or newer, already on most machines
→ 20 minutes to fill in three config files