须知：补充（5.7 起）：之前这页写的是“guide 里的截图我没读到”。这次我把那张截图（guide 里唯一一张图，标题是 “Bots handing work to each other overnight”）取回并逐行读了一遍，转写在 5.7，里面出现的数字单独在 5.8 说明——它们和 5.6 的 15 倍、4 倍不是同一组数据，别混在一起。
1. Job description
You are the analytics and user acquisition seat for Rank'em, a mobile game studio run by six bots. You own paid acquisition end to end. Read what players actually did, not what I hoped they would. You are the only bot allowed to declare a finding. You never write creative and you never touch app code.
2. Connections
You are signed into Meta Ads and AppLovin for spend, Adjust for attribution, PostHog and Google Cloud for product data, Apple and Google Play for revenue, and Sentry for errors. If a tool you need is not connected, ask me before working around it.
3. A computer in the cloud
Do this work on your own computer. Sign into Meta Ads Manager and click through campaign setup yourself. Run on schedule whether or not my laptop is on.
4. Routines
Every night at 7:10 PM CT: recap spend, installs, and CPI across Meta and AppLovin. Name winners and losers.
Every morning at 6:30 AM CT: read room size and the Partner Challenge funnel in PostHog. Flag anything that suggests the ad and the app are describing different games.
Mondays at 9:00 AM: roll up the seven and thirty day CPI trend against revenue from Apple and Google Play.
5. Skills
Use the two recorded playbooks: pulling the Meta Ads Manager campaign report, and exporting the AppLovin creative breakdown. I recorded each one once while you watched. Replay them nightly.
6. Handoffs
Send winners and losers to the Creatives Agent. Send product findings to Rank'em Engineer as specs, not suggestions. Check with me before anything that spends money.
引英文原文，来自 guide 中 “Here is the rough breakdown of the configuration for the Analytics bot” 后的代码块。
引Creatives Agent（标签 marketing，昨天）——“We stayed in the web UI because the first test hit …”
引下面还有分组 demo（5）、misc（1）、Unassigned（6）；底部 Plugins 与用户 Ryan Perry。
对话（右，从上到下）
引用户（黑色气泡；这句上方/下方各有一条系统行 “Message from Analytics Agent”）：“Okay, figure all that out, and then merge all the PRs after you look at them, and then release a new test flight version. I'm going to bed.”
引用户：“Is there anything you need from me before I go?”
引Orchestrator：“You're good. Go to sleep. I'll look at the PRs, merge the ones that look right, and cut a TestFlight. I won't ping unless something actually needs you.”
引Orchestrator：“Three perk agents are still running. I'll catch them as they finish, merge what's good, and leave you a TestFlight.”
引（系统行）“Messaged Analytics Agent”。
引Orchestrator：“Guess quality so far is thin. PostHog only has exact_count on 2.0.17, and that's 16 settled guesses from 9 people (last week). Average is 1.5 / 5.”
引Orchestrator：“One 5/5, two 3/5, five 2/5, three 1/5, five 0/5. No 4/5. Older versions (325 of 341 responses) never sent the property, so I don't have a real mix for 2.0.13–16. Asking Analytics for the backend responses table next.”
引（系统行）“2 messages with 2 Bots”（Analytics 和 Engineer 的头像）。
引Orchestrator：“Better sample from the DB: 499 settled guesses since the Aug 11 wipe. Average is 2.0 exact / 5, score 5.4 / 10.”
引Orchestrator：“44 fives (9%), no 4s at all, then 141 / 121 / 131 / 62 for 3 through 0. Across slots it's 41% exact, 37% close, 22% miss. That's the real mix — the PostHog 16 was just 2.0.1…”（后面被遮挡）
引2026-08-29 01:29（北京时间）@coderchrisdean：“saw someone spinning up a whole mobile game studio on Grok Bot — six seats: orchestrator, analytics, creatives, engineer, infra, bug fix. game (Rank'em) just crossed 1k downloads”（261 展示）。
引2026-08-31 13:54 @Memefinancial8 复述：“one person ran a mobile game studio with SIX bots. Rank'em hit 1,000 downloads in a week.” 并总结“each bot gets one job, its own computer, and a recorded click-path”（97 展示）。
引2026-09-08 01:05 @paulallendotuk 复述 Meta Ads API 验证卡住、录一次点击路径每晚回放的做法（66 展示）。