How playbooks work
Rules every warmup playbook shares - prerequisites, the approval gate, pushback handling and the session report - and how an agent runs one.
A playbook is a multi-day routine an agent runs on the iPhone, one session a day. The three here are warmups: a new or dormant brand account is used the way its owner would use it (profile, browsing, small genuine engagement) before it posts at a steady pace. Each playbook holds the platform-specific parts; this page holds the rules they share.
| Playbook | Length | Posts start |
|---|---|---|
| Instagram warmup | 14 days | Day 3 (story), day 6 (feed) |
| TikTok warmup | 14 days | Day 7, once For You is on-niche |
| Reddit warmup | 21 days | Day 15 |
Playbooks are for brands and agencies operating their own accounts: one account per iPhone, within each platform's terms, with a person approving everything public.
Before any playbook
Done when every item holds:
- The iPhone is set up, and the chat has an AI account (or an agent is connected over MCP).
- The platform's app is installed, and the account owner has signed in on this iPhone. The agent works inside a signed-in app and never handles passwords.
- This is the only account for that platform on the iPhone.
- The niche is written down: 3 to 5 topics, a few peer accounts, the search terms or hashtags the audience uses.
Running a session
From Zeptap's chat: fill in the playbook's prompt for today (the stage table gives the limits) and send it. Start a new chat each day.
From a connected agent (Claude Code, Codex, Cursor): give it the playbook's markdown URL and the day, for example Run day 4 of https://zeptap.com/docs/playbooks/tiktok-warmup.md for @brand. Niche: …. The agent follows the steps below.
Read today's stage
Find the row for today's day in the playbook's stage table. Done when you know the limit for each action and whether today includes a post.
Run the session
Follow the playbook's daily prompt with those limits, under the approval gate. Done when every numbered step of the prompt has run, or a pushback signal has stopped the session.
Report
End on the Home Screen and send the session report. Done when the report is sent.
The approval gate
The agent drafts; the owner approves. Every public or outward action goes through the gate: a follow, a comment, a reply, a post, a story. The agent shows each draft (the post it answers, the text, the account to follow) and acts on the ones approved, exactly as written.
Likes, upvotes, watching and browsing are the only actions that skip the gate, and only up to the day's limit and only on on-niche content.
Outside the playbook's allowed actions (DMs, unfollows, downvotes, edits to settings or the account), the agent asks first.
Pushback
Pushback is any sign the platform wants the account to slow down: a rate-limit message ("Try again later", "You're doing that too much"), a restriction notice, a CAPTCHA, a verification screen, a removal notice. On pushback:
- Stop the session immediately and leave the screen as it is.
- Report what appeared, word for word.
- The owner completes any verification themselves, on the phone or in Zeptap's mirror.
- The next session runs one stage lower in the table.
Each playbook lists its platform's usual pushback signals.
The session report
Day <D> of <playbook>
Watched/browsed: <count and where>
Likes/upvotes: <count>
Follows: <handles>
Comments/replies posted: <count, with links or post titles>
Posted: <what, or none>
Niche fit: <how much of the feed was on-niche>
Pushback: <none, or the exact message>
Next session: <stage, and anything the owner should do first>Text the agent types
Zeptap types US-keyboard ASCII: bios, captions and comments are written with plain letters, digits and punctuation. Put emoji and accented text into the platform's own editor yourself, or in the media before uploading. See Driving the iPhone.