What it is
TikTok Army is a system that does the day-to-day work of running a brand on TikTok. You write a short brief — one TikTok handle, what you want done, optional notes — and the system breaks the job into named steps, each handled by a specialist agent. Some steps are read-only (look at an account, score it, summarize). Some produce drafts (a new listing title, a DM, an ad campaign plan). Anything that would actually post, send, or spend money pauses for you to approve before it runs.
The 15 agents are listed in the catalog (see article: Reading the Risk Scores for the metrics they emit). Each one has a single, well-defined job — there is no general-purpose "do everything" agent. The orchestrator wires them together into workflows. shadowban_sentinel runs inside Profile Audit, Post-Launch Loop, and TikTok Shop Audit; ldr_compliance runs inside TikTok Shop Audit.
Who it's for
- •Brand owners who already have a TikTok presence and want a faster operating loop than a single human or a generalist VA.
- •Operators running multiple client brands who need consistent, auditable processes across accounts.
- •Anyone who wants the work done but wants to stay the final approver on what gets published, sent, or spent.
If you only need to post one video this week, this is overkill. The system pays off when you're running a campaign, optimizing a Shop, or auditing an account regularly.
The four surfaces
Everything you do happens on one of four surfaces in the dashboard.
1. Brief
The intake form. You pick a TikTok handle, choose whether it's a third-party profile (audit a competitor or inspiration) or your own brand (plan and launch work), pick the desired outcome (Profile Audit / Campaign Launch / Post-Launch Loop), and optionally add notes. That's it. The brief becomes the starting input every downstream agent reads from.
2. Workflow
A workflow is the named recipe that runs against a brief. Four are seeded and ready to use:
- •Profile Audit — read-only scan of any handle. Returns a Markdown report covering account health, shadowban risk, trends, audience segments, and compliance. Nothing gets posted or sent. Safe to run on competitors.
- •Campaign Launch — the full plan-and-launch loop for your own brand. Generates listing rewrites, content briefs, ad campaign proposals, and creator outreach DMs. Has four human approval gates (one before each outbound action).
- •Post-Launch Loop — runs on a schedule after a launch. Scores recent posts, triages new comments, rechecks account health and shadowban risk, surfaces next-cycle recommendations.
- •TikTok Shop Audit — operational + content audit for a Shop account. Watches Late Dispatch Rate (TikTok's 10% enforcement line), flags at-risk orders before SLA breach, rechecks account health and shadowban risk, audits captions for FTC and music-licensing issues. Read-only.
When you start a workflow run, the dashboard shows a live DAG (a graph of the steps) that updates in real time as each step runs.
3. Catalog
The list of every agent in the fleet, with what it does, what it needs, what it produces, and where humans show up. This is the reference page. If you're ever wondering "what does the Trend Watcher actually do?" — go here.
4. Approvals
The Approvals queue is where you review anything an agent has drafted that needs your sign-off before it hits the world. Listing rewrites, video captions, ad campaigns with budgets, DMs to creators. Each card shows the proposed output, the reasoning that produced it, and Approve / Reject buttons. You can also reject with feedback, which is recorded for the agent's next attempt.
How to think about it
Three mental models that help.
Agents are specialists, not employees. Each one does one thing very well — Compliance reads captions for FTC issues; Audience Mapper clusters viewers into named segments; Trend Watcher decides what's rising. They don't have memory between runs unless the workflow threads it through. They don't know what the others are doing. They produce structured output that the next step (or the final synthesis) reads.
Workflows are recipes, not routines. A workflow tells the system "for this outcome, run these agents in this order, and pause for me at these points." You can edit the seeded ones or compose your own (a builder feature). The orchestrator handles dependencies and parallelism — if two agents don't depend on each other, they run at the same time.
Approvals are the safety rail, not the bottleneck. The system is designed so the humans-in-the-loop only show up where it actually matters: anything that costs money, anything that goes out under your brand's name, anything that touches a creator. Inside-the-house steps (audit, score, classify, draft) run without pausing.
What's next
- •First time using it? Run a Profile Audit against your own handle. It's read-only and costs cents in mock mode (zero with real Claude on a small account). See How Profile Audit Works.
- •Ready to launch? Read How to Launch a Campaign to understand the four approval gates before kicking it off.
- •Worried about money? Cost and Budget explains how Claude charges roll up and where you'll see them.
- •Worried about compliance? Compliance Warnings covers what gets flagged and what blocks publishing.