Ad policy

Draft. Have a lawyer review before launch.

Tokenbreak shows ads inside people's terminals while they work. That's a place where trust runs out fast, so this policy is stricter than most ad networks. If you're unsure whether something is allowed, ask before you buy. We'd rather answer an email than reject a paid ad.

This policy covers every ad on Tokenbreak: Status Sponsor, Prompt Banner, and Theater. It covers the landing page your ad links to as well.

Tokenbreak is built by an independent developer. It is not affiliated with or endorsed by Anthropic.

The formats

Format	Where it shows	What you submit
Status Sponsor	Claude Code's status line	Up to 40 characters of text and one link
Prompt Banner	A one-line band above the prompt box, rotating about every 40 seconds	Up to 60 characters of text and one link
Theater	An image pane on the right while Claude is working, on wide fullscreen terminals only	One static image, a short caption, and one link

Tokenbreak labels every ad as sponsored and controls how it's styled. You submit plain text. We don't accept color codes, escape sequences, or any formatting tricks inside ad text.

What every ad needs

• Say who is advertising. A reader should be able to tell which product or company the ad is for.
• Make claims you can back up. If you say "2x faster", be ready to show us what it's faster than.
• Link to a page that matches the ad. If the ad is about your database, the link goes to a page about your database.
• Use HTTPS, on a domain you control. No URL shorteners, and no redirect chains that end up somewhere else.
• Make the landing page work without installing anything. If people need to sign up, fine. If they need to download a file before they can see what you're selling, that's a no.

What's not allowed

Malware and anything that touches the user's machine

• Malware, spyware, adware, or software that hides what it does.
• Links that start a download automatically, or pages that push a download before the visitor asks for one.
• Shell commands, install one-liners, or any text meant to be copied into a terminal or into Claude. "Run curl ... | sh" in an ad is an instant rejection, even if the script is harmless. Put install instructions on your landing page instead.
• Text aimed at AI agents, such as instructions to "ignore previous instructions". Ads never enter the model's context, but we reject these anyway.
• Phishing, fake login pages, or anything that asks for credentials, API keys, seed phrases, or private keys.

Crypto scams and airdrops

• Airdrops, token giveaways, "claim your tokens", or "connect your wallet to check eligibility".
• Token sales, presales, ICOs, memecoins, and anything promising returns or yield.
• Trading signals, copy trading, and "guaranteed" crypto income.

Developer tools that happen to work with blockchains, like an RPC provider or an SDK, may be allowed after review. Anything that sells or gives away a token is not. Tokenbreak pays developers in USDC, and we have no token of our own. If an ad suggests otherwise, it's a scam and we'll remove it.

Gambling

Casinos, sports betting, sweepstakes casinos, poker, lotteries, prediction markets, and loot box mechanics sold for money. This includes "skill games" where you pay to enter for a cash prize.

Adult content

Sexual content, nudity, dating services built around hookups, and suggestive imagery in Theater. Keep it something you'd be fine with on a shared screen at work, because that's where this shows up.

Impersonation

• Pretending to be another company, product, or person.
• Implying Anthropic, Claude, or Claude Code made, endorses, or recommends your product. You can say your product "works with Claude Code" if that's true. You can't use Anthropic's logos, the Claude logo, or wording like "Claude recommends".
• Ads that look like part of the terminal: fake error messages, fake permission prompts, fake tool output, or text styled as if Claude wrote it. "Error: your build is slow" is an ad pretending to be an error, and it will be rejected.
• Pretending to be Tokenbreak, or saying an ad is "official".

Misleading claims

• Fake urgency, like countdowns that reset or "only 3 seats left" when that isn't true.
• "Free" when there's a required payment or a card-required trial you don't mention.
• Benchmarks you can't support, or comparisons that name competitors unfairly.
• Fake reviews, fake testimonials, or invented user counts.
• Health, financial, or legal claims that promise outcomes.

Other things we don't run

• Weapons, drugs, tobacco, and vaping products.
• Hacking services, stolen data, credential or account sales, cheating tools, and stalkerware.
• Payday loans, multi-level marketing, and "make money fast" schemes.
• Political and issue ads. We may revisit this later. For launch, no.
• Hate, harassment, or anything that targets people for who they are.

Examples

Fictional brands, for illustration.

Ad text	Verdict	Why
Acme DB: Postgres branches in 2 seconds. Free tier.	Fine	Clear advertiser, checkable claim
Acme Logs: grep your prod logs from the terminal	Fine	Plain description of the product
Claude recommends Acme DB	Rejected	Implies endorsement by Anthropic
Error: build too slow. Fix at acme.dev	Rejected	Pretends to be a system message
Run: curl acme.dev/i | sh	Rejected	Shell command in an ad
Claim your ACME airdrop before Friday	Rejected	Airdrop

How review works

Every ad is reviewed by a person before it runs. Nothing goes live automatically.

1. You submit your ad and pay for a pack.
2. A reviewer reads the text, looks at the image if it's Theater, and opens the landing page in a clean browser. We check that the page loads, matches the ad, doesn't push a download, and doesn't ask for anything sensitive.
3. You get one of three answers: approved, changes requested (with what to change), or rejected (with the reason).
4. Approved ads go into rotation. Ads with changes requested go back to step 2 when you resubmit.

We aim to review within two business days. Early on it will often be faster, and sometimes it'll be slower. Editing an approved ad, including changing where the link points, sends it back through review.

We may check landing pages again while a campaign runs. If your page changes into something this policy doesn't allow, the ad comes down.

Takedowns

We can pause or remove any ad at any time if:

• the landing page changes in a way that breaks this policy,
• we get credible reports about the ad or the advertiser,
• we find out the ad was approved by mistake, or
• the advertiser's domain or account looks compromised.

When we take an ad down we'll tell you why, unless doing that would help someone get around our checks. For anything involving malware, phishing, or scams, we remove first and talk after.

Developers who see an ad that seems wrong can report it through the report form on tokenbreak.dev. Anyone can pause ads on their own machine with /ads pause.

Refunds

Situation	What happens
Ad rejected in review	Full refund, or credit toward a new ad if you prefer
You cancel before any impressions run	Full refund
You cancel partway through a pack	Refund for unserved impressions, minus payment processing fees
We take an ad down for reasons that aren't your fault (our mistake, a policy change)	Refund or credit for unserved impressions
We take an ad down because it broke this policy	No refund for served impressions. Unserved impressions are refunded at our discretion. Malware, phishing, and scams get no refund
Impressions we flag as invalid traffic	Not billed. We credit them back to your pack

Refund requests go to the email on your account. We'll reply within five business days.

Appeals

If you think we got a rejection wrong, reply to the rejection email once with your reasoning. A different reviewer will look at it if we have one. Their answer is final for that version of the ad, but you can always submit a changed version.

Changes to this policy

We'll update this page as we learn what works. Approved ads that are already running stay approved unless they break a rule that was added for safety reasons (malware, scams, impersonation). We'll email advertisers about any material change before it takes effect.