# respect-platform-tos

**Rule:** The swarm never asks any agent to do anything that violates a third-party platform's terms of service. Every platform we touch has terms; we read them and stay inside them.

**Specifically banned (these are TOS violations on at least one major platform):**

- **Fake or paid reviews.** Never write a positive review on behalf of an operator. Never offer compensation, gift cards, discounts, or anything else of value in exchange for a review. Many platforms ban it (Yelp ban list, FTC enforcement actions, Google policy).
- **Review-gating / cherry-picking.** Never use a flow that asks "are you happy?" and only routes the happy ones to public review platforms. Google explicitly prohibits this since 2018. Yelp has prohibited it since forever. We do not build a feature that does this.
- **Bulk auto-responses.** Never post the same response to multiple reviews without per-review variation. Yelp flags this aggressively.
- **Scraping.** Never scrape a platform we lack an API for. If a platform we want is API-less, we either build a manual-upload workflow or skip it. We do not scrape.
- **Account sharing.** Never have one platform account act on behalf of multiple businesses unless the platform's API explicitly supports that (e.g. Google's location group manager). One business = one account.
- **Engagement bait.** No "tag a friend to win" on FB/IG. No "share to enter." These violate Meta's promotional-content rules.
- **Auto-DM / auto-follow.** No automated direct messages, no automated follow-back patterns, no automated DM from social-poster.
- **Compensation in a public response.** Never offer a refund / freebie / coupon in a public response on Google/Yelp/Tripadvisor. The platforms reward private resolution; offering compensation publicly invites brigading. (The brand-style can override in private channels; not in public.)

**Per-platform notes the runtime watches:**

- **Google:** responses are public and tied to the location's GBP. No requesting deletion of negative reviews in the response. Google's policy explicitly cautions against defensive language.
- **Yelp:** does not allow operators to solicit reviews. Strict on bulk and identical responses.
- **Tripadvisor:** allows responses but has its own length and tone guidelines; the response-drafter respects those.
- **Facebook:** posts and responses are subject to Meta's content rules; the social-poster's prompt enforces the relevant subset.

**Enforcement:**
- The runtime maintains a per-platform "what is allowed" matrix and checks every outbound action against it before submitting. If a platform updates its terms, that matrix updates and the runtime re-checks the existing queue.
- A hook breach attempt is logged AND surfaces in the swarm-status page so we can audit prompts that drifted.