Rules

A policy decides. A rule does something about the decision: when something happens, do something. You set rules up in the dashboard, under Rules, in plain words. No code.

A rule is a sentence

When something happens, and a condition holds, in live or test, then run these steps.

• When something is blocked, send a Slack message to #trust-and-safety.
• When something goes to review, email the reviewers.
• When the same user is blocked 3 times in 7 days, pause them.

The dashboard names each rule from its sentence, and keeps the technical details folded away.

Steps

A rule runs its steps in order:

• Slack message: posts to a channel through a Slack incoming webhook URL.
• Call my webhook: a signed POST to your URL, with the same signature and retries as policy webhooks.
• Email: to the addresses you choose.
• Tag the user: labels the person who wrote the prompt.
• Hold for review: puts the decision in the review queue.
• Pause the user: blocks that user's later checks, whatever the policy says. Those decisions carry user_paused: true.

Anything that counts or acts per person needs your id for that person, sent as user on each check (Decisions).

The dry preview

Before you turn a rule on, Test with my last blocked decision runs it against a real decision from your log and shows what each step would have done. It sends nothing: no Slack message, no email, no webhook, no pause.

Rules and policies

A policy says what's allowed, and it's versioned, so every decision traces back to the lines that made it. Rules act once a decision is made. To change what gets blocked, change the policy. To change who hears about it and what happens next, change a rule.