Concepts

Checks and burn

A check is one prompt evaluated across the supported engines. Reads spend nothing. Actions default to confirm:false and return a quote.

What spends

Only check_prompt (exactly one check) and run_scan (one check per currently tracked prompt) spend.

track_prompt is free now and expensive later

The call itself costs nothing. Its schedule creates committed burn. Default interval is 5 days = +6 checks per month per prompt. Monthly +1 only at interval 30.

Contract

A check is one prompt evaluated across all supported engines. Reads spend no checks.
check_prompt spends exactly one check. run_scan spends one check per currently tracked prompt.
Adding a prompt with track_prompt is free at that moment — its schedule creates committed burn from check_interval_days (default 5 days = +6 / prompt / month; monthly +1 only at interval 30).
For operate/test/setup, enroll, scan, then echo the receipt. Stop before spend only for auto-upgrade, 80% burn, or a hard cap.
There is no publish_fix tool. Scope agents draft fixes; they never publish directly to a website or CMS.
Always pass business_id when the account has more than one business. whoami.weekly_brief is the primary business, not necessarily the one the user is talking about.
A prompt is billed only if at least one engine returns a body. Timeout / 429 / 400 cells are error, never omit.
A billed cell can still be unmeasured (timeout, 429, or ChatGPT gpt-4o fallback). Do not force-retry a down engine.

Billed is not the same as measured

A prompt is billed if at least one engine returns a body. A billed cell can still be unmeasured (timeout, 429, or ChatGPT fallback). Do not force-retry a down engine.

quota_exceeded is terminal

Do not retry and do not split the action into smaller calls. Read whoami and offer to reduce cadence.