Agent Spend Precheck Gate

Scenario

Use as a deterministic policy check immediately before a paid model call or tool operation in a buyer-owned runner.

Input and output

Input has remaining budget, proposed spend, per-call ceiling, and a reservation id. Output is APPROVED or BLOCKED, with remaining-after-reservation and explicit reasons.

Execution and recovery

Run the README command. Missing values or negative amounts return a visible error. A BLOCKED decision must be handled by the buyer runner; this offline script cannot stop an external process.

Privacy boundary and acceptance

Use numeric amounts and synthetic ids only. The sample shows an approval and a blocked case across repeated runs. It does not access billing, rotate credentials, call a provider, or promise hard enforcement without integration.