Agent Policy & Guardrail Spec

Write a guardrail policy for this agent: {{agent_description}}. Tools: {{tools_list}}. Data it can access: {{data}}. Users: {{users}}.

For every tool define: risk tier (0–3), allowed arguments and ranges, rate limits per run, which require human approval and from whom, allowed domains/recipients, what is logged, and the exact denial message. Add global rules (budgets, step limits, kill-switch conditions), prompt-injection handling rules, and 12 test cases (allowed, denied, boundary, and attack cases). Output the policy as YAML plus a short rationale for each limit.