Writing a rule of your own

Lesson 6 found that the built-in argument checks only know a handful of tool names. Lesson 7 found the second path. This lesson uses that path to write the rule the built-ins do not have: no refund over a hundred pounds.

A custom rule is a small object with a function inside it. The function gets the request and returns True to allow.

from agent_control_plane.agent_kernel import PolicyRule, ActionType

def small_refunds_only(request):
    return request.parameters.get("amount", 0) <= 100

rule = PolicyRule(rule_id="refund-cap", name="small_refunds_only",
                  description="Refunds over 100 need a human",
                  action_types=[ActionType.API_CALL], validator=small_refunds_only)
print(rule.name, rule.priority)

action_types is what decides whether the rule is consulted at all. A rule listing API_CALL never sees a FILE_READ request, so a rule that seems not to run is usually a rule watching the wrong action type.

Registering it

from agent_control_plane import PolicyEngine
policy = PolicyEngine()
policy.add_custom_rule(rule)

for amount in [40, 4000]:
    print(amount, policy.validate_request(make_request(amount)))

The reason names the rule, which is what makes a refusal readable months later when there are thirty rules and somebody asks why a refund did not go through.

The mistake this lesson exists to prevent

Register the same rule, then ask the gate from lesson 4 instead.

from agent_control_plane import AgentKernel
policy.add_constraint("support", ["issue_refund"])
kernel = AgentKernel(policy_engine=policy)

print(kernel.intercept_tool_execution("support", "issue_refund", {"amount": 4000}))

A four thousand pound refund, allowed, by an engine holding a rule that forbids exactly that. The rule is real and registered; it simply is not on this path.

Rules are sorted by priority when they are added, highest first, and the first one that returns False ends the check.