- BudgetTracker tracks LLM token cost separately from tool cost - LlmClient auto-records cost into request-scoped budget - Router selects tier from classify confidence, intent risk, escalation count - Supervisor short-circuits via regex fast-path when rule confidence is high - Onboarding skill loads user escalation history to inform routing
99 lines
2.5 KiB
TypeScript
99 lines
2.5 KiB
TypeScript
import { describe, expect, it } from "vitest";
|
|
|
|
import type { HandoffReason } from "@agent/types";
|
|
import { decide } from "./trigger.js";
|
|
|
|
describe("escalation trigger", () => {
|
|
it("turns explicit handoff reasons into structured requests", () => {
|
|
const reasons: HandoffReason[] = [
|
|
"out-of-scope",
|
|
"missing-tool",
|
|
"low-confidence",
|
|
"validation-failed",
|
|
"same-agent-revisit",
|
|
"max-hop",
|
|
"high-risk-action",
|
|
"policy-risk",
|
|
];
|
|
|
|
for (const reason of reasons) {
|
|
const decision = decide({
|
|
classified: { confidence: 0.8 },
|
|
reason,
|
|
targetAgent: "supervisor",
|
|
taskSpec: "review",
|
|
});
|
|
|
|
expect(decision).toMatchObject({
|
|
reason,
|
|
shouldHandoff: true,
|
|
handoff: {
|
|
reason,
|
|
targetAgent: "supervisor",
|
|
taskSpec: "review",
|
|
},
|
|
});
|
|
}
|
|
});
|
|
|
|
it("escalates exhausted budgets and validation rejects", () => {
|
|
expect(
|
|
decide({
|
|
budget: {
|
|
estimatedCostUsd: 0,
|
|
handoffs: 4,
|
|
maxCostUsd: 0.5,
|
|
maxHandoffs: 4,
|
|
maxSameAgentVisits: 2,
|
|
maxToolCalls: 20,
|
|
totalCostUsd: 0,
|
|
totalLlmCostUsd: 0,
|
|
totalToolCostUsd: 0,
|
|
toolCalls: 0,
|
|
visits: {},
|
|
},
|
|
}).handoff?.reason,
|
|
).toBe("max-hop");
|
|
|
|
expect(decide({ validation: { status: "reject" } }).handoff?.reason).toBe(
|
|
"validation-failed",
|
|
);
|
|
});
|
|
|
|
it("escalates low-confidence and uncertain high-risk responses", () => {
|
|
expect(
|
|
decide({ classified: { confidence: 0.2 } }).handoff?.reason,
|
|
).toBe("low-confidence");
|
|
|
|
expect(
|
|
decide({
|
|
capabilityCard: {
|
|
canHandle: ["policy"],
|
|
cannotHandle: [],
|
|
domain: "moderation",
|
|
handoffTargets: ["supervisor"],
|
|
outputContract: { format: "text", maxChars: "500" },
|
|
riskTier: "high",
|
|
},
|
|
classified: { confidence: 0.6 },
|
|
}).handoff?.reason,
|
|
).toBe("high-risk-action");
|
|
});
|
|
|
|
it("does not escalate ordinary confident responses", () => {
|
|
expect(
|
|
decide({
|
|
capabilityCard: {
|
|
canHandle: ["help"],
|
|
cannotHandle: [],
|
|
domain: "onboarding",
|
|
handoffTargets: ["supervisor"],
|
|
outputContract: { format: "text", maxChars: "500" },
|
|
riskTier: "medium",
|
|
},
|
|
classified: { confidence: 0.9 },
|
|
}),
|
|
).toEqual({ shouldHandoff: false });
|
|
});
|
|
});
|