auto-model-router 0.33.0 → 0.35.0

package/test/toast-logic.test.ts

@@ -176,6 +176,61 @@ describe("selectToasts", () => {
 		expect(toasts).toHaveLength(1);
 		expect(toasts[0]?.model).toBe("keep");
 	});
+
+	test("task (subagent) turns toast even though they carry their own session id", async () => {
+		// A task's dispatches carry the subagent's own session id and the
+		// isSubagent flag, so the session filter must not drop them.
+		const entries = [
+			dec({ id: "d2", slug: "x/task-model", ompSessionId: "task-1", features: { isSubagent: true } }),
+			dec({ id: "d1", slug: "prior", ompSessionId: "sess-a" }),
+		];
+		const subModels = new Map<string, string | null>();
+		const toasts = selectToasts(entries, "d1", "", "sess-a", true, subModels);
+		expect(toasts).toHaveLength(1);
+		expect(toasts[0]?.model).toBe("x/task-model");
+		expect(toasts[0]!.text).toContain("task");
+	});
+
+	test("a task toasts once per model: the first dispatch and every change after", async () => {
+		const entries = [
+			dec({ id: "d5", slug: "x/m2", ompSessionId: "task-1", features: { isSubagent: true } }),
+			dec({ id: "d4", slug: "x/m1", ompSessionId: "task-1", features: { isSubagent: true } }),
+			dec({ id: "d3", slug: "x/m1", ompSessionId: "task-1", features: { isSubagent: true } }),
+			dec({ id: "d2", slug: "x/m1", ompSessionId: "task-1", features: { isSubagent: true } }),
+		];
+		const subModels = new Map<string, string | null>();
+		const first = selectToasts(entries, "d1", "", "sess-a", true, subModels);
+		// oldest→newest: m1 (first dispatch toasts), m1 (skip), m1 (skip), m2 (change, toasts)
+		expect(first.map((t) => t.model)).toEqual(["x/m1", "x/m2"]);
+		// The next tick remembers the task's last model: the same model again is not news.
+		const again = selectToasts([dec({ id: "d6", slug: "x/m2", ompSessionId: "task-1", features: { isSubagent: true } })], "d5", "", "sess-a", true, subModels);
+		expect(again).toHaveLength(0);
+		// A second task with its own session id toasts its own first dispatch.
+		const other = selectToasts([dec({ id: "d7", slug: "x/m3", ompSessionId: "task-2", features: { isSubagent: true } })], "d6", "", "sess-a", true, subModels);
+		expect(other).toHaveLength(1);
+		expect(other[0]?.model).toBe("x/m3");
+	});
+
+	test("task toasts respect the harness filter: another member's tasks stay out", async () => {
+		const entries = [
+			dec({ id: "d2", slug: "theirs", ompSessionId: "task-9", harnessId: "member-b", features: { isSubagent: true } }),
+			dec({ id: "d1", slug: "mine", ompSessionId: "task-8", harnessId: "member-a", features: { isSubagent: true } }),
+			dec({ id: "d0", slug: "prior", ompSessionId: "sess-a", harnessId: "member-a" }),
+		];
+		const subModels = new Map<string, string | null>();
+		const toasts = selectToasts(entries, "d0", "member-a", "sess-a", true, subModels);
+		expect(toasts).toHaveLength(1);
+		expect(toasts[0]?.model).toBe("mine");
+	});
+
+	test("a main-session turn still toasts every dispatch, unchanged", async () => {
+		const entries = [
+			dec({ id: "d3", slug: "x/m1", ompSessionId: "sess-a" }),
+			dec({ id: "d2", slug: "x/m1", ompSessionId: "sess-a" }),
+			dec({ id: "d1", slug: "prior", ompSessionId: "sess-a" }),
+		];
+		expect(selectToasts(entries, "d1", "", "sess-a", true)).toHaveLength(2);
+	});
 });
 
 describe("toToastText", () => {

package/test/turn.test.ts

@@ -34,7 +34,7 @@ import type {
 function mkConfig(escalation: Partial<EscalationConfig> = {}): RouterConfig {
 	return {
 		server: { host: "127.0.0.1", port: 8787, maxConcurrentTurns: 24, subagentProfile: "auto-sub" },
-		openrouter: { baseUrl: "https://openrouter.ai/api/v1", apiKey: "", title: "test", timeoutMs: 30_000, catalogTtlMs: 3_600_000, catalogRefreshMs: 0 },
+		openrouter: { baseUrl: "https://openrouter.ai/api/v1", apiKey: "", title: "test", timeoutMs: 30_000, catalogTtlMs: 3_600_000, catalogRefreshMs: 0, minCreditsUsd: 0, usagePollMs: 0 },
 		ollama: { enabled: false, baseUrl: "http://127.0.0.1:11434/v1", apiKey: "", timeoutMs: 30_000, catalogTtlMs: 300_000, includeLocal: false, prices: {}, twins: {}, costBias: 1, biasUntilUsage: 0.9, usagePollMs: 0, quotaCooldownMs: 0, rateLimitCooldownMs: 0, planCreditsUsd: 0 },
 		upstreams: [],
 		benchmarks: { enabled: false, artificialAnalysisApiKey: "", benchlm: true, refreshMs: 86_400_000, timeoutMs: 30_000, useLocalScores: false },

package/test/upstreams.test.ts

@@ -193,6 +193,8 @@ describe("the OpenAI-compatible client", () => {
 	test("statuses: OpenAI's insufficient_quota 429 is the account, a plain 429 the moment; 400 context is final", async () => {
 		expect(classifyCompatStatus("x", 429, { error: { code: "insufficient_quota", message: "You exceeded your current quota" } })).toMatchObject({ kind: "quota", retryable: true });
 		expect(classifyCompatStatus("x", 429, { error: { message: "Rate limit reached" } })).toMatchObject({ kind: "rate_limit", retryable: true });
+		// Kimi's balance wording: 429 but the account, so the 15-minute quota cooldown hides the whole upstream.
+		expect(classifyCompatStatus("kimi", 429, { error: { message: "This request would exceed your available credits given your current in-flight requests" } })).toMatchObject({ kind: "quota", retryable: true });
 		expect(classifyCompatStatus("x", 400, { error: { message: "This model's maximum context length is 8192 tokens" } })).toMatchObject({ kind: "context_length", retryable: false });
 		expect(classifyCompatStatus("x", 401, {})).toMatchObject({ kind: "auth", retryable: false });
 		expect(classifyCompatStatus("x", 503, {})).toMatchObject({ kind: "upstream_error", retryable: true });
@@ -238,6 +240,18 @@ describe("the OpenAI-compatible client", () => {
 		expect(caught).toMatchObject({ kind: "quota", retryable: true });
 		expect(broke.available()).toBe(false);
 		expect(broke.lastTrip()?.kind).toBe("quota");
+		const kimiBroke = createCompatClient(cfg, "openai-direct", async () =>
+			Response.json({ error: { message: "This request would exceed your available credits given your current in-flight requests" } }, { status: 429 }),
+		);
+		caught = null;
+		try {
+			await kimiBroke.dispatch({ body: { model: "openai-direct/gpt-4o", messages: [] }, sessionId: "s", signal: new AbortController().signal });
+		} catch (err) {
+			caught = err;
+		}
+		expect(caught).toMatchObject({ kind: "quota", retryable: true });
+		expect(kimiBroke.available()).toBe(false);
+		expect(kimiBroke.lastTrip()?.kind).toBe("quota");
 		// The live key applies to the next call without a new client.
 		applyConfigPatch(cfg, { upstreams: [{ id: "openai-direct", kind: "openai", baseUrl: "https://api.openai.com/v1", apiKey: "sk-new", models: [{ id: "gpt-4o", input: 2.5, output: 10 }] }] } as never);
 		await client.dispatch({ body: { model: "openai-direct/gpt-4o", messages: [], stream: true }, sessionId: "s", signal: new AbortController().signal });
@@ -363,8 +377,20 @@ describe("the Anthropic client", () => {
 		}
 		expect(caught).toMatchObject({ kind: "upstream_error", retryable: true });
 		expect(overloaded.available()).toBe(false);
+		const billedOut = createAnthropicClient(cfg, "anthropic-direct", async () => Response.json({ error: { message: "Your credit balance is too low" } }, { status: 402 }));
+		caught = null;
+		try {
+			await billedOut.dispatch({ body: { model: "anthropic-direct/claude-sonnet-4", messages: [] }, sessionId: "s", signal: new AbortController().signal });
+		} catch (err) {
+			caught = err;
+		}
+		expect(caught).toMatchObject({ kind: "quota", retryable: true });
+		expect(billedOut.available()).toBe(false);
+		expect(billedOut.lastTrip()?.kind).toBe("quota");
 		expect(classifyAnthropicStatus("a", 400, { error: { message: "prompt is too long: 250000 tokens" } })).toMatchObject({ kind: "context_length", retryable: false });
 		expect(classifyAnthropicStatus("a", 401, {})).toMatchObject({ kind: "auth", retryable: false });
+		// The credit card said no (402): the account, not the moment — quota, and the breaker hides the upstream.
+		expect(classifyAnthropicStatus("a", 402, { error: { message: "Your credit balance is too low" } })).toMatchObject({ kind: "quota", retryable: true });
 	});
 
 	test("a subscription upstream: OAuth bearer instead of x-api-key, and the Claude Code identity leads the system blocks", async () => {