Claude Code. Any Model. The most powerful AI coding agent now speaks every language.
test: an empty catalog means no-route, not a guessed chain
Jack Rudenko008c494c354f
diff --git a/packages/cli/src/providers/routing-rules.test.ts b/packages/cli/src/providers/routing-rules.test.tsindex 4db4d6a1..8447a7b6 100644--- a/packages/cli/src/providers/routing-rules.test.ts+++ b/packages/cli/src/providers/routing-rules.test.ts@@ -689,26 +689,16 @@ describe("route()", () => {expect(plan.primary.provider).toBe("openai");});- test("gpt-5 (bare) with OPENAI_CODEX_API_KEY → primary openai-codex", async () => {- // Assert rule composition with an empty catalog (an absent cache file),- // not Codex model support: live Codex returns 400 with "The 'gpt-5' model- // is not supported when using Codex with a ChatGPT account".- //- // In a dev environment where codex-oauth.json exists, codex is genuinely- // credentialed regardless of the fake env key. Skip the strict assertion- // there, matching the sibling credential test. openai-codex declares no- // modelDiscovery, so only the catalog cache, not live discovery, is a factor.- const codexOauth = join(homedir(), ".claudish", "codex-oauth.json");- if (existsSync(codexOauth)) return;-+ test("gpt-5 (bare) with OPENAI_CODEX_API_KEY and no catalog → no-route", async () => {const dir = mkdtempSync(join(tmpdir(), "claudish-routing-empty-catalog-test-"));const cachePath = join(dir, "all-models.json");try {process.env.OPENAI_CODEX_API_KEY = "sk-codex-test";const plan = await route("gpt-5", {}, undefined, cachePath);- expect(plan.kind).toBe("ok");- if (plan.kind !== "ok") return;- expect(plan.primary.provider).toBe("openai-codex");+ expect(plan.kind).toBe("no-route");+ if (plan.kind !== "no-route") return;+ expect(plan.reason).toContain("No model catalog available");+ expect(plan.hint).toContain("claudish --models-refresh");} finally {rmSync(dir, { recursive: true, force: true });}@@ -966,18 +956,7 @@ describe("route()", () => {expect(plan.kind).toBe("no-route");});- test("ok plan returns primary plus fallbacks in order", async () => {- // Assert rule composition with an empty catalog (an absent cache file),- // not Codex model support: live Codex returns 400 with "The 'gpt-5' model- // is not supported when using Codex with a ChatGPT account".- //- // In a dev environment where codex-oauth.json exists, codex is genuinely- // credentialed regardless of the fake env key. Skip the strict assertion- // there, matching the sibling credential test. openai-codex declares no- // modelDiscovery, so only the catalog cache, not live discovery, is a factor.- const codexOauth = join(homedir(), ".claudish", "codex-oauth.json");- if (existsSync(codexOauth)) return;-+ test("gpt-5 (bare) with all fallback credentials and no catalog → no-route", async () => {const dir = mkdtempSync(join(tmpdir(), "claudish-routing-empty-catalog-test-"));const cachePath = join(dir, "all-models.json");try {@@ -985,10 +964,10 @@ describe("route()", () => {process.env.OPENAI_API_KEY = "oai-test";process.env.OPENROUTER_API_KEY = "or-test";const plan = await route("gpt-5", {}, undefined, cachePath);- expect(plan.kind).toBe("ok");- if (plan.kind !== "ok") return;- expect(plan.primary.provider).toBe("openai-codex");- expect(plan.fallbacks.map((r) => r.provider)).toEqual(["openai", "openrouter"]);+ expect(plan.kind).toBe("no-route");+ if (plan.kind !== "no-route") return;+ expect(plan.reason).toContain("No model catalog available");+ expect(plan.hint).toContain("claudish --models-refresh");} finally {rmSync(dir, { recursive: true, force: true });}