muonroi-cli 1.8.1 → 1.8.3

package/dist/src/orchestrator/__tests__/stream-runner.test.js

@@ -1,202 +0,0 @@
-// Phase 12.3-03 — StreamRunner unit tests.
-//
-// Smoke-only: setup short-circuit paths (unknown agent, computer-on-textonly
-// runtime), DI surface invariants. Stream integration is covered by
-// tests/harness/cost-leak-{f1,g1,b3,b4,c3}.spec.ts — those exercise real
-// streamText with MockLanguageModelV3.
-import { beforeAll, describe, expect, it } from "vitest";
-import { getTestModelForProvider, getTestModels, getTestProviders } from "../../__test-helpers__/catalog-fixtures.js";
-import { loadCatalog } from "../../models/registry.js";
-import { computePromptCacheKey } from "../../providers/runtime.js";
-import { StreamRunner } from "../stream-runner.js";
-beforeAll(async () => {
-    await loadCatalog();
-});
-function makeBashStub() {
-    return {
-        getCwd: () => process.cwd(),
-        getSandboxMode: () => "off",
-        getSandboxSettings: () => ({}),
-    };
-}
-function makeDeps(overrides = {}) {
-    const testModels = getTestModels();
-    const testProviders = getTestProviders();
-    return {
-        getProvider: () => (() => null),
-        resolveModelForTask: () => testModels.fast,
-        getModelId: () => testModels.fast,
-        getProviderId: () => testProviders.default,
-        getBash: () => makeBashStub(),
-        getMaxToolRounds: () => 50,
-        getMaxTokens: () => 8192,
-        isBatchApiEnabled: () => false,
-        getCrossTurnDedup: () => null,
-        getReadBudget: () => null,
-        recordUsage: () => { },
-        setCurrentCallId: () => { },
-        setLastProviderOptionsShape: () => { },
-        getSessionId: () => undefined,
-        runTaskRequestBatch: async () => ({
-            success: false,
-            output: "batch path not exercised in this test",
-        }),
-        ...overrides,
-    };
-}
-describe("StreamRunner — setup short-circuit paths", () => {
-    it("returns unknown-agent short-circuit when agent kind is not recognised", async () => {
-        const runner = new StreamRunner(makeDeps());
-        const request = {
-            agent: "definitely-not-a-real-agent",
-            description: "test",
-            prompt: "noop",
-        };
-        const outcome = await runner.setup(request);
-        expect(outcome.kind).toBe("short-circuit");
-        if (outcome.kind === "short-circuit") {
-            expect(outcome.result.success).toBe(false);
-            expect(outcome.result.output).toContain(`Unknown sub-agent "definitely-not-a-real-agent"`);
-            expect(outcome.result.task?.agent).toBe("definitely-not-a-real-agent");
-        }
-    });
-    it("returns the same ToolResult shape that the in-line orchestrator did (parity check)", async () => {
-        // The pre-12.3 orchestrator returned { success: false, output: <msg>,
-        // task: { agent, description, summary: <msg> } } for unknown agents. Pin
-        // that shape so a future refactor that drops `task.summary` is caught.
-        const runner = new StreamRunner(makeDeps());
-        const outcome = await runner.setup({ agent: "??", description: "d", prompt: "p" });
-        expect(outcome.kind).toBe("short-circuit");
-        if (outcome.kind === "short-circuit") {
-            expect(outcome.result.task?.description).toBe("d");
-            expect(outcome.result.task?.summary).toEqual(outcome.result.output);
-        }
-    });
-});
-describe("StreamRunner — cheap-model steering on sub-agent system prompt", () => {
-    // The sub-agent prompt must carry the SAME front-loaded steering stack as the
-    // top-level turn for fast-tier models (regression: previously only the
-    // playbook was injected, leaving the sub-agent without the anti-ramble
-    // workbook or the shell directive). Final prompt order: [ENV] → [CRITICAL
-    // playbook] → [CONVERGENCE workbook] → base.
-    //
-    // `explore` is used because its childMode is "ask" (no MCP load) so setup()
-    // completes synchronously without spawning anything.
-    it("front-loads [ENV] → [CRITICAL] → [CONVERGENCE] for a fast-tier explore sub-agent", async () => {
-        const runner = new StreamRunner(makeDeps());
-        const outcome = await runner.setup({ agent: "explore", description: "scan", prompt: "find auth wiring" });
-        expect(outcome.kind).toBe("prepared");
-        if (outcome.kind !== "prepared")
-            return;
-        const sys = outcome.prepared.childSystem;
-        const iEnv = sys.indexOf("[ENV]");
-        const iCritical = sys.indexOf("[CRITICAL TOOL-USE RULES");
-        const iConvergence = sys.indexOf("[CONVERGENCE");
-        expect(sys.startsWith("[ENV]")).toBe(true);
-        expect(iEnv).toBeGreaterThanOrEqual(0);
-        expect(iCritical).toBeGreaterThan(iEnv);
-        expect(iConvergence).toBeGreaterThan(iCritical);
-        // explore → analyze workbook addendum.
-        expect(sys).toMatch(/do not read the whole codebase/i);
-    });
-});
-describe("StreamRunner — F1 promptCacheKey parity on the sub-agent path", () => {
-    // The top-level turn derives a stable openai.promptCacheKey per turn via
-    // buildTurnProviderOptions({ sessionId }) so every tool round in a session
-    // routes to the same cache node (message-processor.ts). The sub-agent path
-    // previously used only resolve-time childRuntime.providerOptions, which has
-    // NO promptCacheKey — so every explore/verify OpenAI call auto-hashed its
-    // prompt and fragmented the cache, re-billing the unchanging prefix. setup()
-    // must now thread the session id so sub-agents get the SAME stable key.
-    it("threads sessionId → openai.promptCacheKey onto the sub-agent providerOptions", async () => {
-        const sessionId = "sess-fixed-for-f1-parity";
-        const openaiModel = getTestModelForProvider("openai", "fast");
-        const runner = new StreamRunner(makeDeps({
-            resolveModelForTask: () => openaiModel,
-            getSessionId: () => sessionId,
-        }));
-        const outcome = await runner.setup({ agent: "explore", description: "scan", prompt: "find auth wiring" });
-        expect(outcome.kind).toBe("prepared");
-        if (outcome.kind !== "prepared")
-            return;
-        const opts = outcome.prepared.childProviderOptions;
-        expect(opts?.openai?.promptCacheKey).toBe(computePromptCacheKey(sessionId));
-    });
-    it("omits promptCacheKey when there is no session id (headless one-shot)", async () => {
-        const openaiModel = getTestModelForProvider("openai", "fast");
-        const runner = new StreamRunner(makeDeps({
-            resolveModelForTask: () => openaiModel,
-            getSessionId: () => undefined,
-        }));
-        const outcome = await runner.setup({ agent: "explore", description: "scan", prompt: "find auth wiring" });
-        expect(outcome.kind).toBe("prepared");
-        if (outcome.kind !== "prepared")
-            return;
-        const opts = outcome.prepared.childProviderOptions;
-        expect(opts?.openai?.promptCacheKey).toBeUndefined();
-    });
-});
-describe("StreamRunner — DI surface", () => {
-    it("constructs without invoking any dep callbacks", () => {
-        let touched = 0;
-        const trip = () => {
-            touched++;
-            throw new Error("touched");
-        };
-        // Every callback throws — instantiation must not call any of them.
-        const runner = new StreamRunner({
-            getProvider: trip,
-            resolveModelForTask: trip,
-            getModelId: trip,
-            getProviderId: trip,
-            getBash: trip,
-            getMaxToolRounds: trip,
-            getMaxTokens: trip,
-            isBatchApiEnabled: trip,
-            getCrossTurnDedup: trip,
-            getReadBudget: trip,
-            recordUsage: trip,
-            setCurrentCallId: trip,
-            setLastProviderOptionsShape: trip,
-            getSessionId: trip,
-            runTaskRequestBatch: trip,
-        });
-        expect(runner).toBeInstanceOf(StreamRunner);
-        expect(touched).toBe(0);
-    });
-    it("each runner instance is independent (no shared mutable state on the class)", () => {
-        const a = new StreamRunner(makeDeps());
-        const b = new StreamRunner(makeDeps());
-        expect(a).not.toBe(b);
-    });
-    describe("StreamRunner — maxSteps custom overrides", () => {
-        it("respects default maxSteps logic when request.maxToolRounds is undefined", async () => {
-            const deps = makeDeps({ getMaxToolRounds: () => 50 });
-            const runner = new StreamRunner(deps);
-            const outcome = await runner.setup({
-                agent: "explore",
-                description: "test",
-                prompt: "test",
-            });
-            expect(outcome.kind).toBe("prepared");
-            if (outcome.kind === "prepared") {
-                expect(outcome.prepared.maxSteps).toBe(100); // deps.getMaxToolRounds() * 2
-            }
-        });
-        it("uses request.maxToolRounds when it is explicitly defined", async () => {
-            const deps = makeDeps({ getMaxToolRounds: () => 50 });
-            const runner = new StreamRunner(deps);
-            const outcome = await runner.setup({
-                agent: "explore",
-                description: "test",
-                prompt: "test",
-                maxToolRounds: 75,
-            });
-            expect(outcome.kind).toBe("prepared");
-            if (outcome.kind === "prepared") {
-                expect(outcome.prepared.maxSteps).toBe(75);
-            }
-        });
-    });
-});
-//# sourceMappingURL=stream-runner.test.js.map

package/dist/src/orchestrator/__tests__/sub-agent-model-tier.test.d.ts

@@ -1 +0,0 @@
-export {};

package/dist/src/orchestrator/__tests__/sub-agent-model-tier.test.js

@@ -1,45 +0,0 @@
-import { describe, expect, it } from "vitest";
-import { resolveModelForTask, TASK_TIER_PREFS } from "../sub-agent-model-tier.js";
-// A fake provider catalog: "acme" has a full ladder; "premOnly" has premium only.
-function makeLookup() {
-    const table = {
-        acme: {
-            fast: { id: "acme-fast", provider: "acme" },
-            balanced: { id: "acme-balanced", provider: "acme" },
-            premium: { id: "acme-premium", provider: "acme" },
-        },
-        premOnly: {
-            premium: { id: "prem-premium", provider: "premOnly" },
-        },
-    };
-    return (tier, preferProvider) => (preferProvider ? table[preferProvider]?.[tier] : undefined);
-}
-describe("sub-agent model tier policy", () => {
-    const lookup = makeLookup();
-    it("downgrades a general sub-agent to balanced (gap #2 fix)", () => {
-        expect(resolveModelForTask("general", "acme", "acme-premium", lookup)).toBe("acme-balanced");
-    });
-    it("keeps verify on premium (rigor over cost)", () => {
-        expect(resolveModelForTask("verify", "acme", "acme-premium", lookup)).toBe("acme-premium");
-    });
-    it("explore prefers balanced", () => {
-        expect(resolveModelForTask("explore", "acme", "acme-premium", lookup)).toBe("acme-balanced");
-    });
-    it("compact/title prefer the cheapest tier", () => {
-        expect(resolveModelForTask("compact", "acme", "acme-premium", lookup)).toBe("acme-fast");
-        expect(resolveModelForTask("title", "acme", "acme-premium", lookup)).toBe("acme-fast");
-    });
-    it("falls back to the parent model when the provider has no cheaper tier", () => {
-        // premOnly has only premium → general's [balanced, premium] resolves premium.
-        expect(resolveModelForTask("general", "premOnly", "prem-premium", lookup)).toBe("prem-premium");
-    });
-    it("never crosses providers — unknown provider yields the fallback model", () => {
-        expect(resolveModelForTask("general", "unknown", "parent-model", lookup)).toBe("parent-model");
-    });
-    it("policy intent is documented in the tier map", () => {
-        // general downgrades (balanced before premium); verify does not.
-        expect(TASK_TIER_PREFS.general.indexOf("balanced")).toBeLessThan(TASK_TIER_PREFS.general.indexOf("premium"));
-        expect(TASK_TIER_PREFS.verify.indexOf("premium")).toBeLessThan(TASK_TIER_PREFS.verify.indexOf("balanced"));
-    });
-});
-//# sourceMappingURL=sub-agent-model-tier.test.js.map

package/dist/src/orchestrator/__tests__/sub-session-delegation.test.d.ts

@@ -1 +0,0 @@
-export {};