llm provider and test renderer

packages/opencode/specs/property-based-tui-testing-working-notes.md

@@ -0,0 +1,27 @@
+# Property-Based TUI Testing Working Notes
+
+## Mock LLM Provider
+
+- `SessionPrompt` gets model metadata through `Provider.Service.getModel(...)`.
+- Actual generation is routed through `LLM.Service` / `streamText(...)`, so the first mock should return an AI SDK `LanguageModelV3` from the provider service.
+- The normal `Provider.layer` builds providers from config/models.dev/plugin state. For first pass, the simulated graph can replace `Provider.Service` with a smaller simulation provider service instead of trying to flow through provider config.
+- Control state should own an ordered LLM script queue. The provider/model should consume from that queue when the AI SDK calls the language model.
+- First version can support text-only output. Tool calls and stream chunk fidelity can come next.
+- Missing script should fail loudly with a typed simulation error, not silently return an empty assistant message.
+- Implemented `SimulationProvider.layer`, replacing `Provider.Service` in `createSimulatedRoutes`.
+- The provider exposes provider `simulation` and model `mock`.
+- `doGenerate` and `doStream` both consume one queued script through `Simulation.Service.nextLLM()`.
+- Current script support: `text`, `thinking` (treated as text for now), and `error`.
+- Snapshot currently records `llmQueued` and `llmConsumed`, not per-step details yet.
+
+## OpenTUI Fake Renderer
+
+- OpenTUI Solid exposes `testRender(...)` from `@opentui/solid`.
+- The lower-level core API is `createTestRenderer(...)` from `@opentui/core/testing`.
+- `createTestRenderer(...)` returns `renderer`, `mockInput`, `mockMouse`, `renderOnce`, `captureCharFrame`, `captureSpans`, and `resize`.
+- `captureCharFrame()` is the simple screen-buffer string API used heavily in OpenTUI snapshots.
+- `captureSpans()` returns structured lines/spans plus cursor position, which is a better starting point for visible element discovery than parsing raw characters.
+- `mockInput` supports interactions like `typeText`, `pressEnter`, and `pressArrow`.
+- Implemented `TuiSimulation.createSimulationRenderer(...)` beside `thread.ts`. It creates a test renderer and exposes `renderOnce`, `screen`, `spans`, and `destroy`.
+- `thread.ts` checks `OPENCODE_SIMULATION`, creates the fake renderer there, starts the normal worker/backend, and passes the renderer into `tui(...)`.
+- `tui(...)` now accepts an injected `CliRenderer`, test mode, and an `onReady` callback. Production still creates the real renderer.

packages/opencode/specs/property-based-tui-testing.md

@@ -199,7 +199,9 @@ Implementation shape:
 - First pass uses a raw route wrapper at `packages/opencode/src/server/routes/instance/httpapi/simulation.ts` to avoid SDK regeneration while the API shape is still moving.
 - Current control service can reset state, seed filesystem files, register static network responses, and return a snapshot.
 - Register/configure a local mock provider/model through the normal provider path.
-- The mock model reads scripts from simulation control state.
+- The simulated route graph replaces `Provider.Service` with `SimulationProvider.layer`.
+- The mock model reads queued scripts from simulation control state.
+- Current mock provider supports text/thinking/error actions for the first step only. Tool calls and multi-round step selection are still pending.
 - No JSON-in-prompt fallback.
 - Missing script means typed simulation error.
 
@@ -236,10 +238,11 @@ Todos:
 - [x] Add simulation control state and reset semantics.
 - [x] Add gated simulation endpoints for reset, filesystem seed, network register, and snapshot.
 - [x] Decide raw route vs typed HttpApi route. Raw route for first pass; no SDK regeneration yet.
-- [ ] Implement mock provider/model on the normal provider path.
-- [ ] Port the useful stream chunk behavior from the old branch to the current AI SDK interface.
-- [ ] Make missing scripts fail with a typed simulation error.
-- [ ] Record consumed script step in simulation snapshot.
+- [x] Implement mock provider/model on the normal provider path.
+- [x] Make missing scripts fail with a typed simulation error.
+- [x] Record consumed script count in simulation snapshot.
+- [ ] Support tool-call script actions.
+- [ ] Support multi-step script selection after tool result rounds.
 - [ ] Verify `session.prompt_async` exercises real `SessionPrompt` and `SessionProcessor`.
 
 ## OpenTUI Fake Renderer And Interactable Elements
@@ -255,6 +258,8 @@ Known starting points:
 Implementation shape:
 
 - Add a renderer factory/testing hook to `tui(...)` so tests can pass a fake renderer.
+- Current first pass checks `OPENCODE_SIMULATION` in `cli/cmd/tui/thread.ts`, starts the normal worker/backend, and injects an OpenTUI test renderer into `tui(...)`.
+- Fake renderer setup lives in `cli/cmd/tui/simulation.ts` and returns `renderOnce`, `screen`, and `spans` helpers for the thread-side simulation runner.
 - Do not render to a real terminal in simulation mode.
 - Investigate OpenTUI APIs for walking the render tree and extracting focusable/clickable/editable elements.
 - Investigate OpenTUI APIs for reading the screen buffer from the fake renderer.
@@ -262,11 +267,11 @@ Implementation shape:
 
 Todos:
 
-- [ ] Inspect `@opentui/core/testing` `createTestRenderer` capabilities.
-- [ ] Inspect `@opentui/solid` `testRender` capabilities.
-- [ ] Determine how to get a screen buffer string/snapshot from the fake renderer.
-- [ ] Determine how to iterate renderables and identify interactable elements.
-- [ ] Add a minimal renderer factory override to `tui(...)` or app startup.
+- [x] Inspect `@opentui/core/testing` `createTestRenderer` capabilities.
+- [x] Inspect `@opentui/solid` `testRender` capabilities.
+- [x] Determine how to get a screen buffer string/snapshot from the fake renderer.
+- [x] Determine first structured capture API for interactable discovery: `captureSpans()`.
+- [x] Add a minimal renderer factory override to `tui(...)` or app startup.
 - [ ] Expose prompt ref, route, sync state, keymap, and renderer to the simulation harness.
 - [ ] Verify TUI starts in fake renderer with no real terminal output.
 - [ ] Verify screen buffer can be captured after a render.

packages/opencode/src/cli/cmd/tui/app.tsx

@@ -3,7 +3,7 @@ import { createDefaultOpenTuiKeymap } from "@opentui/keymap/opentui"
 import * as Clipboard from "@tui/util/clipboard"
 import * as Selection from "@tui/util/selection"
 import * as TuiAudio from "@tui/util/audio"
-import { createCliRenderer, MouseButton, type CliRendererConfig } from "@opentui/core"
+import { createCliRenderer, MouseButton, type CliRenderer, type CliRendererConfig } from "@opentui/core"
 import { RouteProvider, useRoute } from "@tui/context/route"
 import {
   Switch,
@@ -165,6 +165,9 @@ export function tui(input: {
   fetch?: typeof fetch
   headers?: RequestInit["headers"]
   events?: EventSource
+  renderer?: CliRenderer
+  mode?: "dark" | "light"
+  onReady?: (ctx: { renderer: CliRenderer }) => void
 }) {
   // promise to prevent immediate exit
   // oxlint-disable-next-line no-async-promise-executor -- intentional: async executor used for sequential setup before resolve
@@ -182,10 +185,10 @@ export function tui(input: {
       TuiAudio.dispose()
     }
 
-    const renderer = await createCliRenderer(rendererConfig(input.config))
+    const renderer = input.renderer ?? (await createCliRenderer(rendererConfig(input.config)))
     // Prewarm palette before ThemeProvider mounts so `system` theme avoids a first-paint fallback flash.
     void renderer.getPalette({ size: 16 }).catch(() => undefined)
-    const mode = (await renderer.waitForThemeMode(1000)) ?? "dark"
+    const mode = input.mode ?? (await renderer.waitForThemeMode(1000)) ?? "dark"
 
     const keymap = createDefaultOpenTuiKeymap(renderer)
     const offKeymap = registerOpencodeKeymap(keymap, renderer, input.config)
@@ -256,6 +259,7 @@ export function tui(input: {
         </ErrorBoundary>
       )
     }, renderer)
+    input.onReady?.({ renderer })
   })
 }
 

packages/opencode/src/cli/cmd/tui/simulation.ts

@@ -0,0 +1,30 @@
+import type { CliRenderer } from "@opentui/core"
+import type { CapturedFrame } from "@opentui/core"
+
+export interface SimulationRenderer {
+  readonly renderer: CliRenderer
+  readonly renderOnce: () => Promise<void>
+  readonly screen: () => string
+  readonly spans: () => CapturedFrame
+  readonly destroy: () => void
+}
+
+export async function createSimulationRenderer(): Promise<SimulationRenderer> {
+  const { createTestRenderer } = await import("@opentui/core/testing")
+  const setup = await createTestRenderer({
+    width: Number(process.env.OPENCODE_SIMULATION_TUI_WIDTH) || 100,
+    height: Number(process.env.OPENCODE_SIMULATION_TUI_HEIGHT) || 40,
+    screenMode: "main-screen",
+    consoleMode: "disabled",
+  })
+
+  return {
+    renderer: setup.renderer,
+    renderOnce: setup.renderOnce,
+    screen: setup.captureCharFrame,
+    spans: setup.captureSpans,
+    destroy: () => setup.renderer.destroy(),
+  }
+}
+
+export * as TuiSimulation from "./simulation"

packages/opencode/src/cli/cmd/tui/thread.ts

@@ -21,6 +21,8 @@ import {
   sanitizedProcessEnv,
 } from "@opencode-ai/core/util/opencode-process"
 import { validateSession } from "./validate-session"
+import { Flag } from "@opencode-ai/core/flag/flag"
+import { TuiSimulation } from "./simulation"
 
 declare global {
   const OPENCODE_WORKER_PATH: string
@@ -229,26 +231,38 @@ export const TuiThreadCommand = cmd({
 
       try {
         const { tui } = await import("./app")
-        await tui({
-          url: transport.url,
-          async onSnapshot() {
-            const tui = writeHeapSnapshot("tui.heapsnapshot")
-            const server = await client.call("snapshot", undefined)
-            return [tui, server]
-          },
-          config,
-          directory: cwd,
-          fetch: transport.fetch,
-          events: transport.events,
-          args: {
-            continue: args.continue,
-            sessionID: args.session,
-            agent: args.agent,
-            model: args.model,
-            prompt,
-            fork: args.fork,
-          },
-        })
+        const simulationRenderer = Flag.OPENCODE_SIMULATION ? await TuiSimulation.createSimulationRenderer() : undefined
+        try {
+          await tui({
+            url: transport.url,
+            async onSnapshot() {
+              const tui = writeHeapSnapshot("tui.heapsnapshot")
+              const server = await client.call("snapshot", undefined)
+              return [tui, server]
+            },
+            config,
+            directory: cwd,
+            fetch: transport.fetch,
+            events: transport.events,
+            renderer: simulationRenderer?.renderer,
+            mode: simulationRenderer ? "dark" : undefined,
+            onReady: simulationRenderer
+              ? async () => {
+                  await simulationRenderer.renderOnce()
+                }
+              : undefined,
+            args: {
+              continue: args.continue,
+              sessionID: args.session,
+              agent: args.agent,
+              model: args.model,
+              prompt,
+              fork: args.fork,
+            },
+          })
+        } finally {
+          simulationRenderer?.destroy()
+        }
       } finally {
         await stop()
       }

packages/opencode/src/server/routes/instance/httpapi/server.ts

@@ -64,6 +64,7 @@ import { Workspace } from "@/control-plane/workspace"
 import { SimulationFileSystem } from "@/testing/simulation/filesystem"
 import { SimulationNetwork } from "@/testing/simulation/network"
 import { SimulationNetworkRoutes } from "@/testing/simulation/network-routes"
+import { SimulationProvider } from "@/testing/simulation/provider"
 import { Simulation } from "@/testing/simulation/service"
 import { CorsConfig, isAllowedCorsOrigin, type CorsOptions } from "@/server/cors"
 import { serveUIEffect } from "@/server/shared/ui"
@@ -284,7 +285,7 @@ export function createSimulatedRoutes(corsOptions?: CorsOptions): ReturnType<typ
     Plugin.layer,
     Project.layer,
     ProviderAuth.layer,
-    Provider.layer,
+    SimulationProvider.layer,
     Pty.layer,
     PtyTicket.layer,
     Question.layer,

packages/opencode/src/server/routes/instance/httpapi/simulation.ts

@@ -38,7 +38,12 @@ export const simulationRoute = HttpRouter.use((router) =>
     )
 
     yield* router.add("POST", "/experimental/simulation/llm/enqueue", () =>
-      json(Effect.succeed({ ok: false, skipped: "mock provider scripts are not implemented yet" })),
+      json(
+        Effect.gen(function* () {
+          const input = yield* HttpServerRequest.schemaBodyJson(Simulation.LLMEnqueueInput)
+          return yield* simulation.enqueueLLM(input)
+        }),
+      ),
     )
 
     yield* router.add("GET", "/experimental/simulation/snapshot", () => json(simulation.snapshot()))

packages/opencode/src/testing/simulation/provider.ts

@@ -0,0 +1,142 @@
+import type { LanguageModelV3, LanguageModelV3CallOptions, LanguageModelV3FinishReason } from "@ai-sdk/provider"
+import { Effect, Layer } from "effect"
+import { Provider } from "@/provider/provider"
+import { ModelID, ProviderID } from "@/provider/schema"
+import { Simulation, type LLMScript } from "./service"
+
+const providerID = ProviderID.make("simulation")
+const modelID = ModelID.make("mock")
+
+const model: Provider.Model = {
+  id: modelID,
+  providerID,
+  api: { id: modelID, url: "simulation://mock", npm: "simulation" },
+  name: "Simulation Mock",
+  capabilities: {
+    temperature: true,
+    reasoning: true,
+    attachment: false,
+    toolcall: false,
+    input: { text: true, audio: false, image: false, video: false, pdf: false },
+    output: { text: true, audio: false, image: false, video: false, pdf: false },
+    interleaved: false,
+  },
+  cost: { input: 0, output: 0, cache: { read: 0, write: 0 } },
+  limit: { context: 128_000, output: 32_000 },
+  status: "active",
+  options: {},
+  headers: {},
+  release_date: "2026-01-01",
+  variants: {},
+}
+
+const provider: Provider.Info = {
+  id: providerID,
+  name: "Simulation",
+  source: "custom",
+  env: [],
+  options: {},
+  models: { [modelID]: model },
+}
+
+function text(script: LLMScript) {
+  return script.steps[0]?.flatMap((item) => (item.type === "text" || item.type === "thinking" ? [item.content] : []))
+    .join("") ?? ""
+}
+
+function error(script: LLMScript) {
+  return script.steps[0]?.find((item) => item.type === "error")
+}
+
+function stream(script: LLMScript) {
+  return new ReadableStream({
+    start(controller) {
+      controller.enqueue({ type: "stream-start", warnings: [] })
+      let index = 0
+      for (const item of script.steps[0] ?? []) {
+        index++
+        if (item.type === "error") {
+          controller.enqueue({ type: "error", error: new Error(item.message) })
+          controller.close()
+          return
+        }
+        const id = `simulation-${item.type}-${index}`
+        if (item.type === "thinking") {
+          controller.enqueue({ type: "reasoning-start", id })
+          controller.enqueue({ type: "reasoning-delta", id, delta: item.content })
+          controller.enqueue({ type: "reasoning-end", id })
+          continue
+        }
+        controller.enqueue({ type: "text-start", id })
+        controller.enqueue({ type: "text-delta", id, delta: item.content })
+        controller.enqueue({ type: "text-end", id })
+      }
+      controller.enqueue({ type: "finish", finishReason: finishReason(script), usage: usage(script) })
+      controller.close()
+    },
+  })
+}
+
+function usage(script: LLMScript) {
+  return {
+    inputTokens: {
+      total: script.usage?.inputTokens ?? 0,
+      noCache: script.usage?.inputTokens ?? 0,
+      cacheRead: undefined,
+      cacheWrite: undefined,
+    },
+    outputTokens: {
+      total: script.usage?.outputTokens ?? text(script).length,
+      text: script.usage?.outputTokens ?? text(script).length,
+      reasoning: undefined,
+    },
+    raw: script.usage,
+  }
+}
+
+function finishReason(script: LLMScript): LanguageModelV3FinishReason {
+  return { unified: script.finish === "unknown" ? "other" : (script.finish ?? "stop"), raw: script.finish }
+}
+
+function language(simulation: Simulation.Interface): LanguageModelV3 {
+  return {
+    specificationVersion: "v3",
+    provider: "simulation",
+    modelId: modelID,
+    supportedUrls: {},
+    async doGenerate(_options: LanguageModelV3CallOptions) {
+      const script = await Effect.runPromise(simulation.nextLLM())
+      const err = error(script)
+      if (err?.type === "error") throw new Error(err.message)
+      return {
+        content: [{ type: "text", text: text(script) }],
+        finishReason: finishReason(script),
+        usage: usage(script),
+        warnings: [],
+      }
+    },
+    async doStream(_options: LanguageModelV3CallOptions) {
+      const script = await Effect.runPromise(simulation.nextLLM())
+      return { stream: stream(script) }
+    },
+  }
+}
+
+export const layer = Layer.effect(
+  Provider.Service,
+  Effect.gen(function* () {
+    const simulation = yield* Simulation.Service
+    const lang = language(simulation)
+    return Provider.Service.of({
+      list: () => Effect.succeed({ [providerID]: provider }),
+      getProvider: () => Effect.succeed(provider),
+      getModel: () => Effect.succeed(model),
+      getLanguage: () => Effect.succeed(lang),
+      closest: () => Effect.succeed({ providerID, modelID }),
+      getSmallModel: () => Effect.succeed(model),
+      defaultModel: () => Effect.succeed({ providerID, modelID }),
+    })
+  }),
+)
+
+export * as SimulationProvider from "./provider"

packages/opencode/src/testing/simulation/service.ts

@@ -38,20 +38,56 @@ export const NetworkRegisterInput = Schema.Union([
   }),
 ])
 
+export const LLMScriptAction = Schema.Union([
+  Schema.Struct({ type: Schema.Literal("text"), content: Schema.String }),
+  Schema.Struct({ type: Schema.Literal("thinking"), content: Schema.String }),
+  Schema.Struct({ type: Schema.Literal("error"), message: Schema.String }),
+])
+
+export type LLMScriptAction = typeof LLMScriptAction.Type
+
+export const LLMScript = Schema.Struct({
+  steps: Schema.Array(Schema.Array(LLMScriptAction)),
+  usage: Schema.optional(
+    Schema.Struct({
+      inputTokens: Schema.Number,
+      outputTokens: Schema.Number,
+      totalTokens: Schema.Number,
+    }),
+  ),
+  finish: Schema.optional(Schema.Literals(["stop", "tool-calls", "error", "length", "unknown"])),
+})
+
+export type LLMScript = typeof LLMScript.Type
+
+export const LLMEnqueueInput = Schema.Struct({
+  scripts: Schema.Array(LLMScript),
+})
+
 type FilePath = string
 
 interface State {
   readonly files: readonly FilePath[]
   readonly networkRegistrations: readonly string[]
+  readonly llmScripts: readonly LLMScript[]
+  readonly consumedLLMScripts: number
 }
 
+export class SimulationLLMError extends Schema.TaggedErrorClass<SimulationLLMError>()("SimulationLLMError", {
+  message: Schema.String,
+}) {}
+
 export interface Interface {
   readonly reset: () => Effect.Effect<void>
   readonly seedFilesystem: (input: typeof FilesystemSeedInput.Type) => Effect.Effect<{ files: string[] }, unknown>
   readonly registerNetwork: (input: typeof NetworkRegisterInput.Type) => Effect.Effect<{ registered: string }, unknown>
+  readonly enqueueLLM: (input: typeof LLMEnqueueInput.Type) => Effect.Effect<{ queued: number }>
+  readonly nextLLM: () => Effect.Effect<LLMScript, SimulationLLMError>
   readonly snapshot: () => Effect.Effect<{
     files: readonly string[]
     networkRegistrations: readonly string[]
+    llmQueued: number
+    llmConsumed: number
     network: SimulationNetwork.Snapshot
   }>
 }
@@ -72,11 +108,11 @@ export const layer = Layer.effect(
   Effect.gen(function* () {
     const fs = yield* AppFileSystem.Service
     const network = yield* SimulationNetwork.Service
-    const state = yield* Ref.make<State>({ files: [], networkRegistrations: [] })
+    const state = yield* Ref.make<State>({ files: [], networkRegistrations: [], llmScripts: [], consumedLLMScripts: 0 })
 
     const reset = Effect.fn("Simulation.reset")(function* () {
       yield* network.reset()
-      yield* Ref.set(state, { files: [], networkRegistrations: [] })
+      yield* Ref.set(state, { files: [], networkRegistrations: [], llmScripts: [], consumedLLMScripts: 0 })
     })
 
     const seedFilesystem = Effect.fn("Simulation.seedFilesystem")(function* (input: typeof FilesystemSeedInput.Type) {
@@ -112,12 +148,31 @@ export const layer = Layer.effect(
       return { registered: input.url }
     })
 
+    const enqueueLLM = Effect.fn("Simulation.enqueueLLM")(function* (input: typeof LLMEnqueueInput.Type) {
+      yield* Ref.update(state, (current) => ({ ...current, llmScripts: [...current.llmScripts, ...input.scripts] }))
+      return { queued: input.scripts.length }
+    })
+
+    const nextLLM = Effect.fn("Simulation.nextLLM")(function* () {
+      const current = yield* Ref.get(state)
+      const [script, ...rest] = current.llmScripts
+      if (!script) return yield* new SimulationLLMError({ message: "No LLM script queued" })
+      yield* Ref.set(state, { ...current, llmScripts: rest, consumedLLMScripts: current.consumedLLMScripts + 1 })
+      return script
+    })
+
     const snapshot = Effect.fn("Simulation.snapshot")(function* () {
       const current = yield* Ref.get(state)
-      return { ...current, network: yield* network.snapshot() }
+      return {
+        files: current.files,
+        networkRegistrations: current.networkRegistrations,
+        llmQueued: current.llmScripts.length,
+        llmConsumed: current.consumedLLMScripts,
+        network: yield* network.snapshot(),
+      }
     })
 
-    return Service.of({ reset, seedFilesystem, registerNetwork, snapshot })
+    return Service.of({ reset, seedFilesystem, registerNetwork, enqueueLLM, nextLLM, snapshot })
   }),
 )
 

packages/opencode/test/testing/simulation/service.test.ts

@@ -2,15 +2,18 @@ import { describe, expect } from "bun:test"
 import { AppFileSystem } from "@opencode-ai/core/filesystem"
 import { Effect, Layer } from "effect"
 import { HttpClient, HttpClientRequest } from "effect/unstable/http"
+import { Provider } from "../../../src/provider/provider"
 import { SimulationFileSystem } from "../../../src/testing/simulation/filesystem"
 import { SimulationNetwork } from "../../../src/testing/simulation/network"
+import { SimulationProvider } from "../../../src/testing/simulation/provider"
 import { Simulation } from "../../../src/testing/simulation/service"
 import { testEffect } from "../../lib/effect"
 
 const fsLayer = SimulationFileSystem.layer({ root: "/opencode" })
 const networkLayer = SimulationNetwork.layer({ allowLoopback: false })
 const simulationLayer = Simulation.layer.pipe(Layer.provide(fsLayer), Layer.provide(networkLayer))
-const it = testEffect(Layer.mergeAll(fsLayer, networkLayer, simulationLayer))
+const providerLayer = SimulationProvider.layer.pipe(Layer.provide(simulationLayer))
+const it = testEffect(Layer.mergeAll(fsLayer, networkLayer, simulationLayer, providerLayer))
 
 describe("Simulation", () => {
   it.effect("seeds files into the simulated filesystem", () =>
@@ -64,4 +67,85 @@ describe("Simulation", () => {
       expect(exit._tag).toBe("Failure")
     }),
   )
+
+  it.effect("queues and consumes LLM scripts", () =>
+    Effect.gen(function* () {
+      const simulation = yield* Simulation.Service
+
+      expect(
+        yield* simulation.enqueueLLM({
+          scripts: [{ steps: [[{ type: "text", content: "hello" }]], finish: "stop" }],
+        }),
+      ).toEqual({ queued: 1 })
+      expect((yield* simulation.snapshot()).llmQueued).toBe(1)
+      expect(yield* simulation.nextLLM()).toEqual({ steps: [[{ type: "text", content: "hello" }]], finish: "stop" })
+      expect((yield* simulation.snapshot()).llmConsumed).toBe(1)
+    }),
+  )
+
+  it.effect("simulation provider consumes queued text scripts", () =>
+    Effect.gen(function* () {
+      const simulation = yield* Simulation.Service
+      const provider = yield* Provider.Service
+      const model = yield* provider.defaultModel().pipe(Effect.flatMap((item) => provider.getModel(item.providerID, item.modelID)))
+      const language = yield* provider.getLanguage(model)
+
+      yield* simulation.enqueueLLM({ scripts: [{ steps: [[{ type: "text", content: "assistant text" }]] }] })
+
+      const result = yield* Effect.promise(() => language.doGenerate({ prompt: [], abortSignal: undefined }))
+      expect(result.content).toEqual([{ type: "text", text: "assistant text" }])
+      expect((yield* simulation.snapshot()).llmConsumed).toBe(1)
+    }),
+  )
+
+  it.effect("simulation provider streams queued script actions", () =>
+    Effect.gen(function* () {
+      const simulation = yield* Simulation.Service
+      const provider = yield* Provider.Service
+      const model = yield* provider.defaultModel().pipe(Effect.flatMap((item) => provider.getModel(item.providerID, item.modelID)))
+      const language = yield* provider.getLanguage(model)
+
+      yield* simulation.enqueueLLM({
+        scripts: [
+          {
+            steps: [
+              [
+                { type: "thinking", content: "thinking" },
+                { type: "text", content: "answer" },
+              ],
+            ],
+          },
+        ],
+      })
+
+      const result = yield* Effect.promise(() => language.doStream({ prompt: [], abortSignal: undefined }))
+      const reader = result.stream.getReader()
+      const parts: unknown[] = []
+      while (true) {
+        const next = yield* Effect.promise(() => reader.read())
+        if (next.done) break
+        parts.push(next.value)
+      }
+
+      expect(parts).toEqual([
+        { type: "stream-start", warnings: [] },
+        { type: "reasoning-start", id: "simulation-thinking-1" },
+        { type: "reasoning-delta", id: "simulation-thinking-1", delta: "thinking" },
+        { type: "reasoning-end", id: "simulation-thinking-1" },
+        { type: "text-start", id: "simulation-text-2" },
+        { type: "text-delta", id: "simulation-text-2", delta: "answer" },
+        { type: "text-end", id: "simulation-text-2" },
+        {
+          type: "finish",
+          finishReason: { unified: "stop", raw: undefined },
+          usage: {
+            inputTokens: { total: 0, noCache: 0, cacheRead: undefined, cacheWrite: undefined },
+            outputTokens: { total: "thinkinganswer".length, text: "thinkinganswer".length, reasoning: undefined },
+            raw: undefined,
+          },
+        },
+      ])
+      expect((yield* simulation.snapshot()).llmConsumed).toBe(1)
+    }),
+  )
 })