refactor(flags): migrate output token max to runtime flags (#27680)

Author: nexxeln

packages/core/src/flag/flag.ts

@@ -5,13 +5,6 @@ function truthy(key: string) {
   return value === "true" || value === "1"
 }
 
-function number(key: string) {
-  const value = process.env[key]
-  if (!value) return undefined
-  const parsed = Number(value)
-  return Number.isInteger(parsed) && parsed > 0 ? parsed : undefined
-}
-
 const OPENCODE_EXPERIMENTAL = truthy("OPENCODE_EXPERIMENTAL")
 const OPENCODE_DISABLE_CLAUDE_CODE = truthy("OPENCODE_DISABLE_CLAUDE_CODE")
 const copy = process.env["OPENCODE_EXPERIMENTAL_DISABLE_COPY_ON_SELECT"]
@@ -54,7 +47,6 @@ export const Flag = {
   OPENCODE_EXPERIMENTAL_DISABLE_COPY_ON_SELECT:
     copy === undefined ? process.platform === "win32" : truthy("OPENCODE_EXPERIMENTAL_DISABLE_COPY_ON_SELECT"),
   OPENCODE_ENABLE_EXA: truthy("OPENCODE_ENABLE_EXA") || OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_EXA"),
-  OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX: number("OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX"),
   OPENCODE_EXPERIMENTAL_LSP_TOOL: OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_LSP_TOOL"),
   OPENCODE_EXPERIMENTAL_PLAN_MODE: OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_PLAN_MODE"),
   OPENCODE_EXPERIMENTAL_SCOUT: OPENCODE_EXPERIMENTAL || truthy("OPENCODE_EXPERIMENTAL_SCOUT"),

packages/opencode/src/effect/runtime-flags.ts

@@ -41,6 +41,7 @@ export class Service extends ConfigService.Service<Service>()("@opencode/Runtime
   experimentalEventSystem: enabledByExperimental("OPENCODE_EXPERIMENTAL_EVENT_SYSTEM"),
   experimentalWorkspaces: enabledByExperimental("OPENCODE_EXPERIMENTAL_WORKSPACES"),
   experimentalIconDiscovery: enabledByExperimental("OPENCODE_EXPERIMENTAL_ICON_DISCOVERY"),
+  outputTokenMax: positiveInteger("OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX"),
   bashDefaultTimeoutMs: positiveInteger("OPENCODE_EXPERIMENTAL_BASH_DEFAULT_TIMEOUT_MS"),
   client: Config.string("OPENCODE_CLIENT").pipe(Config.withDefault("cli")),
 }) {}

packages/opencode/src/provider/transform.ts

@@ -4,7 +4,6 @@ import type { JSONSchema7 } from "@ai-sdk/provider"
 import type * as Provider from "./provider"
 import type * as ModelsDev from "@opencode-ai/core/models"
 import { iife } from "@/util/iife"
-import { Flag } from "@opencode-ai/core/flag/flag"
 
 type Modality = NonNullable<ModelsDev.Model["modalities"]>["input"][number]
 
@@ -16,7 +15,7 @@ function mimeToModality(mime: string): Modality | undefined {
   return undefined
 }
 
-export const OUTPUT_TOKEN_MAX = Flag.OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX || 32_000
+export const OUTPUT_TOKEN_MAX = 32_000
 
 export function sanitizeSurrogates(content: string) {
   return content.replace(/[\uD800-\uDBFF](?![\uDC00-\uDFFF])|(?<![\uD800-\uDBFF])[\uDC00-\uDFFF]/g, "\uFFFD")
@@ -1251,8 +1250,8 @@ export function providerOptions(model: Provider.Model, options: { [x: string]: a
   return { [key]: options }
 }
 
-export function maxOutputTokens(model: Provider.Model): number {
-  return Math.min(model.limit.output, OUTPUT_TOKEN_MAX) || OUTPUT_TOKEN_MAX
+export function maxOutputTokens(model: Provider.Model, outputTokenMax = OUTPUT_TOKEN_MAX): number {
+  return Math.min(model.limit.output, outputTokenMax) || outputTokenMax
 }
 
 export function schema(model: Provider.Model, schema: JSONSchema7): JSONSchema7 {

packages/opencode/src/session/compaction.ts

@@ -228,7 +228,12 @@ export const layer = Layer.effect(
       tokens: MessageV2.Assistant["tokens"]
       model: Provider.Model
     }) {
-      return overflow({ cfg: yield* config.get(), tokens: input.tokens, model: input.model })
+      return overflow({
+        cfg: yield* config.get(),
+        tokens: input.tokens,
+        model: input.model,
+        outputTokenMax: flags.outputTokenMax,
+      })
     })
 
     const estimate = Effect.fn("SessionCompaction.estimate")(function* (input: {

packages/opencode/src/session/llm.ts

@@ -173,7 +173,7 @@ const live: Layer.Layer<
             : undefined,
           topP: input.agent.topP ?? ProviderTransform.topP(input.model),
           topK: ProviderTransform.topK(input.model),
-          maxOutputTokens: ProviderTransform.maxOutputTokens(input.model),
+          maxOutputTokens: ProviderTransform.maxOutputTokens(input.model, flags.outputTokenMax),
           options,
         },
       )

packages/opencode/src/session/overflow.ts

@@ -5,18 +5,24 @@ import type { MessageV2 } from "./message-v2"
 
 const COMPACTION_BUFFER = 20_000
 
-export function usable(input: { cfg: Config.Info; model: Provider.Model }) {
+export function usable(input: { cfg: Config.Info; model: Provider.Model; outputTokenMax?: number }) {
   const context = input.model.limit.context
   if (context === 0) return 0
 
   const reserved =
-    input.cfg.compaction?.reserved ?? Math.min(COMPACTION_BUFFER, ProviderTransform.maxOutputTokens(input.model))
+    input.cfg.compaction?.reserved ??
+    Math.min(COMPACTION_BUFFER, ProviderTransform.maxOutputTokens(input.model, input.outputTokenMax))
   return input.model.limit.input
     ? Math.max(0, input.model.limit.input - reserved)
-    : Math.max(0, context - ProviderTransform.maxOutputTokens(input.model))
+    : Math.max(0, context - ProviderTransform.maxOutputTokens(input.model, input.outputTokenMax))
 }
 
-export function isOverflow(input: { cfg: Config.Info; tokens: MessageV2.Assistant["tokens"]; model: Provider.Model }) {
+export function isOverflow(input: {
+  cfg: Config.Info
+  tokens: MessageV2.Assistant["tokens"]
+  model: Provider.Model
+  outputTokenMax?: number
+}) {
   if (input.cfg.compaction?.auto === false) return false
   if (input.model.limit.context === 0) return false
 

packages/opencode/test/effect/runtime-flags.test.ts

@@ -88,6 +88,7 @@ describe("RuntimeFlags", () => {
       expect(flags.enableExa).toBe(false)
       expect(flags.experimentalIconDiscovery).toBe(false)
       expect(flags.experimentalOxfmt).toBe(false)
+      expect(flags.outputTokenMax).toBeUndefined()
       expect(flags.bashDefaultTimeoutMs).toBe(1_000)
       expect(flags.enableExperimentalModels).toBe(false)
       expect(flags.client).toBe("cli")
@@ -183,6 +184,35 @@ describe("RuntimeFlags", () => {
     )
   }
 
+  for (const input of [
+    { name: "absent", config: {}, expected: undefined },
+    {
+      name: "valid positive integer",
+      config: { OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX: "1234" },
+      expected: 1234,
+    },
+    {
+      name: "invalid string",
+      config: { OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX: "nope" },
+      expected: undefined,
+    },
+    { name: "zero", config: { OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX: "0" }, expected: undefined },
+    { name: "negative", config: { OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX: "-1" }, expected: undefined },
+    {
+      name: "non-integer",
+      config: { OPENCODE_EXPERIMENTAL_OUTPUT_TOKEN_MAX: "1.5" },
+      expected: undefined,
+    },
+  ]) {
+    it.effect(`parses outputTokenMax from config: ${input.name}`, () =>
+      Effect.gen(function* () {
+        const flags = yield* readFlags.pipe(Effect.provide(fromConfig(input.config)))
+
+        expect(flags.outputTokenMax).toBe(input.expected)
+      }),
+    )
+  }
+
   it.effect("layer ignores the active ConfigProvider for omitted test overrides", () =>
     Effect.gen(function* () {
       const flags = yield* readFlags.pipe(
@@ -209,6 +239,7 @@ describe("RuntimeFlags", () => {
       expect(flags.enableExa).toBe(false)
       expect(flags.experimentalIconDiscovery).toBe(false)
       expect(flags.experimentalOxfmt).toBe(false)
+      expect(flags.outputTokenMax).toBeUndefined()
       expect(flags.bashDefaultTimeoutMs).toBeUndefined()
       expect(flags.client).toBe("cli")
     }),