fix(opencode): support Luna with ChatGPT OAuth

packages/opencode/src/plugin/openai/codex.ts

@@ -6,15 +6,24 @@ import { setTimeout as sleep } from "node:timers/promises"
 import { createServer } from "http"
 import { OpenAIWebSocketPool } from "./ws-pool"
 import { OauthCallbackPage } from "@opencode-ai/core/oauth/page"
+import { isRecord } from "@/util/record"
 
 const CLIENT_ID = "app_EMoamEEZ73f0CkXaXp7hrann"
 const ISSUER = "https://auth.openai.com"
 const CODEX_API_ENDPOINT = "https://chatgpt.com/backend-api/codex/responses"
 const OAUTH_PORT = 1455
 const OAUTH_POLLING_SAFETY_MARGIN_MS = 3000
+const CODEX_COMPATIBILITY_VERSION = "0.144.0"
+const RESPONSES_LITE_MODEL = "gpt-5.6-luna"
 const ALLOWED_MODELS = new Set(["gpt-5.5", "gpt-5.3-codex-spark", "gpt-5.4", "gpt-5.4-mini"])
 const DISALLOWED_MODELS = new Set(["gpt-5.5-pro"])
 
+type ResponsesLiteRequest = Record<string, unknown> & {
+  input: unknown[]
+  tools?: unknown[]
+  instructions?: string
+}
+
 interface PkceCodes {
   verifier: string
   challenge: string
@@ -263,6 +272,7 @@ function waitForOAuthCallback(pkce: PkceCodes, state: string): Promise<TokenResp
 export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPluginOptions = {}): Promise<Hooks> {
   const issuer = options.issuer ?? ISSUER
   const codexApiEndpoint = options.codexApiEndpoint ?? CODEX_API_ENDPOINT
+  const codexSessionIDs = new Map<string, string>()
   let websocketFetchInstalled = false
   const websocketFetches: Array<ReturnType<typeof OpenAIWebSocketPool.createWebSocketFetch>> = []
 
@@ -273,7 +283,13 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
     },
     async event(input) {
       if (input.event.type !== "session.deleted") return
-      for (const websocketFetch of websocketFetches) websocketFetch.remove(input.event.properties.info.id)
+      const sessionID = input.event.properties.info.id
+      const codexSessionID = codexSessionIDs.get(sessionID)
+      for (const websocketFetch of websocketFetches) {
+        websocketFetch.remove(sessionID)
+        if (codexSessionID) websocketFetch.remove(codexSessionID)
+      }
+      codexSessionIDs.delete(sessionID)
     },
     provider: {
       id: "openai",
@@ -285,6 +301,7 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
             .filter(([, model]) => {
               if (ALLOWED_MODELS.has(model.api.id)) return true
               if (DISALLOWED_MODELS.has(model.api.id)) return false
+              if (model.api.id === "gpt-5.6") return false
               const match = model.api.id.match(/^gpt-(\d+\.\d+)/)
               return match ? parseFloat(match[1]) > 5.4 : false
             })
@@ -416,6 +433,9 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
 
             const requestInit = {
               ...init,
+              body: parsed.pathname.endsWith("/responses")
+                ? prepareResponsesLiteRequest(init?.body, headers, codexSessionIDs)
+                : init?.body,
               headers,
             }
             if (websocketFetch && parsed.pathname.endsWith("/responses")) return websocketFetch(url, requestInit)
@@ -560,3 +580,81 @@ export async function CodexAuthPlugin(input: PluginInput, options: CodexAuthPlug
     },
   }
 }
+
+function prepareResponsesLiteRequest(
+  body: BodyInit | null | undefined,
+  headers: Headers,
+  sessionIDs: Map<string, string>,
+) {
+  if (typeof body !== "string") return body
+  const request = parseResponsesLiteRequest(body)
+  if (!request) return body
+  const sourceSessionID = requireSessionID(headers)
+  const sessionID = sessionIDs.get(sourceSessionID) ?? Bun.randomUUIDv7()
+  sessionIDs.set(sourceSessionID, sessionID)
+
+  stripImageDetail(request.input)
+  request.input = [
+    { type: "additional_tools", role: "developer", tools: request.tools ?? [] },
+    ...(request.instructions
+      ? [
+          {
+            type: "message",
+            role: "developer",
+            content: [{ type: "input_text", text: request.instructions }],
+          },
+        ]
+      : []),
+    ...request.input,
+  ]
+  delete request.tools
+  delete request.instructions
+  request.tool_choice = "auto"
+  request.parallel_tool_calls = false
+  request.prompt_cache_key = sessionID
+  request.reasoning = {
+    ...(isRecord(request.reasoning) ? request.reasoning : {}),
+    context: "all_turns",
+  }
+
+  headers.set("session-id", sessionID)
+  headers.set("x-session-affinity", sessionID)
+  headers.set("version", CODEX_COMPATIBILITY_VERSION)
+  headers.set(OpenAIWebSocketPool.RESPONSES_LITE_HEADER, "true")
+  headers.delete("content-length")
+  return JSON.stringify(request)
+}
+
+function parseResponsesLiteRequest(body: string): ResponsesLiteRequest | undefined {
+  const request: unknown = JSON.parse(body)
+  if (!isRecord(request) || request.model !== RESPONSES_LITE_MODEL) return undefined
+  if (!Array.isArray(request.input)) throw new Error("Responses Lite requires an input array")
+  if (request.tools !== undefined && !Array.isArray(request.tools)) {
+    throw new Error("Responses Lite requires a tools array")
+  }
+  if (request.instructions !== undefined && typeof request.instructions !== "string") {
+    throw new Error("Responses Lite requires string instructions")
+  }
+  return {
+    ...request,
+    input: request.input,
+    tools: request.tools,
+    instructions: request.instructions,
+  }
+}
+
+function requireSessionID(headers: Headers) {
+  const sessionID = headers.get("session-id")
+  if (!sessionID) throw new Error("Responses Lite requires a session-id header")
+  return sessionID
+}
+
+function stripImageDetail(value: unknown) {
+  if (Array.isArray(value)) {
+    value.forEach(stripImageDetail)
+    return
+  }
+  if (!isRecord(value)) return
+  if (value.type === "input_image") delete value.detail
+  Object.values(value).forEach(stripImageDetail)
+}

packages/opencode/src/plugin/openai/ws-pool.ts

@@ -4,6 +4,8 @@ import { isRecord } from "@/util/record"
 import { OpenAIWebSocket } from "./ws"
 
 export const TITLE_HEADER = "x-opencode-title"
+export const RESPONSES_LITE_HEADER = "x-openai-internal-codex-responses-lite"
+const RESPONSES_LITE_CLIENT_METADATA = "ws_request_header_x_openai_internal_codex_responses_lite"
 
 export interface CreateWebSocketFetchOptions {
   httpFetch?: typeof globalThis.fetch
@@ -98,7 +100,7 @@ export function createWebSocketFetch(options?: CreateWebSocketFetchOptions) {
       })
       const response = OpenAIWebSocket.streamResponsesWebSocket({
         socket: entry.socket,
-        body,
+        body: withResponsesLiteMetadata(body, internalHeaders),
         idleTimeout,
         signal: init?.signal ?? undefined,
         onFirstEvent: (error) => resolveFirstEvent(error ?? true),
@@ -194,6 +196,17 @@ export function createWebSocketFetch(options?: CreateWebSocketFetchOptions) {
   return Object.assign(websocketFetch, { close, remove })
 }
 
+function withResponsesLiteMetadata(body: Record<string, unknown>, headers: Record<string, string>) {
+  if (headers[RESPONSES_LITE_HEADER] !== "true") return body
+  return {
+    ...body,
+    client_metadata: {
+      ...(isRecord(body.client_metadata) ? body.client_metadata : {}),
+      [RESPONSES_LITE_CLIENT_METADATA]: "true",
+    },
+  }
+}
+
 function connectionLimitError(event: Record<string, unknown>) {
   if (event.type !== "error" || !isRecord(event.error) || event.error.code !== CONNECTION_LIMIT_REACHED_CODE) return
   return new Error(typeof event.error.message === "string" ? event.error.message : CONNECTION_LIMIT_REACHED_CODE)

packages/opencode/test/plugin/codex.test.ts

@@ -152,13 +152,19 @@ describe("plugin.codex", () => {
   test("uses Codex context limits for OAuth GPT models", async () => {
     const hooks = await CodexAuthPlugin({} as never)
     const limit = { context: 1_050_000, input: 922_000, output: 128_000 }
+    const ids = [
+      "gpt-5.4",
+      "gpt-5.5",
+      "gpt-5.6",
+      "gpt-5.6-luna",
+      "gpt-5.6-luna-pro",
+      "gpt-5.6-sol",
+      "gpt-5.6-sol-pro",
+      "gpt-5.6-terra",
+      "gpt-5.6-terra-pro",
+    ]
     const provider = {
-      models: Object.fromEntries(
-        ["gpt-5.4", "gpt-5.5", "gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna"].map((id) => [
-          id,
-          { id, api: { id }, limit, cost: {} },
-        ]),
-      ),
+      models: Object.fromEntries(ids.map((id) => [id, { id, api: { id }, limit, cost: {} }])),
     }
 
     const models = await hooks.provider!.models!(provider as never, { auth: { type: "oauth" } } as never)
@@ -168,11 +174,155 @@ describe("plugin.codex", () => {
     expect(models["gpt-5.6-sol"]?.limit).toEqual({ context: 500_000, input: 372_000, output: 128_000 })
     expect(models["gpt-5.6-terra"]?.limit).toEqual({ context: 500_000, input: 372_000, output: 128_000 })
     expect(models["gpt-5.6-luna"]?.limit).toEqual({ context: 500_000, input: 372_000, output: 128_000 })
+    expect(models["gpt-5.6"]).toBeUndefined()
+    expect(Object.keys(models)).toEqual(ids.filter((id) => id !== "gpt-5.6"))
     expect(await hooks.provider!.models!(provider as never, { auth: { type: "api" } } as never)).toBe(
       provider.models as never,
     )
   })
 
+  test("applies Responses Lite only to exact Luna requests and keeps its session UUID stable", async () => {
+    const requests: Array<{ body: Record<string, unknown>; headers: Headers }> = []
+    using server = Bun.serve({
+      port: 0,
+      async fetch(request) {
+        requests.push({ body: await request.json(), headers: request.headers })
+        return Response.json({})
+      },
+    })
+    const hooks = await CodexAuthPlugin({} as never, {
+      codexApiEndpoint: new URL("/backend-api/codex/responses", server.url).toString(),
+    })
+    const loaded = await hooks.auth!.loader!(
+      async () => ({
+        type: "oauth",
+        refresh: "refresh",
+        access: "access",
+        expires: Date.now() + 60_000,
+      }),
+      {} as never,
+    )
+    const first = {
+      model: "gpt-5.6-luna",
+      stream: true,
+      input: [
+        {
+          type: "message",
+          role: "user",
+          content: [{ type: "input_image", image_url: "data:image/png;base64,a", detail: "high" }],
+        },
+      ],
+      tools: [],
+      instructions: "Be precise",
+      tool_choice: "required",
+      parallel_tool_calls: true,
+      reasoning: { effort: "high" },
+    }
+    const continuation = {
+      model: "gpt-5.6-luna",
+      stream: true,
+      input: [{ type: "message", role: "user", content: [{ type: "input_text", text: "Continue" }] }],
+      tools: [{ type: "function", name: "search" }],
+      instructions: "",
+    }
+    const headers = {
+      "session-id": "opencode-session",
+      originator: "opencode",
+      "user-agent": "opencode/test",
+    }
+
+    await loaded.fetch!("https://api.openai.com/v1/responses", {
+      method: "POST",
+      headers,
+      body: JSON.stringify(first),
+    })
+    await loaded.fetch!("https://api.openai.com/v1/responses", {
+      method: "POST",
+      headers,
+      body: JSON.stringify(continuation),
+    })
+
+    expect(requests).toHaveLength(2)
+    const sessionID = requests[0].headers.get("session-id")
+    expect(sessionID).toMatch(/^[0-9a-f]{8}-[0-9a-f]{4}-7[0-9a-f]{3}-[89ab][0-9a-f]{3}-[0-9a-f]{12}$/)
+    expect(requests[1].headers.get("session-id")).toBe(sessionID)
+    for (const request of requests) {
+      expect(request.headers.get("x-session-affinity")).toBe(sessionID)
+      expect(request.headers.get("version")).toBe("0.144.0")
+      expect(request.headers.get("x-openai-internal-codex-responses-lite")).toBe("true")
+      expect(request.headers.get("originator")).toBe("opencode")
+      expect(request.headers.get("user-agent")).toBe("opencode/test")
+      expect(request.body.prompt_cache_key).toBe(sessionID)
+      expect(request.body.tool_choice).toBe("auto")
+      expect(request.body.parallel_tool_calls).toBe(false)
+      expect(request.body.tools).toBeUndefined()
+      expect(request.body.instructions).toBeUndefined()
+    }
+    expect(requests[0].body.reasoning).toEqual({ effort: "high", context: "all_turns" })
+    expect(requests[1].body.reasoning).toEqual({ context: "all_turns" })
+    expect(requests[0].body.input).toEqual([
+      { type: "additional_tools", role: "developer", tools: [] },
+      {
+        type: "message",
+        role: "developer",
+        content: [{ type: "input_text", text: "Be precise" }],
+      },
+      {
+        type: "message",
+        role: "user",
+        content: [{ type: "input_image", image_url: "data:image/png;base64,a" }],
+      },
+    ])
+    expect(requests[1].body.input).toEqual([
+      { type: "additional_tools", role: "developer", tools: [{ type: "function", name: "search" }] },
+      { type: "message", role: "user", content: [{ type: "input_text", text: "Continue" }] },
+    ])
+  })
+
+  test("leaves Sol, Terra, and other Responses requests unchanged", async () => {
+    const requests: Array<{ body: string; headers: Headers }> = []
+    using server = Bun.serve({
+      port: 0,
+      async fetch(request) {
+        requests.push({ body: await request.text(), headers: request.headers })
+        return Response.json({})
+      },
+    })
+    const hooks = await CodexAuthPlugin({} as never, {
+      codexApiEndpoint: new URL("/backend-api/codex/responses", server.url).toString(),
+    })
+    const loaded = await hooks.auth!.loader!(
+      async () => ({
+        type: "oauth",
+        refresh: "refresh",
+        access: "access",
+        expires: Date.now() + 60_000,
+      }),
+      {} as never,
+    )
+    const bodies = ["gpt-5.6-sol", "gpt-5.6-terra", "gpt-5.6-luna-pro", "gpt-5.5"].map((model) =>
+      JSON.stringify({ model, stream: true, input: [], tools: [], instructions: "Keep me" }),
+    )
+
+    await Promise.all(
+      bodies.map((body) =>
+        loaded.fetch!("https://api.openai.com/v1/responses", {
+          method: "POST",
+          headers: { "session-id": "opencode-session" },
+          body,
+        }),
+      ),
+    )
+
+    expect(requests.map((request) => request.body).sort()).toEqual(bodies.sort())
+    for (const request of requests) {
+      expect(request.headers.get("session-id")).toBe("opencode-session")
+      expect(request.headers.get("x-session-affinity")).toBeNull()
+      expect(request.headers.get("version")).toBeNull()
+      expect(request.headers.get("x-openai-internal-codex-responses-lite")).toBeNull()
+    }
+  })
+
   test("deduplicates concurrent Codex token refreshes", async () => {
     let auth = {
       type: "oauth" as const,

packages/opencode/test/plugin/openai-ws.test.ts

@@ -166,6 +166,31 @@ describe("plugin.openai.ws-pool", () => {
     fetch.close()
   })
 
+  test("adds the Responses Lite marker to websocket client metadata", async () => {
+    let requestBody: Record<string, unknown> | undefined
+    await using server = await createWebSocketServer((socket) => {
+      socket.once("message", (data) => {
+        requestBody = JSON.parse(data.toString())
+        socket.send(JSON.stringify({ type: "response.completed", response: { id: "resp_luna" } }))
+      })
+    })
+    const fetch = OpenAIWebSocketPool.createWebSocketFetch({ url: server.url })
+
+    const response = await fetch(
+      server.url,
+      streamRequest({ [OpenAIWebSocketPool.RESPONSES_LITE_HEADER]: "true" }, undefined, {
+        client_metadata: { existing: "value" },
+      }),
+    )
+
+    expect(await response.text()).toContain("data: [DONE]")
+    expect(requestBody?.client_metadata).toEqual({
+      existing: "value",
+      ws_request_header_x_openai_internal_codex_responses_lite: "true",
+    })
+    fetch.close()
+  })
+
   test("rotates a socket that exceeds max connection age", async () => {
     let connections = 0
     await using server = await createWebSocketServer((socket) => {
@@ -773,7 +798,11 @@ describe("plugin.openai.ws-pool", () => {
   })
 })
 
-function streamRequest(headers?: Record<string, string>, signal?: AbortSignal): RequestInit {
+function streamRequest(
+  headers?: Record<string, string>,
+  signal?: AbortSignal,
+  body?: Record<string, unknown>,
+): RequestInit {
   return {
     method: "POST",
     headers: {
@@ -781,7 +810,7 @@ function streamRequest(headers?: Record<string, string>, signal?: AbortSignal):
       authorization: "Bearer test",
       ...headers,
     },
-    body: JSON.stringify({ stream: true, input: "hi" }),
+    body: JSON.stringify({ stream: true, input: "hi", ...body }),
     signal,
   }
 }