@4yi/cli 0.1.9 → 0.1.11

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@4yi/cli",
-  "version": "0.1.9",
+  "version": "0.1.11",
   "description": "4YI command-line launcher for OAuth login and OpenCode runtime",
   "type": "module",
   "bin": {

package/src/connect.mjs

@@ -311,6 +311,16 @@ function buildCodexCatalog(models, template) {
       description: `${planModel.display_name || planModel.id} via 4YI Gateway with OpenAI Responses tool support.`,
       supported_in_api: true,
       use_responses_lite: false,
+      // The template is whatever `codex debug models --bundled` returns on this
+      // machine, and that is not a stable schema: 0.147 carries this field,
+      // 0.150 omits it, and connect installs the latest codex -- so it upgrades
+      // its own template source into the version that drops it. Codex.app
+      // requires it and will not start without one, so it is set here rather
+      // than inherited. The endpoint is authoritative when it answers; `true`
+      // is the fallback because every model this catalog can describe is served
+      // through the Responses ingress, which requires `parallel_tool_calls` on
+      // each request.
+      supports_parallel_tool_calls: planModel.supports_parallel_tool_calls ?? true,
       tool_mode: undefined,
     })).map((model) => {
       delete model.tool_mode;
@@ -320,9 +330,16 @@ function buildCodexCatalog(models, template) {
 }
 
 async function checkCodex(session, codexBaseUrl, model) {
-  await requestJson(codexBaseUrl.replace(/\/+$/, ""), "/responses", {
+  const response = await fetch(`${codexBaseUrl.replace(/\/+$/, "")}/responses`, {
     method: "POST",
-    token: session.token,
+    headers: {
+      "content-type": "application/json",
+      authorization: `Bearer ${session.token}`,
+      // The Responses ingress only accepts the native Codex request shape.
+      // Keep this synthetic compatibility check aligned with the real client
+      // so it exercises the same authorization and routing path.
+      originator: "codex_cli_rs",
+    },
     body: JSON.stringify({
       model,
       instructions: "Reply with pong.",
@@ -331,16 +348,39 @@ async function checkCodex(session, codexBaseUrl, model) {
       tool_choice: "auto",
       parallel_tool_calls: true,
       store: false,
-      stream: false,
+      // Safe Responses ingress deliberately requires streaming requests. A
+      // JSON/non-streaming probe is rejected before routing, even though the
+      // same credential and model work in Codex itself.
+      stream: true,
       include: [],
+      client_metadata: {
+        session_id: "4yi-cli-preflight",
+        thread_id: "4yi-cli-preflight",
+      },
     }),
   });
+
+  if (!response.ok) {
+    const text = await response.text();
+    let body;
+    try { body = text ? JSON.parse(text) : null; } catch { body = null; }
+    const message = body?.error?.message || body?.error || text || `HTTP ${response.status}`;
+    const error = new Error(typeof message === "string" ? message : JSON.stringify(message));
+    error.status = response.status;
+    error.body = body;
+    throw error;
+  }
+
+  // A successful native probe is an SSE stream. Compatibility is established
+  // once the response starts; do not consume a model response just for setup.
+  if (response.body) await response.body.cancel();
 }
 
 async function connectCodex({ session, home, codexHome, codexBaseUrl, stdout, skipCheck = false, tooling = {} }) {
   const paths = connectionPaths({ home, codexHome });
   const { models, defaultModel } = await loadCodexModels(session);
   if (!skipCheck) await checkCodex(session, codexBaseUrl, defaultModel);
+  ensureDir(path.dirname(paths.codexConfig));
   const template = parseBundledCatalog(tooling);
   const catalog = buildCodexCatalog(models, template);
   atomicWrite(paths.codexCatalog, `${JSON.stringify(catalog, null, 2)}\n`);
@@ -437,6 +477,7 @@ export function restoreConnection({ target = "all", home = os.homedir(), stdout
 
 export const __testing = {
   buildCodexCatalog,
+  checkCodex,
   commandAvailable,
   desktopAppCandidates,
   ensureToolCli,

package/src/opencode.mjs

@@ -10,6 +10,7 @@ const DEFAULT_CONTEXT_LIMIT = 200000;
 const DEFAULT_OUTPUT_LIMIT = 8192;
 
 function isClaudeModel(model) {
+  if (model?.family === "claude") return true;
   const id = model?.id || "";
   const name = model?.display_name || "";
   return /claude/i.test(id) || /claude/i.test(name);
@@ -23,14 +24,23 @@ function modelOutputLimit(model) {
   return model.output_limit || model.max_output_tokens || model.max_tokens || DEFAULT_OUTPUT_LIMIT;
 }
 
+function modelPickerName(model) {
+  const displayName = model.display_name || model.id;
+  if (model.family === "claude") {
+    return `Claude · ${displayName.replace(/^(?:anthropic[ ._-]*)?claude[ ._-]*/i, "")}`;
+  }
+  if (model.family === "codex") return `Codex · ${displayName}`;
+  return displayName;
+}
+
 export function buildOpenCodeConfig({ modelConfig, preferredModel = null, tokenEnv = "FOURYI_CLI_TOKEN", orgEnv = "FOURYI_ORG_ID" }) {
-  // Register EVERY model the server returned so OpenCode's native switcher
-  // (`Tab` / `/models`) can move between them. Claude stays the default.
+  // Register every server-approved coding model so OpenCode's native switcher
+  // (`Tab` / `/models`) can move between Claude and Codex. Claude stays the default.
   const list = modelConfig.models || [];
   const models = {};
   for (const model of list) {
     const entry = {
-      name: model.display_name || model.id,
+      name: modelPickerName(model),
       limit: {
         context: modelContextLimit(model),
         output: modelOutputLimit(model),
@@ -111,7 +121,7 @@ export function ensureOpenCodeRuntime({ home = os.homedir(), stdout = console.lo
 
 export async function runCode({ session, home = os.homedir(), argv = [], stdout = console.log, preferredModel = null } = {}) {
   if (!session?.token) throw new Error("Not signed in. Run `4yi login`.");
-  const modelConfig = await requestJson(session.baseUrl, "/api/cli/models", { token: session.token });
+  const modelConfig = await requestJson(session.baseUrl, "/api/cli/models?runtime=opencode", { token: session.token });
   const list = modelConfig.models || [];
   if (list.length === 0) throw new Error("No chat models available for this organization.");