Stop forcing fast Codex service tier

Author: xuio

README.md

@@ -11,6 +11,7 @@ The plugin lets Claude Code launch one Codex agent or several Codex agents in pa
 - Codex binary: prefers `/Applications/Codex.app/Contents/Resources/codex` when the Codex desktop app is installed, then falls back to configured overrides and `codex` on `PATH`.
 - Sandbox: `read-only`.
 - Approvals: non-interactive `approval_policy="never"`.
+- Service tier: omitted by default so Codex uses its normal account/default service tier. Pass `service_tier` only when you explicitly want one.
 - Transport: stdio MCP, launched by Claude Code for the active session. No daemon is required.
 - Prompt delivery: stdin, not command-line arguments.
 

dist/index.js

@@ -21450,7 +21450,6 @@ function buildCodexExecArgs(options, outputPath, env = process.env) {
   const model = resolveRequestedModel(options, env);
   const reasoningEffort = options.reasoningEffort ?? defaultReasoningEffort(env);
   const sandbox = options.sandbox ?? "read-only";
-  const serviceTier = options.serviceTier ?? "fast";
   const ephemeral = options.ephemeral ?? true;
   const args = [
     "exec",
@@ -21463,8 +21462,6 @@ function buildCodexExecArgs(options, outputPath, env = process.env) {
     `approval_policy=${tomlString("never")}`,
     "-c",
     `model_reasoning_effort=${tomlString(reasoningEffort)}`,
-    "-c",
-    `service_tier=${tomlString(serviceTier)}`,
     "--output-last-message",
     outputPath
   ];
@@ -21480,6 +21477,9 @@ function buildCodexExecArgs(options, outputPath, env = process.env) {
   if (options.reasoningSummary) {
     args.push("-c", `model_reasoning_summary=${tomlString(options.reasoningSummary)}`);
   }
+  if (options.serviceTier) {
+    args.push("-c", `service_tier=${tomlString(options.serviceTier)}`);
+  }
   if (options.subagentRuntime?.maxThreads !== void 0) {
     args.push("-c", `agents.max_threads=${options.subagentRuntime.maxThreads}`);
   }
@@ -21622,7 +21622,7 @@ async function runAgent(options) {
       modelPreset: options.modelPreset,
       reasoningEffort: options.reasoningEffort ?? defaultReasoningEffort(childEnv),
       sandbox: options.sandbox ?? "read-only",
-      serviceTier: options.serviceTier ?? "fast",
+      serviceTier: options.serviceTier,
       exitCode,
       signal,
       finalMessage: final.text,
@@ -21717,8 +21717,9 @@ var usageGuide = [
   "Default operating rules:",
   "- Keep sandbox read-only unless the user explicitly asks for a different sandbox.",
   "- Approvals are non-interactive; do not expect Codex to ask permission.",
-  '- Prefer model_preset "spark" for fast focused checks, small reviews, UI iteration, and responsive sidecar analysis.',
+  '- Prefer model_preset "spark" for responsive focused checks, small reviews, UI iteration, and sidecar analysis.',
   '- Use reasoning_effort "medium" by default, "low" for simple checks, and "high" or "xhigh" only for difficult analysis.',
+  "- Do not set service_tier by default. Let Codex use its normal account/default service tier unless the user explicitly asks for a service tier.",
   "- Pass project_dir whenever Claude knows the active project directory so Codex works in the same tree as Claude Code.",
   "- Ask Codex for concise results with file paths, line references, and actionable findings when reviewing code.",
   "",
@@ -21770,13 +21771,15 @@ var commonInputSchema = {
     "Exact Codex model, for example gpt-5.3-codex. Omit to use model_preset, the plugin default, or Codex default."
   ),
   model_preset: modelPresetSchema.optional().describe(
-    "Convenience model preset. Use `spark` for fast Codex Spark work; it maps to gpt-5.3-codex-spark."
+    "Convenience model preset. Use `spark` for responsive Codex Spark work; it maps to gpt-5.3-codex-spark."
   ),
   reasoning_effort: reasoningEffortSchema.optional().describe(
     "Codex model reasoning effort. Prefer medium by default, low for simple checks, high/xhigh only for difficult analysis."
   ),
   sandbox: sandboxModeSchema.default("read-only").describe("Codex sandbox mode. Keep read-only unless the user explicitly asks otherwise."),
-  service_tier: serviceTierSchema.default("fast").describe("Codex service tier. Defaults to fast for responsiveness."),
+  service_tier: serviceTierSchema.optional().describe(
+    "Optional Codex service tier. Omit by default; only set this when the user explicitly asks for a service tier."
+  ),
   model_verbosity: modelVerbositySchema.optional().describe("Optional GPT-5 model verbosity override."),
   reasoning_summary: reasoningSummarySchema.optional().describe("Optional Codex reasoning summary setting."),
   cwd: external_exports.string().trim().min(1).optional().describe("Compatibility alias for project_dir."),
@@ -21902,7 +21905,7 @@ server.registerTool(
   "run_agent",
   {
     title: "Run one Codex agent",
-    description: "Launch one OpenAI Codex agent via codex exec. Use automatically when the user asks Claude to use Codex, ask Codex, get a Codex second opinion, run a Codex subagent, use Codex Spark, or delegate one read-only analysis task. Defaults to the Codex desktop app binary when installed, read-only sandbox, fast service tier, and non-interactive approvals.",
+    description: "Launch one OpenAI Codex agent via codex exec. Use automatically when the user asks Claude to use Codex, ask Codex, get a Codex second opinion, run a Codex subagent, use Codex Spark, or delegate one read-only analysis task. Defaults to the Codex desktop app binary when installed, read-only sandbox, Codex's normal service tier, and non-interactive approvals.",
     inputSchema: {
       prompt: external_exports.string().min(1).describe(
         "Concrete instructions for the Codex agent. Include scope, read-only expectation, desired output shape, and file/line reference requirements when reviewing code."
@@ -22046,7 +22049,7 @@ server.registerTool(
         defaultModel: defaultModel(),
         defaultReasoningEffort: defaultReasoningEffort(),
         defaultSandbox: "read-only",
-        defaultServiceTier: "fast",
+        defaultServiceTier: "codex-default",
         modelPresets: {
           codex: "gpt-5.3-codex",
           spark: "gpt-5.3-codex-spark"

skills/codex-subagents/SKILL.md

@@ -24,7 +24,9 @@ When Claude wants Codex to work in the same repository or folder as the active C
 
 Prefer `reasoning_effort: "medium"` for exploration and `high` or `xhigh` only when the task is complex enough to justify the extra latency and token usage.
 
-Use `model_preset: "spark"` for fast, focused work such as UI iteration, narrow exploration, small reviews, and quick sidecar checks.
+Use `model_preset: "spark"` for responsive, focused work such as UI iteration, narrow exploration, small reviews, and quick sidecar checks.
+
+Do not set `service_tier` by default. Let Codex use its normal account/default service tier unless the user explicitly asks for a service tier.
 
 Use `codex_status` only when diagnosing installation or binary resolution, or after a failed Codex tool call. Normal delegation should start with `run_agent` or `run_agents`.
 

src/index.ts

@@ -31,8 +31,9 @@ const usageGuide = [
   "Default operating rules:",
   "- Keep sandbox read-only unless the user explicitly asks for a different sandbox.",
   "- Approvals are non-interactive; do not expect Codex to ask permission.",
-  "- Prefer model_preset \"spark\" for fast focused checks, small reviews, UI iteration, and responsive sidecar analysis.",
+  "- Prefer model_preset \"spark\" for responsive focused checks, small reviews, UI iteration, and sidecar analysis.",
   "- Use reasoning_effort \"medium\" by default, \"low\" for simple checks, and \"high\" or \"xhigh\" only for difficult analysis.",
+  "- Do not set service_tier by default. Let Codex use its normal account/default service tier unless the user explicitly asks for a service tier.",
   "- Pass project_dir whenever Claude knows the active project directory so Codex works in the same tree as Claude Code.",
   "- Ask Codex for concise results with file paths, line references, and actionable findings when reviewing code.",
   "",
@@ -134,7 +135,7 @@ const commonInputSchema = {
   model_preset: modelPresetSchema
     .optional()
     .describe(
-      "Convenience model preset. Use `spark` for fast Codex Spark work; it maps to gpt-5.3-codex-spark.",
+      "Convenience model preset. Use `spark` for responsive Codex Spark work; it maps to gpt-5.3-codex-spark.",
     ),
   reasoning_effort: reasoningEffortSchema
     .optional()
@@ -145,8 +146,10 @@ const commonInputSchema = {
     .default("read-only")
     .describe("Codex sandbox mode. Keep read-only unless the user explicitly asks otherwise."),
   service_tier: serviceTierSchema
-    .default("fast")
-    .describe("Codex service tier. Defaults to fast for responsiveness."),
+    .optional()
+    .describe(
+      "Optional Codex service tier. Omit by default; only set this when the user explicitly asks for a service tier.",
+    ),
   model_verbosity: modelVerbositySchema
     .optional()
     .describe("Optional GPT-5 model verbosity override."),
@@ -376,7 +379,7 @@ server.registerTool(
   {
     title: "Run one Codex agent",
     description:
-      "Launch one OpenAI Codex agent via codex exec. Use automatically when the user asks Claude to use Codex, ask Codex, get a Codex second opinion, run a Codex subagent, use Codex Spark, or delegate one read-only analysis task. Defaults to the Codex desktop app binary when installed, read-only sandbox, fast service tier, and non-interactive approvals.",
+      "Launch one OpenAI Codex agent via codex exec. Use automatically when the user asks Claude to use Codex, ask Codex, get a Codex second opinion, run a Codex subagent, use Codex Spark, or delegate one read-only analysis task. Defaults to the Codex desktop app binary when installed, read-only sandbox, Codex's normal service tier, and non-interactive approvals.",
     inputSchema: {
       prompt: z
         .string()
@@ -545,7 +548,7 @@ server.registerTool(
         defaultModel: defaultModel(),
         defaultReasoningEffort: defaultReasoningEffort(),
         defaultSandbox: "read-only",
-        defaultServiceTier: "fast",
+        defaultServiceTier: "codex-default",
         modelPresets: {
           codex: "gpt-5.3-codex",
           spark: "gpt-5.3-codex-spark",

src/runner.ts

@@ -72,7 +72,7 @@ export interface AgentRunResult {
   modelPreset?: ModelPreset;
   reasoningEffort: ReasoningEffort;
   sandbox: SandboxMode;
-  serviceTier: ServiceTier;
+  serviceTier?: ServiceTier;
   exitCode: number | null;
   signal: NodeJS.Signals | null;
   finalMessage: string;
@@ -199,7 +199,6 @@ export function buildCodexExecArgs(
   const model = resolveRequestedModel(options, env);
   const reasoningEffort = options.reasoningEffort ?? defaultReasoningEffort(env);
   const sandbox = options.sandbox ?? "read-only";
-  const serviceTier = options.serviceTier ?? "fast";
   const ephemeral = options.ephemeral ?? true;
 
   const args = [
@@ -213,8 +212,6 @@ export function buildCodexExecArgs(
     `approval_policy=${tomlString("never")}`,
     "-c",
     `model_reasoning_effort=${tomlString(reasoningEffort)}`,
-    "-c",
-    `service_tier=${tomlString(serviceTier)}`,
     "--output-last-message",
     outputPath,
   ];
@@ -231,6 +228,9 @@ export function buildCodexExecArgs(
   if (options.reasoningSummary) {
     args.push("-c", `model_reasoning_summary=${tomlString(options.reasoningSummary)}`);
   }
+  if (options.serviceTier) {
+    args.push("-c", `service_tier=${tomlString(options.serviceTier)}`);
+  }
   if (options.subagentRuntime?.maxThreads !== undefined) {
     args.push("-c", `agents.max_threads=${options.subagentRuntime.maxThreads}`);
   }
@@ -400,7 +400,7 @@ export async function runAgent(options: AgentRunOptions): Promise<AgentRunResult
       modelPreset: options.modelPreset,
       reasoningEffort: options.reasoningEffort ?? defaultReasoningEffort(childEnv),
       sandbox: options.sandbox ?? "read-only",
-      serviceTier: options.serviceTier ?? "fast",
+      serviceTier: options.serviceTier,
       exitCode,
       signal,
       finalMessage: final.text,

test/claude-autodiscovery.mjs

@@ -78,7 +78,7 @@ try {
 
 Work in this exact project directory: ${projectDir}
 
-Codex should stay read-only and include the token AUTODISCOVERY_OK in its reply. Use a fast focused Codex configuration.
+Codex should stay read-only and include the token AUTODISCOVERY_OK in its reply. Use Codex Spark, but do not set an explicit service_tier.
 
 After the Codex result comes back, return exactly one compact JSON object and no markdown. Shape: {"ok": boolean, "tokenSeen": boolean, "model": string, "cwd": string}. Set ok true when the Codex tool call completed successfully.`;
 
@@ -156,6 +156,11 @@ After the Codex result comes back, return exactly one compact JSON object and no
     "Fake Codex should be launched with the Spark preset",
     calls[0],
   );
+  assert(
+    !calls[0].args.some((arg) => arg.includes("service_tier=")),
+    "Fake Codex should not be launched with an explicit service tier by default",
+    calls[0],
+  );
 
   console.log(
     `Claude autodiscovery passed in ${envelope.duration_ms}ms, cost $${envelope.total_cost_usd}`,

test/reliability-matrix.mjs

@@ -71,6 +71,12 @@ try {
   assert(defaultAgent?.ok, "run_agent should succeed with CLAUDE_PROJECT_DIR", defaultProject);
   assert(defaultAgent.cwd === projectDir, "run_agent should default to CLAUDE_PROJECT_DIR", defaultAgent);
   assert(defaultAgent.sandbox === "read-only", "run_agent should default to read-only", defaultAgent);
+  assert(defaultAgent.serviceTier === undefined, "run_agent should not default service_tier", defaultAgent);
+  assert(
+    !defaultAgent.commandPreview.some((arg) => arg.includes("service_tier=")),
+    "run_agent should omit service_tier unless explicitly requested",
+    defaultAgent.commandPreview,
+  );
   assert(
     defaultAgent.eventSummary?.commands?.[0]?.command === "rg example",
     "run_agent should parse command events",

test/runner.test.ts

@@ -23,7 +23,7 @@ afterEach(async () => {
 });
 
 describe("buildCodexExecArgs", () => {
-  it("defaults to read-only, non-interactive approvals, fast tier, and stdin prompt input", () => {
+  it("defaults to read-only, non-interactive approvals, Codex default service tier, and stdin prompt input", () => {
     const args = buildCodexExecArgs({ cwd: "/repo" }, "/tmp/out.md", {});
 
     expect(args).toContain("exec");
@@ -32,7 +32,7 @@ describe("buildCodexExecArgs", () => {
     expect(args[args.indexOf("--sandbox") + 1]).toBe("read-only");
     expect(args).toContain('approval_policy="never"');
     expect(args).toContain('model_reasoning_effort="medium"');
-    expect(args).toContain('service_tier="fast"');
+    expect(args.some((arg) => arg.includes("service_tier="))).toBe(false);
     expect(args).toContain("--cd");
     expect(args[args.indexOf("--cd") + 1]).toBe("/repo");
     expect(args.at(-1)).toBe("-");