Merge pull request #493 from corbitsdev/cl-6572-show-a-default-reasoning-effort-in-the-prompt-when-the-model

Show default reasoning effort in the prompt for the live model

Author: TheGreatAxios

src/config.test.ts

@@ -3,7 +3,7 @@ import { mkdtemp, mkdir, writeFile, rm } from "node:fs/promises";
 import { tmpdir } from "node:os";
 import { join } from "node:path";
 
-import { buildBifrostSource, buildOpenAISource, buildProviderCatalog, catalogEntryAsProviderSettings, CliHelpError, CLI_HELP_TEXT, KEYLESS_API_KEY, loadConfig, providerCatalogToSettings, runtimeSettingsWithCatalog, SOURCE_MAX_TOKENS } from "./config/index.js";
+import { buildBifrostSource, buildOpenAISource, buildXaiSource, buildProviderCatalog, catalogEntryAsProviderSettings, CliHelpError, CLI_HELP_TEXT, KEYLESS_API_KEY, loadConfig, providerCatalogToSettings, runtimeSettingsWithCatalog, SOURCE_MAX_TOKENS } from "./config/index.js";
 import type { Config, UnconfiguredConfig } from "./config/index.js";
 import { mergeProviderIntoSettings, type ResolvedProvider, type Settings } from "./config/settings.js";
 import { OPENCODE_GO_BASE_URL } from "../packages/opencode-go/src/index.js";
@@ -677,6 +677,29 @@ describe("buildBifrostSource", () => {
   });
 });
 
+describe("buildXaiSource", () => {
+  test("omits reasoning_effort when effort is absent", () => {
+    const source = buildXaiSource({ id: "xai/work", apiKey: "tok", model: "grok-4.6" });
+    expect(source.provider).toBe("grok-responses");
+    expect(source.defaults?.providerOptions).not.toHaveProperty("reasoning_effort");
+  });
+
+  test("sets providerOptions.reasoning_effort when effort is present", () => {
+    const source = buildXaiSource({
+      id: "xai/work",
+      apiKey: "tok",
+      model: "grok-4.6",
+      reasoningEffort: "low",
+    });
+    expect(source.defaults?.providerOptions).toMatchObject({ reasoning_effort: "low" });
+  });
+
+  test("does not invent high when effort is absent", () => {
+    const source = buildXaiSource({ id: "xai/work", apiKey: "tok", model: "grok-4.6" });
+    expect(source.defaults?.providerOptions?.["reasoning_effort"]).toBeUndefined();
+  });
+});
+
 describe("buildProviderCatalog", () => {
   const resolved: ResolvedProvider = {
     providerName: "fp",

src/config/index.ts

@@ -168,10 +168,12 @@ export function buildXaiSource(fields: {
   id: string;
   apiKey: string;
   model: string;
+  reasoningEffort?: ReasoningEffort;
 }): InferenceSource {
   const userId = xaiUserIdFromAccessToken(fields.apiKey);
   const providerOptions: Record<string, unknown> = {};
   if (userId !== undefined) providerOptions[GROK_USER_ID_OPTION] = userId;
+  if (fields.reasoningEffort !== undefined) providerOptions["reasoning_effort"] = fields.reasoningEffort;
   return {
     id: fields.id,
     provider: GROK_RESPONSES_PROVIDER,

src/config/inference-sources.ts

@@ -10,7 +10,7 @@ import {
   type ProviderCatalogEntry,
 } from "./index.js";
 import type { Settings } from "./settings.js";
-import type { ReasoningEffort } from "../provider/reasoning-effort.js";
+import { resolveSessionEffort, type ReasoningEffort } from "../provider/reasoning-effort.js";
 import { SOURCE_MAX_TOKENS } from "./index.js";
 import { isOpenCodeGoProvider } from "../../packages/opencode-go/src/index.js";
 import { resolveDefaultModel } from "./providers.js";
@@ -77,7 +77,11 @@ export function buildInferenceSourceForRef(
   if (baseURL === undefined) return null;
 
   const maxTokens = maxTokensFor(settings, ref.provider, ref.model);
-  const effort = ref.reasoningEffort ?? ctx.reasoningEffort;
+  const configured = ref.reasoningEffort ?? ctx.reasoningEffort;
+  const effort =
+    configured !== undefined
+      ? resolveSessionEffort(ref.model, configured, entry?.codexProfile !== undefined)
+      : undefined;
 
   if (entry?.codexProfile !== undefined) {
     return buildCodexSource({
@@ -94,6 +98,7 @@ export function buildInferenceSourceForRef(
       id: ref.provider,
       apiKey: entry.apiKey ?? "",
       model: ref.model,
+      ...(effort !== undefined ? { reasoningEffort: effort } : {}),
     });
   }
   if (

src/exec/runner.ts

@@ -507,6 +507,9 @@ export async function runExec(config: Config): Promise<ExecResult> {
               id: config.providerName,
               apiKey: config.apiKey,
               model: config.model,
+              ...(config.reasoningEffort !== undefined
+                ? { reasoningEffort: config.reasoningEffort }
+                : {}),
             })
           : buildOpenAICompatibleInitialSource();
 

src/provider/grok-responses-adapter.test.ts

@@ -75,4 +75,43 @@ describe("createGrokResponsesAdapter", () => {
     expect(body.include).toEqual(["reasoning.encrypted_content"]);
     expect(body.reasoning).toEqual({ summary: "detailed" });
   });
+
+  test("forwards providerOptions.reasoning_effort onto reasoning.effort", () => {
+    const adapter = createGrokResponsesAdapter(source);
+    const turns: ConversationTurn[] = [
+      {
+        role: "user",
+        timestamp: 0,
+        content: [{ type: "text", text: "hello" }],
+      },
+    ];
+
+    const request = adapter.buildRequest(turns, "grok-4.6", {
+      providerOptions: { reasoning_effort: "low" },
+    });
+    const body = JSON.parse(request.body) as {
+      reasoning?: { effort?: string; summary?: string };
+    };
+
+    expect(body.reasoning).toEqual({ effort: "low", summary: "detailed" });
+  });
+
+  test("does not invent high when no reasoning_effort is set", () => {
+    const adapter = createGrokResponsesAdapter(source);
+    const turns: ConversationTurn[] = [
+      {
+        role: "user",
+        timestamp: 0,
+        content: [{ type: "text", text: "hello" }],
+      },
+    ];
+
+    const request = adapter.buildRequest(turns, "grok-4.6", {});
+    const body = JSON.parse(request.body) as {
+      reasoning?: { effort?: string; summary?: string };
+    };
+
+    expect(body.reasoning).toEqual({ summary: "detailed" });
+    expect(body.reasoning?.effort).toBeUndefined();
+  });
 });

src/provider/grok-responses-adapter.ts

@@ -152,6 +152,10 @@ function buildRequest(
   const input = systemMessage !== undefined ? [systemMessage, ...conversation] : conversation;
   const tools = toResponsesTools(options);
 
+  const reasoning: { summary: "detailed"; effort?: string } = { summary: "detailed" };
+  const effort = optionString(options, "reasoning_effort");
+  if (effort !== undefined) reasoning.effort = effort;
+
   const body: Record<string, unknown> = {
     model,
     input,
@@ -160,8 +164,9 @@ function buildRequest(
     include: ["reasoning.encrypted_content"],
     // "detailed" streams denser summary deltas than "auto". Grok bills full
     // thinking tokens but only returns summarized text; sparse auto summaries
-    // left the stall/activity clocks quiet for 60–120s mid-think.
-    reasoning: { summary: "detailed" },
+    // left the stall/activity clocks quiet for 60–120s mid-think. Effort is
+    // forwarded when the source set it — this adapter does not invent a default.
+    reasoning,
   };
   if (tools !== undefined) {
     body["tools"] = tools;

src/provider/reasoning-effort.test.ts

@@ -11,7 +11,10 @@ import {
   clampEffort,
   pickEffortFromCascade,
   resolveEffortForRole,
+  defaultEffortForModel,
+  resolveSessionEffort,
 } from "./reasoning-effort.js";
+import { composePromptActionBarModelLabel } from "../tui/components/prompt-action-bar-label.js";
 
 describe("REASONING_EFFORTS", () => {
   test("is ordered from least to most effort", () => {
@@ -120,15 +123,41 @@ describe("cycleReasoningEffort", () => {
   afterEach(() => setModelReasoningCapabilities({}));
 
   test("walks the gpt-5 ladder and wraps", () => {
-    expect(cycleReasoningEffort("gpt-5", undefined)).toBe("minimal");
     expect(cycleReasoningEffort("gpt-5", "minimal")).toBe("low");
     expect(cycleReasoningEffort("gpt-5", "low")).toBe("medium");
     expect(cycleReasoningEffort("gpt-5", "medium")).toBe("high");
     expect(cycleReasoningEffort("gpt-5", "high")).toBe("minimal");
   });
 
-  test("starts at the first supported level when current is unsupported", () => {
-    expect(cycleReasoningEffort("gpt-5", "xhigh")).toBe("minimal");
+  test("unset gpt-5 cycles from the implicit medium default to high", () => {
+    expect(cycleReasoningEffort("gpt-5", undefined)).toBe("high");
+  });
+
+  test("unset grok cycles from implicit high, matching an explicit high", () => {
+    expect(cycleReasoningEffort("grok-4.6", undefined)).toBe(
+      cycleReasoningEffort("grok-4.6", "high"),
+    );
+    expect(cycleReasoningEffort("grok-4.6", "high")).toBe("low");
+  });
+
+  test("unset gpt-5.1 chat cycles from implicit none to minimal", () => {
+    expect(cycleReasoningEffort("gpt-5.1", undefined)).toBe("minimal");
+  });
+
+  test("leftover unsupported effort cycles from the family default", () => {
+    expect(cycleReasoningEffort("gpt-5", "xhigh")).toBe("high");
+    expect(cycleReasoningEffort("gpt-5", "xhigh")).toBe(cycleReasoningEffort("gpt-5", "medium"));
+  });
+
+  test("grok leftover minimal cycles the same as unset / high", () => {
+    expect(cycleReasoningEffort("grok-4.6", "minimal")).toBe(cycleReasoningEffort("grok-4.6", undefined));
+    expect(cycleReasoningEffort("grok-4.6", "minimal")).toBe(cycleReasoningEffort("grok-4.6", "high"));
+    expect(cycleReasoningEffort("grok-4.6", "minimal")).toBe("low");
+  });
+
+  test("unknown models with rungs still start at supported[0] when no default exists", () => {
+    expect(defaultEffortForModel("some-random-model")).toBeUndefined();
+    expect(cycleReasoningEffort("some-random-model", undefined)).toBe("low");
   });
 
   test("returns undefined for a non-reasoning model", () => {
@@ -334,3 +363,88 @@ describe("resolveEffortForRole", () => {
     ).toBe("high");
   });
 });
+
+describe("defaultEffortForModel", () => {
+  afterEach(() => setModelReasoningCapabilities({}));
+
+  test("grok family defaults to high", () => {
+    expect(defaultEffortForModel("grok-4.6")).toBe("high");
+    expect(defaultEffortForModel("grok-4.5")).toBe("high");
+  });
+
+  test("gpt-5 and o-series default to medium", () => {
+    expect(defaultEffortForModel("gpt-5")).toBe("medium");
+    expect(defaultEffortForModel("o1")).toBe("medium");
+    expect(defaultEffortForModel("o3-mini")).toBe("medium");
+    expect(defaultEffortForModel("o4-mini")).toBe("medium");
+  });
+
+  test("gpt-5.1 chat defaults to none when none is on the ladder", () => {
+    expect(supportedEfforts("gpt-5.1").includes("none")).toBe(true);
+    expect(defaultEffortForModel("gpt-5.1")).toBe("none");
+    expect(defaultEffortForModel("gpt-5.1", false)).toBe("none");
+  });
+
+  test("Codex defaults to medium", () => {
+    expect(defaultEffortForModel("gpt-5.6-sol", true)).toBe("medium");
+    expect(defaultEffortForModel("gpt-5.1-codex", true)).toBe("medium");
+  });
+
+  test("empty ladder yields undefined", () => {
+    setModelReasoningCapabilities({ "chat-only-model": false });
+    expect(defaultEffortForModel("chat-only-model")).toBeUndefined();
+  });
+
+  test("unknown models with rungs have no family default", () => {
+    expect(supportedEfforts("some-random-model").length).toBeGreaterThan(0);
+    expect(defaultEffortForModel("some-random-model")).toBeUndefined();
+  });
+});
+
+describe("resolveSessionEffort", () => {
+  afterEach(() => setModelReasoningCapabilities({}));
+
+  test("empty ladder yields undefined even when configured", () => {
+    setModelReasoningCapabilities({ "chat-only-model": false });
+    expect(resolveSessionEffort("chat-only-model", "high")).toBeUndefined();
+  });
+
+  test("keeps a supported configured level", () => {
+    expect(resolveSessionEffort("gpt-5", "low")).toBe("low");
+    expect(resolveSessionEffort("grok-4.6", "low")).toBe("low");
+  });
+
+  test("falls back to the family default when unset or unsupported", () => {
+    expect(resolveSessionEffort("gpt-5", undefined)).toBe("medium");
+    expect(resolveSessionEffort("gpt-5", "xhigh")).toBe("medium");
+    expect(resolveSessionEffort("grok-4.6", undefined)).toBe("high");
+    expect(resolveSessionEffort("gpt-5.1", undefined)).toBe("none");
+    expect(resolveSessionEffort("gpt-5.6-sol", undefined, true)).toBe("medium");
+    expect(resolveSessionEffort("some-random-model", undefined)).toBeUndefined();
+  });
+});
+
+describe("prompt action bar effort label", () => {
+  test("joiner stays a dumb concatenation of the resolved session effort", () => {
+    const effort = resolveSessionEffort("grok-4.6", undefined);
+    expect(effort).toBe("high");
+    expect(
+      composePromptActionBarModelLabel({
+        profile: "xai/work",
+        model: "grok-4.6",
+        ...(effort !== undefined ? { effort } : {}),
+      }),
+    ).toBe("xai/work · grok-4.6 · high");
+  });
+
+  test("shows gpt-5 medium without seeding a configured effort", () => {
+    const effort = resolveSessionEffort("gpt-5", undefined);
+    expect(effort).toBe("medium");
+    expect(
+      composePromptActionBarModelLabel({
+        model: "gpt-5",
+        ...(effort !== undefined ? { effort } : {}),
+      }),
+    ).toBe("gpt-5 · medium");
+  });
+});

src/provider/reasoning-effort.ts

@@ -107,6 +107,10 @@ export function validateEffort(
  * Next effort on the model's supported ladder (wraps around). Returns undefined
  * when the model supports no reasoning effort — callers flash a status and leave
  * the session config alone.
+ *
+ * Walks from resolveSessionEffort: unset and leftover unsupported current sit
+ * on the family default, then the next rung. When there is no family default
+ * but the ladder is non-empty, start at supported[0].
  */
 export function cycleReasoningEffort(
   model: string,
@@ -115,13 +119,53 @@ export function cycleReasoningEffort(
 ): ReasoningEffort | undefined {
   const supported = supportedEfforts(model, undefined, isCodex);
   if (supported.length === 0) return undefined;
-  if (current === undefined || !supported.includes(current)) {
+  const implicit = resolveSessionEffort(model, current, isCodex);
+  if (implicit === undefined || !supported.includes(implicit)) {
     return supported[0];
   }
-  const idx = supported.indexOf(current);
+  const idx = supported.indexOf(implicit);
   return supported[(idx + 1) % supported.length];
 }
 
+/**
+ * Product default effort for a live session model. Distinct from role defaults
+ * (`defaultEffortForDirector`): this is what the prompt shows and what Shift+Tab
+ * advances from when the operator has not picked a level.
+ *
+ * Family table: grok* → high; Codex → medium; gpt-5.1 chat (`none` on the
+ * ladder, not Codex) → none; gpt-5/o1/o3/o4 → medium. Unknown models with a
+ * conservative rung set stay undefined so we do not invent a family default.
+ */
+export function defaultEffortForModel(
+  model: string,
+  isCodex = false,
+): ReasoningEffort | undefined {
+  const supported = supportedEfforts(model, undefined, isCodex);
+  if (supported.length === 0) return undefined;
+  const pick = (desired: ReasoningEffort): ReasoningEffort | undefined =>
+    supported.includes(desired) ? desired : undefined;
+  if (model.startsWith("grok")) return pick("high");
+  if (!isCodex && supported.includes("none")) return "none";
+  if (isCodex || isKnownOpenAIReasoningModel(model)) return pick("medium");
+  return undefined;
+}
+
+/**
+ * Effort the session is currently on: a configured level when the model accepts
+ * it, otherwise the family default. Empty ladders stay undefined. Does not
+ * write back into session config — display and request wiring read this.
+ */
+export function resolveSessionEffort(
+  model: string,
+  configured: ReasoningEffort | undefined,
+  isCodex = false,
+): ReasoningEffort | undefined {
+  const supported = supportedEfforts(model, undefined, isCodex);
+  if (supported.length === 0) return undefined;
+  if (configured !== undefined && supported.includes(configured)) return configured;
+  return defaultEffortForModel(model, isCodex);
+}
+
 // ---------------------------------------------------------------------------
 // Role-based product defaults (CL-5162)
 //