@cartanova/qgrid-cli 2.10.0 → 2.10.1

package/bundle/src/application/token/token.model.ts

@@ -50,7 +50,7 @@ class TokenModelClass extends BaseModelClass<
 
   @api({ httpMethod: "GET", clients: ["axios", "tanstack-query"], resourceName: "Token" })
   async findById<T extends TokenSubsetKey>(subset: T, id: number): Promise<TokenSubsetMapping[T]> {
-    const { rows } = await this.findMany(subset, { id, num: 1, page: 1 });
+    const { rows } = await this.findMany(subset, { id, num: 1, page: 1, queryMode: "list" });
     if (!rows[0]) {
       throw new NotFoundException(SD("error.entityNotFound")("Token", id));
     }
@@ -61,7 +61,12 @@ class TokenModelClass extends BaseModelClass<
     subset: T,
     listParams: TokenListParams,
   ): Promise<TokenSubsetMapping[T] | null> {
-    const { rows } = await this.findMany(subset, { ...listParams, num: 1, page: 1 });
+    const { rows } = await this.findMany(subset, {
+      ...listParams,
+      num: 1,
+      page: 1,
+      queryMode: "list",
+    });
     return rows[0] ?? null;
   }
 
@@ -78,7 +83,7 @@ class TokenModelClass extends BaseModelClass<
       ...rawParams,
     } satisfies TokenListParams;
 
-    const { qb, onSubset: _ } = this.getSubsetQueries(subset);
+    const { qb } = this.getSubsetQueries(subset);
 
     if (params.id) {
       qb.whereIn("tokens.id", asArray(params.id));
@@ -105,11 +110,7 @@ class TokenModelClass extends BaseModelClass<
       }
     }
 
-    const enhancers = this.createEnhancers({
-      A: (row) => ({ ...row }),
-    });
-
-    return this.executeSubsetQuery({ subset, qb, params, enhancers, debug: false });
+    return this.executeSubsetQuery({ subset, qb, params });
   }
 
   async findByAccountIdentifier<T extends TokenSubsetKey>(
@@ -121,13 +122,10 @@ class TokenModelClass extends BaseModelClass<
     qb.where("tokens.provider", provider);
     const jsonKey = provider === "anthropic" ? "accountUuid" : "accountId";
     qb.whereRaw(`tokens.credentials->>'${jsonKey}' = ?`, [accountId]);
-    const enhancers = this.createEnhancers({ A: (row) => ({ ...row }) });
     const result = await this.executeSubsetQuery({
       subset,
       qb,
-      params: { num: 100, page: 1 },
-      enhancers,
-      debug: false,
+      params: { num: 0, page: 1, queryMode: "list" },
     });
     return result.rows;
   }
@@ -139,13 +137,10 @@ class TokenModelClass extends BaseModelClass<
     const { qb } = this.getSubsetQueries(subset);
     qb.where("tokens.active", true);
     qb.where("tokens.provider", provider);
-    const enhancers = this.createEnhancers({ A: (row) => ({ ...row }) });
     const result = await this.executeSubsetQuery({
       subset,
       qb,
-      params: { num: 100, page: 1 },
-      enhancers,
-      debug: false,
+      params: { num: 0, page: 1, queryMode: "list" },
     });
     return result.rows;
   }
@@ -159,13 +154,10 @@ class TokenModelClass extends BaseModelClass<
     qb.where("tokens.active", true);
     qb.where("tokens.provider", provider);
     qb.where("tokens.name", name);
-    const enhancers = this.createEnhancers({ A: (row) => ({ ...row }) });
     const result = await this.executeSubsetQuery({
       subset,
       qb,
-      params: { num: 1, page: 1 },
-      enhancers,
-      debug: false,
+      params: { num: 1, page: 1, queryMode: "list" },
     });
     return result.rows[0];
   }
@@ -175,13 +167,10 @@ class TokenModelClass extends BaseModelClass<
     const { qb } = this.getSubsetQueries(subset);
     qb.where("tokens.reauth_required", true);
     qb.orderBy("tokens.name", "asc");
-    const enhancers = this.createEnhancers({ A: (row) => ({ ...row }) });
     const result = await this.executeSubsetQuery({
       subset,
       qb,
-      params: { num: 100, page: 1 },
-      enhancers,
-      debug: false,
+      params: { num: 0, page: 1, queryMode: "list" },
     });
     return result.rows;
   }
@@ -189,13 +178,10 @@ class TokenModelClass extends BaseModelClass<
   async findActive<T extends TokenSubsetKey>(subset: T): Promise<TokenSubsetMapping[T][]> {
     const { qb } = this.getSubsetQueries(subset);
     qb.where("tokens.active", true);
-    const enhancers = this.createEnhancers({ A: (row) => ({ ...row }) });
     const result = await this.executeSubsetQuery({
       subset,
       qb,
-      params: { num: 100, page: 1 },
-      enhancers,
-      debug: false,
+      params: { num: 0, page: 1, queryMode: "list" },
     });
     return result.rows;
   }
@@ -246,24 +232,53 @@ class TokenModelClass extends BaseModelClass<
     accountId: string | undefined,
     saveParams: TokenSaveParams,
   ): Promise<void> {
-    let keepaliveEnabled = saveParams.keepalive_enabled;
-    if (accountId) {
-      const oldEntries = await this.findByAccountIdentifier("A", provider, accountId);
-      if (keepaliveEnabled === undefined && oldEntries.length > 0) {
-        keepaliveEnabled = oldEntries.some((entry) => entry.keepalive_enabled);
+    await this.getPuri("w").transaction(async (trx) => {
+      let replacement = saveParams;
+      if (accountId) {
+        const jsonKey = provider === "anthropic" ? "accountUuid" : "accountId";
+        const oldEntries = await trx
+          .from("tokens")
+          .select({
+            id: "id",
+            quota_threshold: "quota_threshold",
+            weight: "weight",
+            ord: "ord",
+            keepalive_enabled: "keepalive_enabled",
+          })
+          .where("provider", provider)
+          .whereRaw(`credentials->>'${jsonKey}' = ?`, [accountId])
+          .orderBy("id", "asc")
+          .forUpdate();
+        const first = oldEntries[0];
+        if (first) {
+          replacement = {
+            ...saveParams,
+            quota_threshold:
+              saveParams.quota_threshold === undefined
+                ? first.quota_threshold
+                : saveParams.quota_threshold,
+            weight: saveParams.weight ?? first.weight,
+            ord: saveParams.ord ?? first.ord,
+            keepalive_enabled:
+              saveParams.keepalive_enabled ?? oldEntries.some((entry) => entry.keepalive_enabled),
+          };
+          await trx
+            .from("tokens")
+            .whereIn(
+              "id",
+              oldEntries.map((entry) => entry.id),
+            )
+            .delete();
+        }
+      } else {
+        logger.warn(
+          `${provider} login without account identifier: dedup skipped for ${saveParams.name ?? "unnamed"}`,
+        );
       }
-      if (oldEntries.length > 0) await this.del(oldEntries.map((o) => o.id));
-    } else {
-      logger.warn(
-        `${provider} login without account identifier: dedup skipped for ${saveParams.name ?? "unnamed"}`,
-      );
-    }
 
-    await this.save([
-      keepaliveEnabled === undefined
-        ? saveParams
-        : { ...saveParams, keepalive_enabled: keepaliveEnabled },
-    ]);
+      trx.ubRegister("tokens", applyCreateDefaults(replacement));
+      await trx.ubUpsert("tokens");
+    });
   }
 
   /**

package/bundle/src/server-error-handler.ts

@@ -2,8 +2,7 @@ import { getLogger } from "@logtape/logtape";
 import { type FastifyReply } from "fastify";
 import { SoException } from "sonamu";
 
-// dispatcher 기동은 dev0 기준 1~2분 걸린다(워커 25개 × spawn 간격). 그보다 짧게 잡으면
-// 재시도가 같은 503 을 다시 받는다.
+// provider 초기화 중 즉시 재요청이 몰리지 않도록 재시도 간격을 안내한다.
 const RETRY_AFTER_SECONDS = "30";
 
 // statusCode 를 신뢰하는 범위를 Sonamu 예외로 한정한다. qgrid 는 프록시라 upstream HTTP

package/bundle/src/sonamu.config.ts

@@ -130,37 +130,10 @@ export default defineConfig({
       },
       guardHandler: () => {},
       cacheControlHandler: (req) => {
-        switch (req.type) {
-          case "assets":
-            // Hash 포함된 파일: 영구 캐시
-            if (req.path.match(/-[a-f0-9]+\./)) {
-              return CachePresets.immutable;
-            }
-            return CachePresets.longLived;
-
-          case "api":
-            // GET 요청만 캐싱 고려
-            if (req.method === "GET") {
-              // 특정 경로는 짧은 캐시
-              if (req.path.startsWith("/api/static-data")) {
-                return CachePresets.shortLived;
-              }
-              if (req.path.startsWith("/api/terms")) {
-                return CachePresets.mediumLived;
-              }
-            }
-            // 기본: 캐시 없음
-            return CachePresets.noCache;
-
-          case "ssr":
-            // SSR 페이지: 10초 캐시
-            return CachePresets.ssr;
-
-          case "csr":
-            // CSR fallback (index.html): 1분 캐시
-            return CachePresets.shortLived;
-        }
-        return CachePresets.noCache;
+        if (req.type === "api") return CachePresets.noCache;
+        if (req.type === "csr") return CachePresets.shortLived;
+        // Sonamu serves fingerprinted assets with its default immutable policy.
+        return undefined;
       },
     },
     cache: {

package/bundle/src/utils/providers/anthropic/anthropic-constants.test.ts

@@ -36,6 +36,7 @@ describe("Anthropic 1M context policy", () => {
     expect(supports1MContext("claude-opus-5-5")).toBe(true);
     expect(supports1MContext("claude-sonnet-5")).toBe(true);
     expect(supports1MContext("claude-sonnet-5-5")).toBe(true);
+    expect(supports1MContext("claude-haiku-5-5")).toBe(true);
     expect(supports1MContext("claude-sonnet-4-6")).toBe(true);
     expect(supports1MContext("claude-opus-4-6")).toBe(true);
     expect(supports1MContext("claude-opus-4-8")).toBe(true);
@@ -53,6 +54,7 @@ describe("Anthropic 1M context policy", () => {
     expect(needsCli1mSuffix("claude-opus-5-5")).toBe(false);
     expect(needsCli1mSuffix("claude-sonnet-5")).toBe(false);
     expect(needsCli1mSuffix("claude-sonnet-5-5")).toBe(false);
+    expect(needsCli1mSuffix("claude-haiku-5-5")).toBe(false);
     expect(needsCli1mSuffix("claude-sonnet-4-6")).toBe(true);
     expect(needsCli1mSuffix("claude-opus-4-6")).toBe(true);
     expect(needsCli1mSuffix("claude-opus-4-8")).toBe(false);
@@ -82,7 +84,7 @@ describe("Anthropic 1M context policy", () => {
 });
 
 describe("Anthropic thinking policy", () => {
-  it("Fable 5/5.1, Opus 5, Opus 5.5, Sonnet 5.5 는 adaptive thinking 을 보존한다", () => {
+  it("Fable 5/5.1, Opus 5, Opus 5.5, Sonnet 5.5, Haiku 5.5 는 adaptive thinking 을 보존한다", () => {
     expect(usesAdaptiveThinking("claude-fable-5-1")).toBe(true);
     expect(usesAdaptiveThinking("anthropic/claude-fable-5-1")).toBe(true);
     expect(usesAdaptiveThinking("claude-fable-5")).toBe(true);
@@ -93,6 +95,9 @@ describe("Anthropic thinking policy", () => {
     expect(usesAdaptiveThinking("anthropic/claude-opus-5-5[1m]")).toBe(true);
     expect(usesAdaptiveThinking("claude-sonnet-5-5")).toBe(true);
     expect(usesAdaptiveThinking("anthropic/claude-sonnet-5-5[1m]")).toBe(true);
+    expect(usesAdaptiveThinking("claude-haiku-5-5")).toBe(true);
+    expect(usesAdaptiveThinking("anthropic/claude-haiku-5-5")).toBe(true);
+    expect(usesAdaptiveThinking("claude-haiku-4-5")).toBe(false);
     expect(usesAdaptiveThinking("claude-sonnet-5")).toBe(false);
     expect(usesAdaptiveThinking("claude-opus-4-8")).toBe(false);
   });

package/bundle/src/utils/providers/anthropic/anthropic-constants.ts

@@ -51,14 +51,16 @@ const ONE_MILLION_CONTEXT_MODELS = new Set([
   "claude-opus-5-5",
   "claude-sonnet-5-5",
   "claude-sonnet-5",
+  "claude-haiku-5-5",
   "claude-sonnet-4-6",
   "claude-opus-4-6",
   "claude-opus-4-8",
 ]);
 
 const CLI_ONE_MILLION_SUFFIX_MODELS = new Set(["claude-sonnet-4-6", "claude-opus-4-6"]);
-// Fable 5/5.1, Opus 5.5, Sonnet 5.5 는 thinking disabled 를 거부한다(Fable: CLI 카탈로그 `rejects_disabled_thinking`,
-// Opus 5.5: 공식 사양 "can't be turned off", 2026-09-22 출시. Sonnet 5.5: API `{type: "disabled"}` 400, CLI 카탈로그 `rejects_disabled_thinking`, 2026-09-29 출시). Opus 5 는
+// Fable 5/5.1, Opus 5.5, Sonnet 5.5, Haiku 5.5 는 thinking disabled 를 거부한다(Fable: CLI 카탈로그 `rejects_disabled_thinking`,
+// Opus 5.5: 공식 사양 "can't be turned off", 2026-09-22 출시. Sonnet 5.5: API `{type: "disabled"}` 400, CLI 카탈로그 `rejects_disabled_thinking`, 2026-09-28 출시.
+// Haiku 5.5: API 는 high 이하에서만 disabled 허용·xhigh/max 400, Claude Code 2.1.293 카탈로그는 `rejects_disabled_thinking`, 2026-10-07 출시). Opus 5 는
 // disabled 를 high 이하에서 지원하지만, 공식 기본값·권장 동작인 adaptive thinking 을 유지한다. qgrid
 // 기본 effort=low 가 비용/지연 제어를 맡으며, 이 정책은 disabled+xhigh/max 400 오류와 disabled 시
 // tool/XML 출력 오염도 피한다.
@@ -68,6 +70,7 @@ const ADAPTIVE_THINKING_MODELS = new Set([
   "claude-opus-5",
   "claude-opus-5-5",
   "claude-sonnet-5-5",
+  "claude-haiku-5-5",
 ]);
 
 // Fable 계열: 구독 플랜에서 usage credits 가 필요한 최상위 티어. 구독 usage API 의

package/bundle/src/utils/providers/anthropic/anthropic-dispatcher.test.ts

@@ -153,24 +153,6 @@ describe("AnthropicDispatcher", () => {
     expect(call.input).toEqual([{ type: "text", text: "hi", text_elements: [] }]);
   });
 
-  it("reuse/reuseInput 이 실려 와도 무시하고 coldInput/coldHistory 로 실행한다", async () => {
-    const d = new AnthropicDispatcher();
-    d.onTokenAdded(1, "tok-A", creds(), null, 1);
-
-    await d.generate(
-      baseReq({
-        coldHistory: [{ type: "message", role: "assistant", content: [] }],
-        reuse: { workerId: 1, threadId: "S1", epoch: 0 },
-        reuseInput: [{ type: "text", text: "delta", text_elements: [] }],
-      }),
-    );
-
-    const call = firstRunRequest();
-    expect(call).not.toHaveProperty("resumeSessionId");
-    expect(call.coldHistory).toBeDefined();
-    expect(call.input).toEqual([{ type: "text", text: "hi", text_elements: [] }]);
-  });
-
   it("structured(outputSchema): jsonSchema 로 직렬화되어 전달", async () => {
     const d = new AnthropicDispatcher();
     d.onTokenAdded(1, "tok-A", creds(), null, 1);
@@ -245,14 +227,7 @@ describe("AnthropicDispatcher", () => {
     d.onTokenAdded(1, "tok-A", creds(), null, 1);
     runClaudeSessionMock.mockRejectedValueOnce(new Error("closed without result"));
 
-    await expect(
-      d.generate(
-        baseReq({
-          reuse: { workerId: 1, threadId: "S1", epoch: 0 },
-          reuseInput: [{ type: "text", text: "delta", text_elements: [] }],
-        }),
-      ),
-    ).rejects.toThrow("closed without result");
+    await expect(d.generate(baseReq())).rejects.toThrow("closed without result");
     expect(runClaudeSessionMock).toHaveBeenCalledTimes(1);
   });
 

package/bundle/src/utils/providers/common/model-cost.test.ts

@@ -11,6 +11,25 @@ describe("calculateCostUsd", () => {
       calculateCostUsd("openai/gpt-6-astra", { inputTokens, outputTokens: 1_000 }),
     ).toBeCloseTo(expected, 10);
   });
+  it.each([
+    [100_000, 0.0105],
+    [100_001, 0.0525005],
+  ])("Haiku 5.5 applies the over-100K prompt tier to the whole request (%i input)", (inputTokens, expected) => {
+    expect(
+      calculateCostUsd("anthropic/claude-haiku-5-5", { inputTokens, outputTokens: 1_000 }),
+    ).toBeCloseTo(expected, 10);
+  });
+  it("Haiku 5.5 over-100K tier multiplies cache read and cache write rates by 5x", () => {
+    // 50K uncached × $0.50 + 1K output × $2.50 + 200K cache read × $0.05 + 50K 1h cache write × $1.
+    expect(
+      calculateCostUsd("anthropic/claude-haiku-5-5", {
+        inputTokens: 300_000,
+        outputTokens: 1_000,
+        cachedInputTokens: 200_000,
+        cacheCreationInputTokens: 50_000,
+      }),
+    ).toBeCloseTo(0.0875, 10);
+  });
   it.each([
     ["gpt-6-astra", 10, 50, 1, 12.5],
     ["gpt-6.1-sol", 2, 10, 0.1, 2.5],
@@ -60,7 +79,9 @@ describe("calculateCostUsd", () => {
     // Opus 5.5 는 cache read 만 0.05x 특례($0.20), cache write 배율은 표준(5m $5 / 1h $8).
     ["claude-opus-5-5", 4, 20, 0.2, 8],
     ["claude-sonnet-5", 2, 10, 0.2, 4],
-    ["claude-sonnet-5-5", 2, 10, 0.2, 4],
+    // Sonnet 5.5 는 2026-10-07 부터 cache read 0.05x($0.10), cache write 배율은 표준(5m $2.50 / 1h $4).
+    ["claude-sonnet-5-5", 2, 10, 0.1, 4],
+    ["claude-haiku-5-5", 0.1, 0.5, 0.01, 0.2],
   ])(
     "%s official Anthropic rates for 5m/1h cache writes",
     (model, inputTokens, outputTokens, cachedInputTokens, cacheCreationInputTokens) => {

package/bundle/src/utils/providers/common/model-cost.ts

@@ -157,7 +157,7 @@ const OPENAI_COSTS: Record<string, OpenAIModelSpec> = {
 function anthropicCosts(
   inputTokens: number,
   outputTokens: number,
-  overrides: Pick<Partial<ModelCosts>, "cachedInputTokens"> = {},
+  overrides: Pick<Partial<ModelCosts>, "cachedInputTokens" | "longContext"> = {},
 ): ModelCosts {
   return {
     inputTokens,
@@ -169,10 +169,13 @@ function anthropicCosts(
     cacheCreationInputTokens: inputTokens * 2,
     cacheCreationInputTokens5m: inputTokens * 1.25,
     cacheCreationInputTokens1h: inputTokens * 2,
+    // Haiku 5.5 처럼 프롬프트 길이 티어가 있는 모델만 long-context 배율을 가진다.
+    ...(overrides.longContext ? { longContext: overrides.longContext } : {}),
   };
 }
 
-// 가격 출처: https://platform.claude.com/docs/en/about-claude/pricing (2026-09-23 확인, 기존 모델 단가 변동 없음)
+// 가격 출처: https://platform.claude.com/docs/en/about-claude/pricing (2026-10-08 확인. 2026-10-07 에 Sonnet 5.5
+// cache read 가 $0.20→$0.10(0.05x)으로 인하됐고, 그 외 기존 모델 단가는 변동 없음)
 const ANTHROPIC_COSTS: Record<string, ModelCosts> = {
   // Fable 5.1 (2026-09-01 출시): input/output 은 Fable 5 와 같고 cache read 만 0.025x($0.25) 특례.
   // @see https://platform.claude.com/docs/en/models/fable-5-1/overview
@@ -200,9 +203,23 @@ const ANTHROPIC_COSTS: Record<string, ModelCosts> = {
   // Sonnet 5 의 introductory $2/$10 이 정식 단가로 확정됐다. 2026-09-01 에 예정됐던 $3/$15 인상은
   // 취소됐으므로 날짜 분기 없이 고정 단가로 계산한다.
   "claude-sonnet-5": anthropicCosts(2, 10),
-  // Sonnet 5.5 (2026-09-29 출시): Sonnet 5 와 같은 $2/$10, cache read 는 표준 0.1x($0.20).
-  // @see https://platform.claude.com/docs/en/about-claude/pricing
-  "claude-sonnet-5-5": anthropicCosts(2, 10),
+  // Sonnet 5.5 (2026-09-28 출시): Sonnet 5 와 같은 $2/$10. cache read 는 출시 시 표준 0.1x($0.20)였다가
+  // 2026-10-07 Haiku 5.5 출시와 함께 Opus 5.5 와 같은 0.05x($0.10)로 인하됐다(API 릴리스 노트). 비용은 요청 시점
+  // 단가로 기록되므로 날짜 분기 없이 현행 단가만 둔다. cache write 는 표준 배율($2.50/$4).
+  // @see https://platform.claude.com/docs/en/release-notes/overview (2026-10-07)
+  "claude-sonnet-5-5": anthropicCosts(2, 10, { cachedInputTokens: 0.1 }),
+  // Haiku 5.5 (2026-10-07 출시): 프롬프트 길이 티어 단가. 100K 이하 $0.10/$0.50, cache read 0.1x($0.01),
+  // cache write 5m $0.125 / 1h $0.20. 입력이 100K 를 넘으면 input/cache read/cache write/output 모두 5x
+  // ($0.50/$0.05/$0.625·$1/$2.50)가 요청 전체에 적용된다. 1M 컨텍스트 모델 중 유일하게 길이별 단가를 쓴다.
+  // @see https://platform.claude.com/docs/en/models/haiku-5-5/overview
+  "claude-haiku-5-5": anthropicCosts(0.1, 0.5, {
+    longContext: {
+      threshold: 100_000,
+      inputMultiplier: 5,
+      cachedInputMultiplier: 5,
+      outputMultiplier: 5,
+    },
+  }),
 };
 
 // gpt-5.3-codex-spark 는 research preview 로 공식 token 단가가 아직 final 이 아니다.

package/bundle/src/utils/providers/common/provider-dispatcher.ts

@@ -1,7 +1,7 @@
 /**
  * ProviderDispatcher — provider 별 LLM 요청 실행 인터페이스.
  *
- * MVP 에서는 stream() 만. getRateLimits(), listModels() 는 future.
+ * Provider 실행 결과와 stream callback의 공통 계약.
  */
 
 import {
@@ -18,8 +18,7 @@ export type ProviderTokenUsageBreakdown = TokenUsageBreakdown & {
   cacheCreationInputTokens1h?: number;
 };
 
-// thread 재사용 라우팅 좌표. 상위(qgrid.dispatcher)에서 conv 핸들 검증을 통과한 경우에만 전달.
-// dispatcher 는 이 좌표가 가리키는 worker 의 기존 thread 에 turn 만 실행한다.
+// 공개 thread 좌표의 provider 결과 부분. 이름은 호환성을 위해 유지한다.
 export interface ReuseThreadCoord {
   workerId: number;
   threadId: string;
@@ -61,9 +60,8 @@ export interface GenerateRequest {
   // provider 실행 제한(ms). Queue selection 이후의 active provider request 전체에 적용한다.
   timeoutMs?: number;
   abortSignal?: AbortSignal;
-  // 첫 turn / 재사용 폴백 시 보낼 input — 전체 prompt. 항상 설정.
+  // 매 turn에 전달하는 현재 input과 전체 history.
   coldInput: Array<UserInput>;
-  // 첫 turn / 폴백 시 inject 할 전체 history.
   coldHistory?: Array<JsonValue>;
   // Provider-neutral cache affinity hints. Direct OpenAI currently receives these while still
   // replaying coldInput + coldHistory on every turn; a dispatcher may use them for token choice
@@ -72,13 +70,8 @@ export interface GenerateRequest {
   preferredTokenId?: number;
   // 사람이 지정한 exact target. Cache affinity 선호와 달리 부적격 시 다른 토큰으로 대체하지 않는다.
   requirePreferredToken?: boolean;
-  // 재사용 좌표 + delta input. 둘은 한 쌍 — 검증 통과 시에만 설정한다.
-  // dispatcher 가 worker/thread 생존을 재검증해 성공하면 reuseInput(delta)을, 실패하면
-  // coldInput + coldHistory 로 폴백한다(전체 history 로 문맥 복구).
-  reuse?: ReuseThreadCoord;
-  reuseInput?: Array<UserInput>;
   // OpenAI image_generation tool 을 켠다(OpenAI 경로 전용, opt-in).
-  // 이 플래그가 있으면 dispatcher 는 항상 cold thread 로 실행하고 재사용 라우팅을 건너뛴다.
+  // 이 플래그가 있으면 OpenAI prompt-cache affinity 라우팅을 건너뛴다.
   imageGeneration?: boolean | "auto";
   imageGenerationOptions?: ImageGenerationOptions;
 }