@bitkyc08/opencodex 2.42.0 → 2.43.0

package/src/bridge.ts

@@ -16,9 +16,9 @@ import {
   type OcxErrorPayload,
 } from "./lib/errors";
 import { redactSecretString } from "./lib/redact";
-import { repairFreeformToolInput } from "./responses/apply-patch-envelope";
+import { mayBecomePatchEnvelope, repairFreeformToolInput } from "./responses/apply-patch-envelope";
 import { encodeCompactionSummary } from "./responses/compaction";
-import { compileCodeModeHelperInput } from "./responses/code-mode-helper-compat";
+import { compileCodeModeHelperInput, resolveCodeModeHelperName } from "./responses/code-mode-helper-compat";
 import { isTruncatedStopReason, truncationReasonFor } from "./responses/truncated-stop-reason";
 import { encodeReasoningEnvelope, type ReasoningEnvelope } from "./responses/reasoning-envelope";
 import { rememberReasoningForCall } from "./responses/reasoning-replay-cache";
@@ -33,7 +33,7 @@ import {
   stripCitationMarkers,
   type CitationMarkerFilter,
 } from "./responses/citation-markers";
-import { normalizeDeclaredToolName } from "./types";
+import { declaresCodeModeExec, normalizeDeclaredToolName } from "./types";
 import { usageDisplayTotalTokens } from "./usage/totals";
 import { appendSafeWebSearchSource, safeWebSearchSources } from "./web-search/sources";
 import {
@@ -277,9 +277,12 @@ export function bridgeToResponsesSSE(
     toolName: string,
     namespace?: string,
     codeModeHelperName?: string,
-  ): string => codeModeHelperName
-    ? compileCodeModeHelperInput(args, codeModeHelperName)
-    : repairFreeformToolInput(args, toolName, namespace);
+  ): string => {
+    const helper = resolveCodeModeHelperName(codeModeHelperName, toolName, args, namespace, options?.declaredToolNames);
+    return helper
+      ? compileCodeModeHelperInput(args, helper)
+      : repairFreeformToolInput(args, toolName, namespace);
+  };
   // Best-effort unwrap of a PARTIAL freeform arg buffer for live input streaming
   // (`response.custom_tool_call_input.delta` — codex-rs uses it for UI preview only;
   // the completed custom_tool_call item stays authoritative). Compact `{"input":"...`
@@ -1172,7 +1175,13 @@ export function bridgeToResponsesSSE(
                   if (!FREEFORM_WRAP_PREFIX.startsWith(currentToolCall.args)) {
                     const full = freeformPartialInput(currentToolCall.args);
                     const emitted = currentToolCall.inputEmitted ?? "";
-                    if (full.startsWith(emitted) && full.length > emitted.length) {
+                    // Also hold a buffer that could still become a complete patch envelope:
+                    // at completion such a body is recompiled into an apply_patch helper call,
+                    // and streaming the envelope bytes first would be that same rewind.
+                    const mayCompile = declaresCodeModeExec(options?.declaredToolNames)
+                      && !currentToolCall.namespace
+                      && currentToolCall.name === "exec";
+                    if (!(mayCompile && mayBecomePatchEnvelope(full)) && full.startsWith(emitted) && full.length > emitted.length) {
                       emit("response.custom_tool_call_input.delta", {
                         item_id: currentToolCall.itemId, output_index: currentToolCall.outputIndex,
                         delta: full.slice(emitted.length),
@@ -1666,9 +1675,12 @@ function buildResponseJSONWithBudget(
     toolName: string,
     namespace?: string,
     codeModeHelperName?: string,
-  ): string => codeModeHelperName
-    ? compileCodeModeHelperInput(args, codeModeHelperName)
-    : repairFreeformToolInput(args, toolName, namespace);
+  ): string => {
+    const helper = resolveCodeModeHelperName(codeModeHelperName, toolName, args, namespace, options?.declaredToolNames);
+    return helper
+      ? compileCodeModeHelperInput(args, helper)
+      : repairFreeformToolInput(args, toolName, namespace);
+  };
   const parseArgsObj = (args: string): Record<string, unknown> => {
     try { const o = JSON.parse(args); return o && typeof o === "object" ? o : {}; } catch { return {}; }
   };

package/src/chat/inbound.ts

@@ -73,13 +73,18 @@ function userContentToBlocks(content: unknown): Rec[] {
       continue;
     }
     if (!isRec(raw)) continue;
-    if ((raw.type === "text" || raw.type === "input_text") && typeof raw.text === "string") {
+    if ((raw.type === "text" || raw.type === "input_text" || raw.type === "output_text") && typeof raw.text === "string") {
       blocks.push({ type: "input_text", text: raw.text });
       continue;
     }
     const imageUrl = imageUrlFromPart(raw);
     if (imageUrl) {
-      blocks.push({ type: "input_image", image_url: imageUrl });
+      const detail = isRec(raw.image_url) ? raw.image_url.detail : raw.detail;
+      blocks.push({
+        type: "input_image",
+        image_url: imageUrl,
+        ...(detail === "auto" || detail === "low" || detail === "high" ? { detail } : {}),
+      });
       continue;
     }
     const videoUrl = videoUrlFromPart(raw);
@@ -275,7 +280,10 @@ export function chatCompletionsToResponsesBody(raw: unknown): Rec {
           : typeof msg.tool_use_id === "string" ? msg.tool_use_id
           : "";
         if (!callId) throw new ChatCompletionsRequestError("tool messages require tool_call_id");
-        const output = typeof msg.content === "string" ? msg.content : contentToText(msg.content);
+        const blocks = userContentToBlocks(msg.content);
+        const output = blocks.some(part => part.type === "input_image")
+          ? blocks.filter(part => part.type === "input_text" || part.type === "input_image")
+          : contentToText(msg.content);
         input.push({ type: "function_call_output", call_id: callId, output });
         break;
       }

package/src/claude/inbound-content-options.ts

@@ -0,0 +1,60 @@
+import { isClaudeWebSearchToolName } from "./outbound";
+import { AnthropicRequestError, isRec, type Rec } from "./inbound-records";
+
+export function systemToInstructions(system: unknown): string | undefined {
+  if (typeof system === "string") return system.length > 0 ? system : undefined;
+  if (Array.isArray(system)) {
+    const parts: string[] = [];
+    for (const block of system) {
+      if (isRec(block) && block.type === "text" && typeof block.text === "string") parts.push(block.text);
+    }
+    return parts.length > 0 ? parts.join("\n\n") : undefined;
+  }
+  return undefined;
+}
+
+export function toolsToResponses(tools: unknown): Rec[] | undefined {
+  if (!Array.isArray(tools) || tools.length === 0) return undefined;
+  const out: Rec[] = [];
+  for (const raw of tools) {
+    if (!isRec(raw)) continue;
+    const type = typeof raw.type === "string" ? raw.type : "";
+    if (type.startsWith("web_search")) {
+      out.push({ type: "web_search" }); // hosted sidecar path
+      continue;
+    }
+    if (typeof raw.name === "string" && raw.name.length > 0 && isRec(raw.input_schema)) {
+      out.push({
+        type: "function",
+        name: raw.name,
+        ...(typeof raw.description === "string" ? { description: raw.description } : {}),
+        parameters: raw.input_schema as Record<string, unknown>,
+      });
+      continue;
+    }
+    // Other server tools (bash_*, text_editor_*, ...) have no routed equivalent: drop.
+  }
+  return out.length > 0 ? out : undefined;
+}
+
+export function toolChoiceToResponses(choice: unknown, body: Rec): void {
+  if (!isRec(choice)) return;
+  if (choice.disable_parallel_tool_use === true) body.parallel_tool_calls = false;
+  switch (choice.type) {
+    case "auto": body.tool_choice = "auto"; break;
+    case "none": body.tool_choice = "none"; break;
+    case "any": body.tool_choice = "required"; break;
+    case "tool":
+      if (typeof choice.name !== "string" || choice.name.length === 0) {
+        throw new AnthropicRequestError("tool_choice.tool requires a name");
+      }
+      // Anthropic represents hosted WebSearch as a named tool choice, while
+      // Responses requires the choice type to match the hosted declaration.
+      // Preserve forced-tool intent rather than weakening it to `auto`.
+      body.tool_choice = isClaudeWebSearchToolName(choice.name)
+        ? { type: "web_search" }
+        : { type: "function", name: choice.name };
+      break;
+    default: break;
+  }
+}

package/src/claude/inbound-model-options.ts

@@ -0,0 +1,142 @@
+import type { OcxClaudeCodeConfig } from "../types";
+import { isAnthropicOutputSchema } from "../adapters/anthropic-output-schema";
+import { resolveAlias } from "./alias";
+import { stripOneMillionMarker } from "./context-windows";
+import { resolveDesktop3pAlias } from "./desktop-3p";
+import { isRec, type Rec } from "./inbound-records";
+
+function isClaudeClassifierModel(model: string): boolean {
+  const stripped = model.replace(/-\d{8}$/, "");
+  return /^claude-opus-[45]/.test(stripped);
+}
+
+/**
+ * Explicitly configured classifier route for Claude Code Auto Mode safety checks (#1697).
+ *
+ * Only OPERATOR-DECLARED targets are used: `classifierModel`, then the ordered
+ * `classifierFallbacks`. Both are qualified `provider/model` strings the operator chose, so
+ * routing them crosses no boundary the operator did not ask for.
+ *
+ * Deliberately NOT here: inferring a provider from `claudeCode.model`. That value is the
+ * injected/default config slot, not the provider the live session actually selected, so it goes
+ * stale the moment the user changes the model picker -- and acting on it would silently move a
+ * classifier turn onto a provider with its own privacy and billing consequences. Live session
+ * affinity needs the request/session state this function does not have; it is tracked as
+ * follow-up work rather than approximated from static config.
+ */
+function configuredClassifierRoute(cc?: OcxClaudeCodeConfig): string | undefined {
+  const explicit = typeof cc?.classifierModel === "string" ? cc.classifierModel.trim() : "";
+  if (explicit.length > 0) return explicit;
+  if (Array.isArray(cc?.classifierFallbacks)) {
+    for (const candidate of cc.classifierFallbacks) {
+      if (typeof candidate === "string" && candidate.trim().length > 0) return candidate.trim();
+    }
+  }
+  return undefined;
+}
+
+/** Alias first, then modelMap: exact id, then date-suffix-stripped (`-\d{8}$`), then classifier affinity/config, else passthrough. */
+export function resolveInboundModel(model: string, cc?: OcxClaudeCodeConfig): string {
+  // Defensive: Desktop/CLI strip the [1m] context-variant marker client-side, but a
+  // leaking build must not break alias decode (devlog 138 — the 1M signal is the
+  // anthropic-beta header, never the id). Case-insensitive: the CLI matches /\[1m\]/i.
+  model = stripOneMillionMarker(model);
+  const aliased = resolveAlias(model);
+  if (aliased) return aliased;
+  // Desktop 3P aliases: claude-opus-4-{code} → provider/model route key
+  const desktop3p = resolveDesktop3pAlias(model);
+  if (desktop3p) {
+    // Native pseudo-provider returns bare slug; routed returns provider/model
+    const sep = desktop3p.indexOf("/");
+    if (sep > 0 && desktop3p.slice(0, sep) === "native") return desktop3p.slice(sep + 1);
+    return desktop3p;
+  }
+  const map = cc?.modelMap ?? {};
+  const exact = map[model];
+  if (typeof exact === "string" && exact.length > 0) return exact;
+  const stripped = model.replace(/-\d{8}$/, "");
+  const dateless = map[stripped];
+  if (typeof dateless === "string" && dateless.length > 0) return dateless;
+
+  // Claude Code Auto Mode classifier routing (#1697). Bare classifier checks such as
+  // `claude-opus-5` carry no provider, so without this they fall through to defaultProvider --
+  // which may not speak Anthropic at all. Only an operator-declared target is used.
+  if (isClaudeClassifierModel(model)) {
+    const configured = configuredClassifierRoute(cc);
+    if (configured) return configured;
+  }
+  return model;
+}
+
+/** budget_tokens ladder -> Responses reasoning effort (003: real API min is 1024; never forward raw). */
+export function effortForThinkingBudget(budget: number): string {
+  if (budget <= 4096) return "low";
+  if (budget <= 16384) return "medium";
+  return "high";
+}
+
+/**
+ * Adaptive-thinking wire (devlog 080): Claude Code /effort sends
+ * `thinking:{type:"adaptive"}` + `output_config:{effort:"..."}` (verified by local
+ * capture of claude 2.1.207 and CLIProxyAPI#1540). Forward the level verbatim when it
+ * is a known Responses effort; unknown strings are dropped so downstream defaults win.
+ */
+const OUTPUT_CONFIG_EFFORTS = new Set(["minimal", "low", "medium", "high", "xhigh", "max", "ultra"]);
+export function effortFromOutputConfig(outputConfig: unknown): string | undefined {
+  if (!isRec(outputConfig)) return undefined;
+  const effort = outputConfig.effort;
+  return typeof effort === "string" && OUTPUT_CONFIG_EFFORTS.has(effort) ? effort : undefined;
+}
+
+export function formatFromOutputConfig(outputConfig: unknown): Rec | undefined {
+  if (!isRec(outputConfig) || !isRec(outputConfig.format)) return undefined;
+  const format = outputConfig.format;
+  if (
+    format.type !== "json_schema"
+    || !isRec(format.schema)
+    || !isAnthropicOutputSchema(format.schema)
+  ) return undefined;
+  return { type: "json_schema", name: "response", schema: format.schema };
+}
+
+/**
+ * ocx-route directive (devlog 072): injected agent-definition bodies carry
+ * `<!-- ocx-route: <model> -->` because Claude Code 2.1.207 ignores custom
+ * gateway ids in agent frontmatter (live-proven fallback to sonnet). The body
+ * rides the subagent's system prompt, so the proxy re-routes here. Only the
+ * FIRST directive wins; the scan is bounded to the system field.
+ */
+const OCX_ROUTE_RE = /<!--\s*ocx-route:\s*([^\s]+)\s*-->/;
+const OCX_EFFORT_RE = /<!--\s*ocx-effort:\s*(low|medium|high|xhigh|max)\s*-->/;
+
+function systemText(body: unknown): string | null {
+  if (!isRec(body)) return null;
+  const system = body.system;
+  if (typeof system === "string") return system || null;
+  if (!Array.isArray(system)) return null;
+  const text = system
+    .filter((b): b is Rec => isRec(b) && b.type === "text" && typeof b.text === "string")
+    .map(b => b.text as string)
+    .join("\n");
+  return text || null;
+}
+
+export function extractOcxRouteDirective(body: unknown): string | null {
+  const text = systemText(body);
+  if (!text) return null;
+  const match = OCX_ROUTE_RE.exec(text);
+  return match ? match[1]! : null;
+}
+
+/**
+ * Claude Code 2.1.220 collapses custom-agent frontmatter `effort: max` and
+ * `effort: xhigh` into the legacy `thinking.budget_tokens` shape. Preserve the
+ * exact generated-agent setting through the same trusted system-body channel as
+ * ocx-route so the inbound translator can restore `output_config.effort`.
+ */
+export function extractOcxEffortDirective(body: unknown): NonNullable<OcxClaudeCodeConfig["subagentEffort"]> | null {
+  const text = systemText(body);
+  if (!text) return null;
+  const match = OCX_EFFORT_RE.exec(text);
+  return match ? match[1] as NonNullable<OcxClaudeCodeConfig["subagentEffort"]> : null;
+}

package/src/claude/inbound-records.ts

@@ -0,0 +1,7 @@
+export class AnthropicRequestError extends Error {}
+
+export type Rec = Record<string, unknown>;
+
+export function isRec(v: unknown): v is Rec {
+  return !!v && typeof v === "object" && !Array.isArray(v);
+}

package/src/claude/inbound.ts

@@ -10,130 +10,24 @@
  *  - top_k is accepted and silently dropped (no Responses equivalent, CCR parity).
  */
 import type { OcxClaudeCodeConfig } from "../types";
-import { isAnthropicOutputSchema } from "../adapters/anthropic-output-schema";
-import { resolveAlias } from "./alias";
-import { stripOneMillionMarker } from "./context-windows";
-import { resolveDesktop3pAlias } from "./desktop-3p";
-import { isClaudeWebSearchToolName } from "./outbound";
 import { createHash } from "node:crypto";
 
-export class AnthropicRequestError extends Error {}
+export { AnthropicRequestError } from "./inbound-records";
+export { resolveInboundModel, effortForThinkingBudget, effortFromOutputConfig, extractOcxRouteDirective, extractOcxEffortDirective } from "./inbound-model-options";
+import { AnthropicRequestError, isRec, type Rec } from "./inbound-records";
+import { resolveInboundModel, effortForThinkingBudget, effortFromOutputConfig, formatFromOutputConfig } from "./inbound-model-options";
+import { systemToInstructions, toolsToResponses, toolChoiceToResponses } from "./inbound-content-options";
 
-type Rec = Record<string, unknown>;
 
-function isRec(v: unknown): v is Rec {
-  return !!v && typeof v === "object" && !Array.isArray(v);
-}
-
-function isClaudeClassifierModel(model: string): boolean {
-  const stripped = model.replace(/-\d{8}$/, "");
-  return /^claude-opus-[45]/.test(stripped);
-}
-
-/**
- * Explicitly configured classifier route for Claude Code Auto Mode safety checks (#1697).
- *
- * Only OPERATOR-DECLARED targets are used: `classifierModel`, then the ordered
- * `classifierFallbacks`. Both are qualified `provider/model` strings the operator chose, so
- * routing them crosses no boundary the operator did not ask for.
- *
- * Deliberately NOT here: inferring a provider from `claudeCode.model`. That value is the
- * injected/default config slot, not the provider the live session actually selected, so it goes
- * stale the moment the user changes the model picker -- and acting on it would silently move a
- * classifier turn onto a provider with its own privacy and billing consequences. Live session
- * affinity needs the request/session state this function does not have; it is tracked as
- * follow-up work rather than approximated from static config.
- */
-function configuredClassifierRoute(cc?: OcxClaudeCodeConfig): string | undefined {
-  const explicit = typeof cc?.classifierModel === "string" ? cc.classifierModel.trim() : "";
-  if (explicit.length > 0) return explicit;
-  if (Array.isArray(cc?.classifierFallbacks)) {
-    for (const candidate of cc.classifierFallbacks) {
-      if (typeof candidate === "string" && candidate.trim().length > 0) return candidate.trim();
-    }
-  }
-  return undefined;
-}
-
-/** Alias first, then modelMap: exact id, then date-suffix-stripped (`-\d{8}$`), then classifier affinity/config, else passthrough. */
-export function resolveInboundModel(model: string, cc?: OcxClaudeCodeConfig): string {
-  // Defensive: Desktop/CLI strip the [1m] context-variant marker client-side, but a
-  // leaking build must not break alias decode (devlog 138 — the 1M signal is the
-  // anthropic-beta header, never the id). Case-insensitive: the CLI matches /\[1m\]/i.
-  model = stripOneMillionMarker(model);
-  const aliased = resolveAlias(model);
-  if (aliased) return aliased;
-  // Desktop 3P aliases: claude-opus-4-{code} → provider/model route key
-  const desktop3p = resolveDesktop3pAlias(model);
-  if (desktop3p) {
-    // Native pseudo-provider returns bare slug; routed returns provider/model
-    const sep = desktop3p.indexOf("/");
-    if (sep > 0 && desktop3p.slice(0, sep) === "native") return desktop3p.slice(sep + 1);
-    return desktop3p;
-  }
-  const map = cc?.modelMap ?? {};
-  const exact = map[model];
-  if (typeof exact === "string" && exact.length > 0) return exact;
-  const stripped = model.replace(/-\d{8}$/, "");
-  const dateless = map[stripped];
-  if (typeof dateless === "string" && dateless.length > 0) return dateless;
-
-  // Claude Code Auto Mode classifier routing (#1697). Bare classifier checks such as
-  // `claude-opus-5` carry no provider, so without this they fall through to defaultProvider --
-  // which may not speak Anthropic at all. Only an operator-declared target is used.
-  if (isClaudeClassifierModel(model)) {
-    const configured = configuredClassifierRoute(cc);
-    if (configured) return configured;
-  }
-  return model;
-}
-
-/** budget_tokens ladder -> Responses reasoning effort (003: real API min is 1024; never forward raw). */
-export function effortForThinkingBudget(budget: number): string {
-  if (budget <= 4096) return "low";
-  if (budget <= 16384) return "medium";
-  return "high";
-}
-
-/**
- * Adaptive-thinking wire (devlog 080): Claude Code /effort sends
- * `thinking:{type:"adaptive"}` + `output_config:{effort:"..."}` (verified by local
- * capture of claude 2.1.207 and CLIProxyAPI#1540). Forward the level verbatim when it
- * is a known Responses effort; unknown strings are dropped so downstream defaults win.
- */
-const OUTPUT_CONFIG_EFFORTS = new Set(["minimal", "low", "medium", "high", "xhigh", "max", "ultra"]);
-export function effortFromOutputConfig(outputConfig: unknown): string | undefined {
-  if (!isRec(outputConfig)) return undefined;
-  const effort = outputConfig.effort;
-  return typeof effort === "string" && OUTPUT_CONFIG_EFFORTS.has(effort) ? effort : undefined;
-}
-
-function formatFromOutputConfig(outputConfig: unknown): Rec | undefined {
-  if (!isRec(outputConfig) || !isRec(outputConfig.format)) return undefined;
-  const format = outputConfig.format;
-  if (
-    format.type !== "json_schema"
-    || !isRec(format.schema)
-    || !isAnthropicOutputSchema(format.schema)
-  ) return undefined;
-  return { type: "json_schema", name: "response", schema: format.schema };
-}
-
-function systemToInstructions(system: unknown): string | undefined {
-  if (typeof system === "string") return system.length > 0 ? system : undefined;
-  if (Array.isArray(system)) {
-    const parts: string[] = [];
-    for (const block of system) {
-      if (isRec(block) && block.type === "text" && typeof block.text === "string") parts.push(block.text);
-    }
-    return parts.length > 0 ? parts.join("\n\n") : undefined;
-  }
-  return undefined;
-}
 
 function imageBlockToInputImage(block: Rec): Rec | null {
   const source = block.source;
   if (!isRec(source)) return null;
+  if (source.type === "file") {
+    throw new AnthropicRequestError(
+      "File-backed images require native Anthropic passthrough; use base64 or URL images on translated routes.",
+    );
+  }
   if (source.type === "base64" && typeof source.data === "string") {
     const media = typeof source.media_type === "string" ? source.media_type : "image/png";
     return { type: "input_image", image_url: `data:${media};base64,${source.data}` };
@@ -194,47 +88,6 @@ export function effectiveBlockedSkillNames(cc?: Pick<OcxClaudeCodeConfig, "block
     .filter(name => name.length > 0))];
 }
 
-/**
- * ocx-route directive (devlog 072): injected agent-definition bodies carry
- * `<!-- ocx-route: <model> -->` because Claude Code 2.1.207 ignores custom
- * gateway ids in agent frontmatter (live-proven fallback to sonnet). The body
- * rides the subagent's system prompt, so the proxy re-routes here. Only the
- * FIRST directive wins; the scan is bounded to the system field.
- */
-const OCX_ROUTE_RE = /<!--\s*ocx-route:\s*([^\s]+)\s*-->/;
-const OCX_EFFORT_RE = /<!--\s*ocx-effort:\s*(low|medium|high|xhigh|max)\s*-->/;
-
-function systemText(body: unknown): string | null {
-  if (!isRec(body)) return null;
-  const system = body.system;
-  if (typeof system === "string") return system || null;
-  if (!Array.isArray(system)) return null;
-  const text = system
-    .filter((b): b is Rec => isRec(b) && b.type === "text" && typeof b.text === "string")
-    .map(b => b.text as string)
-    .join("\n");
-  return text || null;
-}
-
-export function extractOcxRouteDirective(body: unknown): string | null {
-  const text = systemText(body);
-  if (!text) return null;
-  const match = OCX_ROUTE_RE.exec(text);
-  return match ? match[1]! : null;
-}
-
-/**
- * Claude Code 2.1.220 collapses custom-agent frontmatter `effort: max` and
- * `effort: xhigh` into the legacy `thinking.budget_tokens` shape. Preserve the
- * exact generated-agent setting through the same trusted system-body channel as
- * ocx-route so the inbound translator can restore `output_config.effort`.
- */
-export function extractOcxEffortDirective(body: unknown): NonNullable<OcxClaudeCodeConfig["subagentEffort"]> | null {
-  const text = systemText(body);
-  if (!text) return null;
-  const match = OCX_EFFORT_RE.exec(text);
-  return match ? match[1] as NonNullable<OcxClaudeCodeConfig["subagentEffort"]> : null;
-}
 
 /** Injected-skill payloads below this size are never stubbed (not worth it). */
 const SKILL_ELISION_MIN_CHARS = 10_000;
@@ -391,51 +244,6 @@ function assistantMessageToItems(content: unknown, input: Rec[]): void {
   flush();
 }
 
-function toolsToResponses(tools: unknown): Rec[] | undefined {
-  if (!Array.isArray(tools) || tools.length === 0) return undefined;
-  const out: Rec[] = [];
-  for (const raw of tools) {
-    if (!isRec(raw)) continue;
-    const type = typeof raw.type === "string" ? raw.type : "";
-    if (type.startsWith("web_search")) {
-      out.push({ type: "web_search" }); // hosted sidecar path
-      continue;
-    }
-    if (typeof raw.name === "string" && raw.name.length > 0 && isRec(raw.input_schema)) {
-      out.push({
-        type: "function",
-        name: raw.name,
-        ...(typeof raw.description === "string" ? { description: raw.description } : {}),
-        parameters: raw.input_schema as Record<string, unknown>,
-      });
-      continue;
-    }
-    // Other server tools (bash_*, text_editor_*, ...) have no routed equivalent: drop.
-  }
-  return out.length > 0 ? out : undefined;
-}
-
-function toolChoiceToResponses(choice: unknown, body: Rec): void {
-  if (!isRec(choice)) return;
-  if (choice.disable_parallel_tool_use === true) body.parallel_tool_calls = false;
-  switch (choice.type) {
-    case "auto": body.tool_choice = "auto"; break;
-    case "none": body.tool_choice = "none"; break;
-    case "any": body.tool_choice = "required"; break;
-    case "tool":
-      if (typeof choice.name !== "string" || choice.name.length === 0) {
-        throw new AnthropicRequestError("tool_choice.tool requires a name");
-      }
-      // Anthropic represents hosted WebSearch as a named tool choice, while
-      // Responses requires the choice type to match the hosted declaration.
-      // Preserve forced-tool intent rather than weakening it to `auto`.
-      body.tool_choice = isClaudeWebSearchToolName(choice.name)
-        ? { type: "web_search" }
-        : { type: "function", name: choice.name };
-      break;
-    default: break;
-  }
-}
 
 /** Recursive canonical JSON (keys sorted at every depth) — stable cache-cohort input. */
 function canonicalJson(value: unknown): string {

package/src/claude/model-info.ts

@@ -111,9 +111,32 @@ export function buildAnthropicModelInfos(
   aliasForRoute: (provider: string, modelId: string) => string = desktop3pAlias,
   nativeContextCap?: NativeContextLimitsInput,
   fastMode?: boolean,
+  // Presence is the feature gate: the caller passes undefined when `fastRows` is off, so a
+  // default install publishes nothing. The predicate answers ELIGIBILITY, not enablement.
+  fastRows?: (model: CatalogModel | { provider: string; id: string }) => boolean,
 ): AnthropicModelInfo[] {
   const out: AnthropicModelInfo[] = [];
   const seen = new Set<string>();
+  // Every id the loops below will really emit, computed BEFORE either runs. `seen` alone is
+  // not enough: it grows as they run, so whether a synthetic id collided with a real one
+  // would depend on iteration order. With both `foo` and a real `foo--fast` in the roster,
+  // the synthetic id for `foo` IS the real model's id, and whichever ran first would win it.
+  const realDiscoveryIds = new Set<string>([
+    ...nativeSlugs.map(slug => (
+      idStyle === "readable" ? claudeCodeNativeAlias(slug) : aliasForRoute("native", slug)
+    )),
+    ...routedModels.map(m => {
+      // The same asymmetry the routed loop applies: readable uses the LISTED id, so a
+      // fastMode-rewritten Cursor row is counted under the id it is really published as,
+      // while Desktop 3P hashes the RAW id.
+      const listed = fastMode === true && m.provider === "cursor" && idStyle === "readable"
+        ? cursorFastIdFor(m.id) ?? m.id
+        : m.id;
+      return idStyle === "readable"
+        ? claudeCodeAlias(m.provider, listed)
+        : aliasForRoute(m.provider, m.id);
+    }),
+  ]);
   // [1m] picker variant (devlog 260712 B1): Claude Code accounts exactly 1M for ids
   // carrying the marker (2.1.207 binary: /\[1m\]/i → 1e6, compaction preserved), so
   // ONLY models with an authoritative >=1M window get a second selectable row —
@@ -140,6 +163,21 @@ export function buildAnthropicModelInfos(
       : ONE_MILLION;
     out.push({ ...base, id, display_name: `${base.display_name} · 1M`, max_input_tokens: advertised });
   };
+  /**
+   * Publish a Fast sibling beside a row, following `push1mVariant` rather than the
+   * `fastMode` rewrite below: `fastMode` is a global switch with no per-request choice, so
+   * it REPLACES the listed id, while a selector has to leave the default pickable beside it.
+   *
+   * Because it only ADDS a row, it is safe for the Desktop 3P hashed style too — the
+   * exclusion `fastMode` needs exists because rewriting a hash strands a saved selection.
+   */
+  const pushFastVariant = (base: AnthropicModelInfo) => {
+    const id = `${base.id}--fast`;
+    // A real model always wins its own id, whatever the iteration order.
+    if (realDiscoveryIds.has(id) || seen.has(id)) return;
+    seen.add(id);
+    out.push({ ...base, id, display_name: `${base.display_name} · Fast` });
+  };
   for (const slug of nativeSlugs) {
     const id = idStyle === "readable" ? claudeCodeNativeAlias(slug) : aliasForRoute("native", slug);
     if (seen.has(id)) continue;
@@ -151,6 +189,9 @@ export function buildAnthropicModelInfos(
     const info = modelInfo(id, `${slug} (native)`, nativeEffectiveLadder(slug), true, nativeMaxInput ?? nativeWindow);
     out.push(info);
     push1mVariant(info, nativeWindow, nativeMaxInput);
+    // Natives too, not only routed rows: gpt-5.6-sol is the flagship Fast model, and
+    // omitting it would leave this surface without the model the feature exists for.
+    if (fastRows?.({ provider: "native", id: slug }) === true) pushFastVariant(info);
   }
   for (const m of routedModels) {
     // Global Fast has no toggle on this surface, so the fast identity is what gets listed —
@@ -180,6 +221,10 @@ export function buildAnthropicModelInfos(
     // Anthropic passthrough guard (audit 021 #3): never auto-widen canonical claude
     // routes — only a genuine >=1M window earns the variant row there.
     push1mVariant(info, m.contextWindow, routedMaxInput);
+    // The whole model is passed, not a (provider, id) pair: a combo row lives in its own
+    // namespace with no config.providers entry, so the caller classifies it from the
+    // aggregated supportsServiceTier the row already carries.
+    if (fastRows?.(m) === true) pushFastVariant(info);
   }
   return out;
 }